Make pg_stripe_size a per-pool config

This makes sense for some SSDs like Intel D3-4510 because they don't
like overwrites of the same sector:

$ fio -direct=1 -rw=write -bs=4k -size=4k -loops=100000 -iodepth=1
  write: IOPS=3142, BW=12.3MiB/s (12.9MB/s)(97.9MiB/7977msec)

$ fio -direct=1 -rw=write -bs=4k -size=128k -loops=100000 -iodepth=1
  write: IOPS=20.8k, BW=81.4MiB/s (85.3MB/s)(543MiB/6675msec)

.gitmodules

@@ -0,0 +1,6 @@
+[submodule "cpp-btree"]
+	path = cpp-btree
+	url = ../cpp-btree.git
+[submodule "json11"]
+	path = json11
+	url = ../json11.git

GPL-2.0.txt

@@ -0,0 +1,339 @@
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 2, June 1991
+
+ Copyright (C) 1989, 1991 Free Software Foundation, Inc.,
+ 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The licenses for most software are designed to take away your
+freedom to share and change it.  By contrast, the GNU General Public
+License is intended to guarantee your freedom to share and change free
+software--to make sure the software is free for all its users.  This
+General Public License applies to most of the Free Software
+Foundation's software and to any other program whose authors commit to
+using it.  (Some other Free Software Foundation software is covered by
+the GNU Lesser General Public License instead.)  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+this service if you wish), that you receive source code or can get it
+if you want it, that you can change the software or use pieces of it
+in new free programs; and that you know you can do these things.
+
+  To protect your rights, we need to make restrictions that forbid
+anyone to deny you these rights or to ask you to surrender the rights.
+These restrictions translate to certain responsibilities for you if you
+distribute copies of the software, or if you modify it.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must give the recipients all the rights that
+you have.  You must make sure that they, too, receive or can get the
+source code.  And you must show them these terms so they know their
+rights.
+
+  We protect your rights with two steps: (1) copyright the software, and
+(2) offer you this license which gives you legal permission to copy,
+distribute and/or modify the software.
+
+  Also, for each author's protection and ours, we want to make certain
+that everyone understands that there is no warranty for this free
+software.  If the software is modified by someone else and passed on, we
+want its recipients to know that what they have is not the original, so
+that any problems introduced by others will not reflect on the original
+authors' reputations.
+
+  Finally, any free program is threatened constantly by software
+patents.  We wish to avoid the danger that redistributors of a free
+program will individually obtain patent licenses, in effect making the
+program proprietary.  To prevent this, we have made it clear that any
+patent must be licensed for everyone's free use or not licensed at all.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                    GNU GENERAL PUBLIC LICENSE
+   TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
+
+  0. This License applies to any program or other work which contains
+a notice placed by the copyright holder saying it may be distributed
+under the terms of this General Public License.  The "Program", below,
+refers to any such program or work, and a "work based on the Program"
+means either the Program or any derivative work under copyright law:
+that is to say, a work containing the Program or a portion of it,
+either verbatim or with modifications and/or translated into another
+language.  (Hereinafter, translation is included without limitation in
+the term "modification".)  Each licensee is addressed as "you".
+
+Activities other than copying, distribution and modification are not
+covered by this License; they are outside its scope.  The act of
+running the Program is not restricted, and the output from the Program
+is covered only if its contents constitute a work based on the
+Program (independent of having been made by running the Program).
+Whether that is true depends on what the Program does.
+
+  1. You may copy and distribute verbatim copies of the Program's
+source code as you receive it, in any medium, provided that you
+conspicuously and appropriately publish on each copy an appropriate
+copyright notice and disclaimer of warranty; keep intact all the
+notices that refer to this License and to the absence of any warranty;
+and give any other recipients of the Program a copy of this License
+along with the Program.
+
+You may charge a fee for the physical act of transferring a copy, and
+you may at your option offer warranty protection in exchange for a fee.
+
+  2. You may modify your copy or copies of the Program or any portion
+of it, thus forming a work based on the Program, and copy and
+distribute such modifications or work under the terms of Section 1
+above, provided that you also meet all of these conditions:
+
+    a) You must cause the modified files to carry prominent notices
+    stating that you changed the files and the date of any change.
+
+    b) You must cause any work that you distribute or publish, that in
+    whole or in part contains or is derived from the Program or any
+    part thereof, to be licensed as a whole at no charge to all third
+    parties under the terms of this License.
+
+    c) If the modified program normally reads commands interactively
+    when run, you must cause it, when started running for such
+    interactive use in the most ordinary way, to print or display an
+    announcement including an appropriate copyright notice and a
+    notice that there is no warranty (or else, saying that you provide
+    a warranty) and that users may redistribute the program under
+    these conditions, and telling the user how to view a copy of this
+    License.  (Exception: if the Program itself is interactive but
+    does not normally print such an announcement, your work based on
+    the Program is not required to print an announcement.)
+
+These requirements apply to the modified work as a whole.  If
+identifiable sections of that work are not derived from the Program,
+and can be reasonably considered independent and separate works in
+themselves, then this License, and its terms, do not apply to those
+sections when you distribute them as separate works.  But when you
+distribute the same sections as part of a whole which is a work based
+on the Program, the distribution of the whole must be on the terms of
+this License, whose permissions for other licensees extend to the
+entire whole, and thus to each and every part regardless of who wrote it.
+
+Thus, it is not the intent of this section to claim rights or contest
+your rights to work written entirely by you; rather, the intent is to
+exercise the right to control the distribution of derivative or
+collective works based on the Program.
+
+In addition, mere aggregation of another work not based on the Program
+with the Program (or with a work based on the Program) on a volume of
+a storage or distribution medium does not bring the other work under
+the scope of this License.
+
+  3. You may copy and distribute the Program (or a work based on it,
+under Section 2) in object code or executable form under the terms of
+Sections 1 and 2 above provided that you also do one of the following:
+
+    a) Accompany it with the complete corresponding machine-readable
+    source code, which must be distributed under the terms of Sections
+    1 and 2 above on a medium customarily used for software interchange; or,
+
+    b) Accompany it with a written offer, valid for at least three
+    years, to give any third party, for a charge no more than your
+    cost of physically performing source distribution, a complete
+    machine-readable copy of the corresponding source code, to be
+    distributed under the terms of Sections 1 and 2 above on a medium
+    customarily used for software interchange; or,
+
+    c) Accompany it with the information you received as to the offer
+    to distribute corresponding source code.  (This alternative is
+    allowed only for noncommercial distribution and only if you
+    received the program in object code or executable form with such
+    an offer, in accord with Subsection b above.)
+
+The source code for a work means the preferred form of the work for
+making modifications to it.  For an executable work, complete source
+code means all the source code for all modules it contains, plus any
+associated interface definition files, plus the scripts used to
+control compilation and installation of the executable.  However, as a
+special exception, the source code distributed need not include
+anything that is normally distributed (in either source or binary
+form) with the major components (compiler, kernel, and so on) of the
+operating system on which the executable runs, unless that component
+itself accompanies the executable.
+
+If distribution of executable or object code is made by offering
+access to copy from a designated place, then offering equivalent
+access to copy the source code from the same place counts as
+distribution of the source code, even though third parties are not
+compelled to copy the source along with the object code.
+
+  4. You may not copy, modify, sublicense, or distribute the Program
+except as expressly provided under this License.  Any attempt
+otherwise to copy, modify, sublicense or distribute the Program is
+void, and will automatically terminate your rights under this License.
+However, parties who have received copies, or rights, from you under
+this License will not have their licenses terminated so long as such
+parties remain in full compliance.
+
+  5. You are not required to accept this License, since you have not
+signed it.  However, nothing else grants you permission to modify or
+distribute the Program or its derivative works.  These actions are
+prohibited by law if you do not accept this License.  Therefore, by
+modifying or distributing the Program (or any work based on the
+Program), you indicate your acceptance of this License to do so, and
+all its terms and conditions for copying, distributing or modifying
+the Program or works based on it.
+
+  6. Each time you redistribute the Program (or any work based on the
+Program), the recipient automatically receives a license from the
+original licensor to copy, distribute or modify the Program subject to
+these terms and conditions.  You may not impose any further
+restrictions on the recipients' exercise of the rights granted herein.
+You are not responsible for enforcing compliance by third parties to
+this License.
+
+  7. If, as a consequence of a court judgment or allegation of patent
+infringement or for any other reason (not limited to patent issues),
+conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot
+distribute so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you
+may not distribute the Program at all.  For example, if a patent
+license would not permit royalty-free redistribution of the Program by
+all those who receive copies directly or indirectly through you, then
+the only way you could satisfy both it and this License would be to
+refrain entirely from distribution of the Program.
+
+If any portion of this section is held invalid or unenforceable under
+any particular circumstance, the balance of the section is intended to
+apply and the section as a whole is intended to apply in other
+circumstances.
+
+It is not the purpose of this section to induce you to infringe any
+patents or other property right claims or to contest validity of any
+such claims; this section has the sole purpose of protecting the
+integrity of the free software distribution system, which is
+implemented by public license practices.  Many people have made
+generous contributions to the wide range of software distributed
+through that system in reliance on consistent application of that
+system; it is up to the author/donor to decide if he or she is willing
+to distribute software through any other system and a licensee cannot
+impose that choice.
+
+This section is intended to make thoroughly clear what is believed to
+be a consequence of the rest of this License.
+
+  8. If the distribution and/or use of the Program is restricted in
+certain countries either by patents or by copyrighted interfaces, the
+original copyright holder who places the Program under this License
+may add an explicit geographical distribution limitation excluding
+those countries, so that distribution is permitted only in or among
+countries not thus excluded.  In such case, this License incorporates
+the limitation as if written in the body of this License.
+
+  9. The Free Software Foundation may publish revised and/or new versions
+of the General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+Each version is given a distinguishing version number.  If the Program
+specifies a version number of this License which applies to it and "any
+later version", you have the option of following the terms and conditions
+either of that version or of any later version published by the Free
+Software Foundation.  If the Program does not specify a version number of
+this License, you may choose any version ever published by the Free Software
+Foundation.
+
+  10. If you wish to incorporate parts of the Program into other free
+programs whose distribution conditions are different, write to the author
+to ask for permission.  For software which is copyrighted by the Free
+Software Foundation, write to the Free Software Foundation; we sometimes
+make exceptions for this.  Our decision will be guided by the two goals
+of preserving the free status of all derivatives of our free software and
+of promoting the sharing and reuse of software generally.
+
+                            NO WARRANTY
+
+  11. BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY
+FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW.  EXCEPT WHEN
+OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES
+PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED
+OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.  THE ENTIRE RISK AS
+TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU.  SHOULD THE
+PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING,
+REPAIR OR CORRECTION.
+
+  12. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR
+REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES,
+INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING
+OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED
+TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY
+YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER
+PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGES.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+convey the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+Also add information on how to contact you by electronic and paper mail.
+
+If the program is interactive, make it output a short notice like this
+when it starts in an interactive mode:
+
+    Gnomovision version 69, Copyright (C) year name of author
+    Gnomovision comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, the commands you use may
+be called something other than `show w' and `show c'; they could even be
+mouse-clicks or menu items--whatever suits your program.
+
+You should also get your employer (if you work as a programmer) or your
+school, if any, to sign a "copyright disclaimer" for the program, if
+necessary.  Here is a sample; alter the names:
+
+  Yoyodyne, Inc., hereby disclaims all copyright interest in the program
+  `Gnomovision' (which makes passes at compilers) written by James Hacker.
+
+  <signature of Ty Coon>, 1 April 1989
+  Ty Coon, President of Vice
+
+This General Public License does not permit incorporating your program into
+proprietary programs.  If your program is a subroutine library, you may
+consider it more useful to permit linking proprietary applications with the
+library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.

Make-gen.pl

@@ -0,0 +1,46 @@
+#!/usr/bin/perl
+
+use strict;
+
+my $deps = {};
+for my $line (split /\n/, `grep '^#include "' *.cpp *.h`)
+{
+    if ($line =~ /^([^:]+):\#include "([^"]+)"/s)
+    {
+        $deps->{$1}->{$2} = 1;
+    }
+}
+
+my $added;
+do
+{
+    $added = 0;
+    for my $file (keys %$deps)
+    {
+        for my $dep (keys %{$deps->{$file}})
+        {
+            if ($deps->{$dep})
+            {
+                for my $subdep (keys %{$deps->{$dep}})
+                {
+                    if (!$deps->{$file}->{$subdep})
+                    {
+                        $added = 1;
+                        $deps->{$file}->{$subdep} = 1;
+                    }
+                }
+            }
+        }
+    }
+} while ($added);
+
+for my $file (sort keys %$deps)
+{
+    if ($file =~ /\.cpp$/)
+    {
+        my $obj = $file;
+        $obj =~ s/\.cpp$/.o/s;
+        print "$obj: $file ".join(" ", sort keys %{$deps->{$file}})."\n";
+        print "\tg++ \$(CXXFLAGS) -c -o \$\@ \$\<\n";
+    }
+}

Makefile

@@ -2,85 +2,54 @@ BLOCKSTORE_OBJS := allocator.o blockstore.o blockstore_impl.o blockstore_init.o
 	blockstore_write.o blockstore_sync.o blockstore_stable.o blockstore_rollback.o blockstore_flush.o crc32c.o ringloop.o
 # -fsanitize=address
 CXXFLAGS := -g -O3 -Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fPIC -fdiagnostics-color=always
-all: $(BLOCKSTORE_OBJS) libfio_blockstore.so osd libfio_sec_osd.so stub_osd stub_bench osd_test dump_journal
+all: libfio_blockstore.so osd libfio_sec_osd.so libfio_cluster.so stub_osd stub_uring_osd stub_bench osd_test dump_journal qemu_driver.so nbd_proxy
 clean:
 	rm -f *.o
 
-crc32c.o: crc32c.c
-	g++ $(CXXFLAGS) -c -o $@ $<
-json11.o: json11/json11.cpp
-	g++ $(CXXFLAGS) -c -o json11.o json11/json11.cpp
-allocator.o: allocator.cpp allocator.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-ringloop.o: ringloop.cpp ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-timerfd_interval.o: timerfd_interval.cpp timerfd_interval.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-timerfd_manager.o: timerfd_manager.cpp timerfd_manager.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-
-%.o: %.cpp allocator.h blockstore_flush.h blockstore.h blockstore_impl.h blockstore_init.h blockstore_journal.h crc32c.h ringloop.h object_id.h
-	g++ $(CXXFLAGS) -c -o $@ $<
 dump_journal: dump_journal.cpp crc32c.o blockstore_journal.h
 	g++ $(CXXFLAGS) -o $@ $< crc32c.o
 
 libblockstore.so: $(BLOCKSTORE_OBJS)
-	g++ $(CXXFLAGS) -o libblockstore.so -shared $(BLOCKSTORE_OBJS) -ltcmalloc_minimal -luring
+	g++ $(CXXFLAGS) -o $@ -shared $(BLOCKSTORE_OBJS) -ltcmalloc_minimal -luring
-libfio_blockstore.so: ./libblockstore.so fio_engine.cpp json11.o
+libfio_blockstore.so: ./libblockstore.so fio_engine.o json11.o
-	g++ $(CXXFLAGS) -shared -o libfio_blockstore.so fio_engine.cpp json11.o ./libblockstore.so -ltcmalloc_minimal -luring
+	g++ $(CXXFLAGS) -shared -o $@ fio_engine.o json11.o ./libblockstore.so -ltcmalloc_minimal -luring
 
-OSD_OBJS := osd.o osd_secondary.o osd_receive.o osd_send.o osd_peering.o osd_flush.o osd_peering_pg.o \
+OSD_OBJS := osd.o osd_secondary.o msgr_receive.o msgr_send.o osd_peering.o osd_flush.o osd_peering_pg.o \
-	osd_primary.o osd_primary_subops.o etcd_state_client.o cluster_client.o osd_cluster.o http_client.o pg_states.o \
+	osd_primary.o osd_primary_subops.o etcd_state_client.o messenger.o osd_cluster.o http_client.o osd_ops.o pg_states.o \
-	osd_rmw.o json11.o base64.o timerfd_manager.o
+	osd_rmw.o json11.o base64.o timerfd_manager.o epoll_manager.o
-base64.o: base64.cpp base64.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_secondary.o: osd_secondary.cpp osd.h osd_ops.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_receive.o: osd_receive.cpp osd.h osd_ops.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_send.o: osd_send.cpp osd.h osd_ops.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_peering.o: osd_peering.cpp osd.h osd_ops.h osd_peering_pg.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_cluster.o: osd_cluster.cpp osd.h osd_ops.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-http_client.o: http_client.cpp http_client.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-etcd_state_client.o: etcd_state_client.cpp etcd_state_client.h http_client.h pg_states.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-cluster_client.o: cluster_client.cpp cluster_client.h osd_ops.h timerfd_manager.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_flush.o: osd_flush.cpp osd.h osd_ops.h osd_peering_pg.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_peering_pg.o: osd_peering_pg.cpp object_id.h osd_peering_pg.h pg_states.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-pg_states.o: pg_states.cpp pg_states.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_rmw.o: osd_rmw.cpp osd_rmw.h xor.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_rmw_test: osd_rmw_test.cpp osd_rmw.cpp osd_rmw.h xor.h
-	g++ $(CXXFLAGS) -o $@ $<
-osd_primary.o: osd_primary.cpp osd_primary.h osd_rmw.h osd.h osd_ops.h osd_peering_pg.h xor.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_primary_subops.o: osd_primary_subops.cpp osd_primary.h osd_rmw.h osd.h osd_ops.h osd_peering_pg.h xor.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd.o: osd.cpp osd.h http_client.h osd_ops.h osd_peering_pg.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
 osd: ./libblockstore.so osd_main.cpp osd.h osd_ops.h $(OSD_OBJS)
-	g++ $(CXXFLAGS) -o osd osd_main.cpp $(OSD_OBJS) ./libblockstore.so -ltcmalloc_minimal -luring
+	g++ $(CXXFLAGS) -o $@ osd_main.cpp $(OSD_OBJS) ./libblockstore.so -ltcmalloc_minimal -luring
-stub_osd: stub_osd.cpp osd_ops.h rw_blocking.o
+
-	g++ $(CXXFLAGS) -o stub_osd stub_osd.cpp rw_blocking.o -ltcmalloc_minimal
+stub_osd: stub_osd.o rw_blocking.o
+	g++ $(CXXFLAGS) -o $@ stub_osd.o rw_blocking.o -ltcmalloc_minimal
+
+STUB_URING_OSD_OBJS := stub_uring_osd.o epoll_manager.o messenger.o msgr_send.o msgr_receive.o ringloop.o timerfd_manager.o json11.o
+stub_uring_osd: $(STUB_URING_OSD_OBJS)
+	g++ $(CXXFLAGS) -o $@ -ltcmalloc_minimal $(STUB_URING_OSD_OBJS) -luring
 stub_bench: stub_bench.cpp osd_ops.h rw_blocking.o
-	g++ $(CXXFLAGS) -o stub_bench stub_bench.cpp rw_blocking.o -ltcmalloc_minimal
+	g++ $(CXXFLAGS) -o $@ stub_bench.cpp rw_blocking.o -ltcmalloc_minimal
-rw_blocking.o: rw_blocking.cpp rw_blocking.h
-	g++ $(CXXFLAGS) -c -o $@ $<
 osd_test: osd_test.cpp osd_ops.h rw_blocking.o
-	g++ $(CXXFLAGS) -o osd_test osd_test.cpp rw_blocking.o -ltcmalloc_minimal
+	g++ $(CXXFLAGS) -o $@ osd_test.cpp rw_blocking.o -ltcmalloc_minimal
 osd_peering_pg_test: osd_peering_pg_test.cpp osd_peering_pg.o
 	g++ $(CXXFLAGS) -o $@ $< osd_peering_pg.o -ltcmalloc_minimal
 
-libfio_sec_osd.so: fio_sec_osd.cpp osd_ops.h rw_blocking.o
+libfio_sec_osd.so: fio_sec_osd.o rw_blocking.o
-	g++ $(CXXFLAGS) -ltcmalloc_minimal -shared -o libfio_sec_osd.so fio_sec_osd.cpp rw_blocking.o -luring
+	g++ $(CXXFLAGS) -ltcmalloc_minimal -shared -o $@ fio_sec_osd.o rw_blocking.o
+
+FIO_CLUSTER_OBJS := cluster_client.o epoll_manager.o etcd_state_client.o \
+	messenger.o msgr_send.o msgr_receive.o ringloop.o json11.o http_client.o osd_ops.o pg_states.o timerfd_manager.o base64.o
+libfio_cluster.so: fio_cluster.o $(FIO_CLUSTER_OBJS)
+	g++ $(CXXFLAGS) -ltcmalloc_minimal -shared -o $@ $< $(FIO_CLUSTER_OBJS) -luring
+
+nbd_proxy: nbd_proxy.o $(FIO_CLUSTER_OBJS)
+	g++ $(CXXFLAGS) -ltcmalloc_minimal -o $@ $< $(FIO_CLUSTER_OBJS) -luring
+
+qemu_driver.o: qemu_driver.c qemu_proxy.h
+	gcc -I qemu/b/qemu `pkg-config glib-2.0 --cflags` \
+		-I qemu/include $(CXXFLAGS) -c -o $@ $<
+
+qemu_driver.so: qemu_driver.o qemu_proxy.o $(FIO_CLUSTER_OBJS)
+	g++ $(CXXFLAGS) -ltcmalloc_minimal -shared -o $@ $< $(FIO_CLUSTER_OBJS) qemu_driver.o qemu_proxy.o -luring
 
 test_blockstore: ./libblockstore.so test_blockstore.cpp timerfd_interval.o
 	g++ $(CXXFLAGS) -o test_blockstore test_blockstore.cpp timerfd_interval.o ./libblockstore.so -ltcmalloc_minimal -luring
@@ -88,3 +57,113 @@ test: test.cpp osd_peering_pg.o
 	g++ $(CXXFLAGS) -o test test.cpp osd_peering_pg.o -luring -lm
 test_allocator: test_allocator.cpp allocator.o
 	g++ $(CXXFLAGS) -o test_allocator test_allocator.cpp allocator.o
+
+crc32c.o: crc32c.c crc32c.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+json11.o: json11/json11.cpp
+	g++ $(CXXFLAGS) -c -o json11.o json11/json11.cpp
+
+# Autogenerated
+
+allocator.o: allocator.cpp allocator.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+base64.o: base64.cpp base64.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore.o: blockstore.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore_flush.o: blockstore_flush.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore_impl.o: blockstore_impl.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore_init.o: blockstore_init.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore_journal.o: blockstore_journal.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore_open.o: blockstore_open.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore_read.o: blockstore_read.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore_rollback.o: blockstore_rollback.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore_stable.o: blockstore_stable.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore_sync.o: blockstore_sync.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+blockstore_write.o: blockstore_write.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+cluster_client.o: cluster_client.cpp cluster_client.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd_id.h osd_ops.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+dump_journal.o: dump_journal.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+epoll_manager.o: epoll_manager.cpp epoll_manager.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+etcd_state_client.o: etcd_state_client.cpp base64.h etcd_state_client.h http_client.h json11/json11.hpp object_id.h osd_id.h osd_ops.h pg_states.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+fio_cluster.o: fio_cluster.cpp cluster_client.h epoll_manager.h etcd_state_client.h fio/fio.h fio/optgroup.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd_id.h osd_ops.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+fio_engine.o: fio_engine.cpp blockstore.h fio/fio.h fio/optgroup.h json11/json11.hpp object_id.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+fio_sec_osd.o: fio_sec_osd.cpp fio/fio.h fio/optgroup.h object_id.h osd_id.h osd_ops.h rw_blocking.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+http_client.o: http_client.cpp http_client.h json11/json11.hpp timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+messenger.o: messenger.cpp json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd_id.h osd_ops.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+msgr_receive.o: msgr_receive.cpp json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd_id.h osd_ops.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+msgr_send.o: msgr_send.cpp json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd_id.h osd_ops.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+nbd_proxy.o: nbd_proxy.cpp cluster_client.h epoll_manager.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd_id.h osd_ops.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd.o: osd.cpp blockstore.h cpp-btree/btree_map.h epoll_manager.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_cluster.o: osd_cluster.cpp base64.h blockstore.h cpp-btree/btree_map.h epoll_manager.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_flush.o: osd_flush.cpp blockstore.h cpp-btree/btree_map.h epoll_manager.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_main.o: osd_main.cpp blockstore.h cpp-btree/btree_map.h epoll_manager.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_ops.o: osd_ops.cpp object_id.h osd_id.h osd_ops.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_peering.o: osd_peering.cpp base64.h blockstore.h cpp-btree/btree_map.h epoll_manager.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_peering_pg.o: osd_peering_pg.cpp cpp-btree/btree_map.h object_id.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_peering_pg_test.o: osd_peering_pg_test.cpp cpp-btree/btree_map.h object_id.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_primary.o: osd_primary.cpp blockstore.h cpp-btree/btree_map.h epoll_manager.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd.h osd_id.h osd_ops.h osd_peering_pg.h osd_primary.h osd_rmw.h pg_states.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_primary_subops.o: osd_primary_subops.cpp blockstore.h cpp-btree/btree_map.h epoll_manager.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd.h osd_id.h osd_ops.h osd_peering_pg.h osd_primary.h osd_rmw.h pg_states.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_rmw.o: osd_rmw.cpp malloc_or_die.h object_id.h osd_id.h osd_rmw.h xor.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_rmw_test.o: osd_rmw_test.cpp malloc_or_die.h object_id.h osd_id.h osd_rmw.cpp osd_rmw.h test_pattern.h xor.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_secondary.o: osd_secondary.cpp blockstore.h cpp-btree/btree_map.h epoll_manager.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+osd_test.o: osd_test.cpp object_id.h osd_id.h osd_ops.h rw_blocking.h test_pattern.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+pg_states.o: pg_states.cpp pg_states.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+qemu_proxy.o: qemu_proxy.cpp cluster_client.h etcd_state_client.h http_client.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd_id.h osd_ops.h qemu_proxy.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+ringloop.o: ringloop.cpp ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+rw_blocking.o: rw_blocking.cpp rw_blocking.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+stub_bench.o: stub_bench.cpp object_id.h osd_id.h osd_ops.h rw_blocking.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+stub_osd.o: stub_osd.cpp object_id.h osd_id.h osd_ops.h rw_blocking.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+stub_uring_osd.o: stub_uring_osd.cpp epoll_manager.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd_id.h osd_ops.h ringloop.h timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+test.o: test.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+test_allocator.o: test_allocator.cpp allocator.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+test_blockstore.o: test_blockstore.cpp blockstore.h object_id.h ringloop.h timerfd_interval.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+timerfd_interval.o: timerfd_interval.cpp ringloop.h timerfd_interval.h
+	g++ $(CXXFLAGS) -c -o $@ $<
+timerfd_manager.o: timerfd_manager.cpp timerfd_manager.h
+	g++ $(CXXFLAGS) -c -o $@ $<

README.md

@@ -0,0 +1,387 @@
+## Vitastor
+
+## The Idea
+
+Make Software-Defined Block Storage Great Again.
+
+Vitastor is a small, simple and fast clustered block storage (storage for VM drives),
+architecturally similar to Ceph which means strong consistency, primary-replication, symmetric
+clustering and automatic data distribution over any number of drives of any size
+with configurable redundancy (replication or erasure codes/XOR).
+
+## Features
+
+Vitastor is currently a pre-release, a lot of features are missing and you can still expect
+breaking changes in the future. However, the following is implemented:
+
+- Basic part: highly-available block storage with symmetric clustering and no SPOF
+- Performance ;-D
+- Two redundancy schemes: Replication and XOR n+1 (simplest case of EC)
+- Configuration via simple JSON data structures in etcd
+- Automatic data distribution over OSDs, with support for:
+  - Mathematical optimization for better uniformity and less data movement
+  - Multiple pools
+  - Placement tree
+  - Configurable failure domains
+- Recovery of degraded blocks
+- Rebalancing (data movement between OSDs)
+- Lazy fsync support
+- I/O statistics reporting to etcd
+- Generic user-space client library
+- QEMU driver (built out-of-tree)
+- Loadable fio engine for benchmarks (also built out-of-tree)
+- NBD proxy for kernel mounts
+
+## Roadmap
+
+- Packaging for Debian and, probably, CentOS too
+- OSD creation tool (OSDs currently have to be created by hand)
+- Inode deletion tool (currently you can't delete anything :))
+- Other administrative tools
+- Per-inode I/O and space usage statistics
+- jerasure EC support with any number of data and parity drives in a group
+- Parallel usage of multiple network interfaces
+- Proxmox and OpenNebula plugins
+- iSCSI proxy
+- Inode metadata storage in etcd
+- Snapshots and copy-on-write image clones
+- Operation timeouts and better failure detection
+- Checksums
+- SSD+HDD optimizations, possibly including tiered storage and soft journal flushes
+- RDMA and NVDIMM support
+- Compression (possibly)
+- Read caching using system page cache (possibly)
+
+## Architecture
+
+Similarities:
+
+- Just like Ceph, Vitastor has Pools, PGs, OSDs, Monitors, Failure Domains, Placement Tree.
+- Just like Ceph, Vitastor is transactional (even though there's a "lazy fsync mode" which
+  doesn't implicitly flush every operation to disks).
+- OSDs also have journal and metadata and they can also be put on separate drives.
+- Just like in Ceph, client library attempts to recover from any cluster failure so
+  you can basically reboot the whole cluster and only pause, but not crash, your clients
+  (I consider this a bug if the client crashes in that case).
+
+Some basic terms for people not familiar with Ceph:
+
+- OSD (Object Storage Daemon) is a process that stores data and serves read/write requests.
+- PG (Placement Group) is a container for data that (normally) shares the same replicas.
+- Pool is a container for data that has the same redundancy scheme and placement rules.
+- Monitor is a separate daemon that watches cluster state and handles failures.
+- Failure Domain is a group of OSDs that you allow to fail. It's "host" by default.
+- Placement Tree groups OSDs in a hierarchy to later split them into Failure Domains.
+
+Architectural differences from Ceph:
+
+- Vitastor's primary focus is on SSDs. Proper SSD+HDD optimizations may be added in the future, though.
+- Vitastor OSD is (and will always be) single-threaded. If you want to dedicate more than 1 core
+  per drive you should run multiple OSDs each on a different partition of the drive.
+  Vitastor isn't CPU-hungry though (as opposed to Ceph), so 1 core is sufficient in a lot of cases.
+- Metadata and journal are always kept in memory. Metadata size depends linearly on drive capacity
+  and data store block size which is 128 KB by default. With 128 KB blocks, metadata should occupy
+  around 512 MB per 1 TB (which is still less than Ceph wants). Journal doesn't have to be big,
+  the example test below was conducted with only 16 MB journal. A big journal is probably even
+  harmful as dirty write metadata also take some memory.
+- Vitastor storage layer doesn't have internal copy-on-write or redirect-write. I know that maybe
+  it's possible to create a good copy-on-write storage, but it's much harder and makes performance
+  less deterministic, so CoW isn't used in Vitastor.
+- The basic layer of Vitastor is block storage with fixed-size blocks, not object storage with
+  rich semantics like in Ceph (RADOS).
+- There's a "lazy fsync" mode which allows to batch writes before flushing them to the disk.
+  This allows to use Vitastor with desktop SSDs, but still lowers performance due to additional
+  network roundtrips, so use server SSDs with capacitor-based power loss protection
+  ("Advanced Power Loss Protection") for best performance.
+- PGs are ephemeral. This means that they aren't stored on data disks and only exist in memory
+  while OSDs are running.
+- Recovery process is per-object (per-block), not per-PG. Also there are no PGLOGs.
+- Monitors don't store data. Cluster configuration and state is stored in etcd in simple human-readable
+  JSON structures. Monitors only watch cluster state and handle data movement.
+  Thus Vitastor's Monitor isn't a critical component of the system and is more similar to Ceph's Manager.
+  Vitastor's Monitor is implemented in node.js.
+- PG distribution isn't based on consistent hashes. All PG mappings are stored in etcd.
+  Rebalancing PGs between OSDs is done by mathematical optimization - data distribution problem
+  is reduced to a linear programming problem and solved by lp_solve. This allows for almost
+  perfect (96-99% uniformity compared to Ceph's 80-90%) data distribution in most cases, ability
+  to map PGs by hand without breaking rebalancing logic, reduced OSD peer-to-peer communication
+  (on average, OSDs have fewer peers) and less data movement. It also probably has a drawback -
+  this method may fail in very large clusters, but up to several hundreds of OSDs it's perfectly fine.
+  It's also easy to add consistent hashes in the future if something proves their necessity.
+- There's no separate CRUSH layer. You select pool redundancy scheme, placement root, failure domain
+  and so on directly in pool configuration.
+
+## Understanding Storage Performance
+
+The most important thing for fast storage is latency, not parallel iops.
+
+The best possible latency is achieved with one thread and queue depth of 1 which basically means
+"client load as low as possible". In this case IOPS = 1/latency, and this number doesn't
+scale with number of servers, drives, server processes or threads and so on.
+Single-threaded IOPS and latency numbers only depend on *how fast a single daemon is*.
+
+Why is it important? It's important because some of the applications *can't* use
+queue depth greater than 1 because their task isn't parallelizable. A notable example
+is any ACID DBMS because all of them write their WALs sequentially with fsync()s.
+
+fsync, by the way, is another important thing often missing in benchmarks. The point is
+that drives have cache buffers and don't guarantee that your data is actually persisted
+until you call fsync() which is translated to a FLUSH CACHE command by the OS.
+
+Desktop SSDs are very fast without fsync - NVMes, for example, can process ~80000 write
+operations per second with queue depth of 1 without fsync - but they're really slow with
+fsync because they have to actually write data to flash chips when you call fsync. Typical
+number is around 1000-2000 iops with fsync.
+
+Server SSDs often have supercapacitors that act as a built-in UPS and allow the drive
+to flush its DRAM cache to the persistent flash storage when a power loss occurs.
+This makes them perform equally well with and without fsync. This feature is called
+"Advanced Power Loss Protection" by Intel; other vendors either call it similarly
+or directly as "Full Capacitor-Based Power Loss Protection".
+
+All software-defined storages that I currently know are slow in terms of latency.
+Notable examples are Ceph and internal SDSes used by cloud providers like Amazon, Google,
+Yandex and so on. They're all slow and can only reach ~0.3ms read and ~0.6ms 4 KB write latency
+with best-in-slot hardware.
+
+And that's in the SSD era when you can buy an SSD that has ~0.04ms latency for 100 $.
+
+I use the following 6 commands with small variations to benchmark any storage:
+
+- Linear write:
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4M -iodepth=32 -rw=write -runtime=60 -filename=/dev/sdX`
+- Linear read:
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4M -iodepth=32 -rw=read -runtime=60 -filename=/dev/sdX`
+- Random write latency (this hurts storages the most):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=1 -fsync=1 -rw=randwrite -runtime=60 -filename=/dev/sdX`
+- Random read latency:
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=1 -rw=randread -runtime=60 -filename=/dev/sdX`
+- Parallel write iops (use numjobs if a single CPU core is insufficient to saturate the load):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=128 [-numjobs=4 -group_reporting] -rw=randwrite -runtime=60 -filename=/dev/sdX`
+- Parallel read iops (use numjobs if a single CPU core is insufficient to saturate the load):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=128 [-numjobs=4 -group_reporting] -rw=randread -runtime=60 -filename=/dev/sdX`
+
+## Vitastor's Theoretical Maximum Random Access Performance
+
+Replicated setups:
+- Single-threaded (T1Q1) read latency: 1 network roundtrip + 1 disk read.
+- Single-threaded write+fsync latency:
+  - With immediate commit: 2 network roundtrips + 1 disk write.
+  - With lazy commit: 4 network roundtrips + 1 disk write + 1 disk flush.
+- Saturated parallel read iops: min(network bandwidth, sum(disk read iops)).
+- Saturated parallel write iops: min(network bandwidth, sum(disk write iops / number of replicas / write amplification)).
+
+EC/XOR setups:
+- Single-threaded (T1Q1) read latency: 1.5 network roundtrips + 1 disk read.
+- Single-threaded write+fsync latency:
+  - With immediate commit: 3.5 network roundtrips + 1 disk read + 2 disk writes.
+  - With lazy commit: 5.5 network roundtrips + 1 disk read + 2 disk writes + 2 disk fsyncs.
+  - 0.5 in actually (k-1)/k which means that an additional roundtrip doesn't happen when
+    the read sub-operation can be served locally.
+- Saturated parallel read iops: min(network bandwidth, sum(disk read iops)).
+- Saturated parallel write iops: min(network bandwidth, sum(disk write iops * number of data drives / (number of data + parity drives) / write amplification)).
+  In fact, you should put disk write iops under the condition of ~10% reads / ~90% writes in this formula.
+
+Write amplification for 4 KB blocks is usually 3-5 in Vitastor:
+1. Journal block write
+2. Journal data write
+3. Metadata block write
+4. Another journal block write for EC/XOR setups
+5. Data block write
+
+If you manage to get an SSD which handles 512 byte blocks well (Optane?) you may
+lower 1, 3 and 4 to 512 bytes (1/8 of data size) and get WA as low as 2.375.
+
+Lazy fsync also reduces WA for parallel workloads because journal blocks are only
+written when they fill up or fsync is requested.
+
+## Example Comparison with Ceph
+
+Hardware configuration: 4 nodes, each with:
+- 6x SATA SSD Intel D3-4510 3.84 TB
+- 2x Xeon Gold 6242 (16 cores @ 2.8 GHz)
+- 384 GB RAM
+- 1x 25 GbE network interface (Mellanox ConnectX-4 LX), connected to a Juniper QFX5200 switch
+
+CPU powersaving was disabled. Both Vitastor and Ceph were configured with 2 OSDs per 1 SSD.
+
+All of the results below apply to 4 KB blocks.
+
+Raw drive performance:
+- T1Q1 write ~27000 iops (~0.037ms latency)
+- T1Q1 read ~9800 iops (~0.101ms latency)
+- T1Q32 write ~60000 iops
+- T1Q32 read ~81700 iops
+
+Ceph 15.2.4 (Bluestore):
+- T1Q1 write ~1000 iops (~1ms latency)
+- T1Q1 read ~1750 iops (~0.57ms latency)
+- T8Q64 write ~100000 iops, total CPU usage by OSDs about 40 virtual cores on each node
+- T8Q64 read ~480000 iops, total CPU usage by OSDs about 40 virtual cores on each node
+
+T8Q64 tests were conducted over 8 400GB RBD images from all hosts (every host was running 2 instances of fio).
+This is because Ceph has performance penalties related to running multiple clients over a single RBD image.
+
+cephx_sign_messages was set to false during tests, RocksDB and Bluestore settings were left at defaults.
+
+In fact, not that bad for Ceph. These servers are an example of well-balanced Ceph nodes.
+However, CPU usage and I/O latency were through the roof, as usual.
+
+Vitastor:
+- T1Q1 write: 7087 iops (0.14ms latency)
+- T1Q1 read: 6838 iops (0.145ms latency)
+- T2Q64 write: 162000 iops, total CPU usage by OSDs about 3 virtual cores on each node
+- T8Q64 read: 895000 iops, total CPU usage by OSDs about 4 virtual cores on each node
+
+T8Q64 read test was conducted over 1 larger inode (3.2T) from all hosts (every host was running 2 instances of fio).
+Vitastor has no performance penalties related to running multiple clients over a single inode.
+If conducted from one node with all primary OSDs moved to other nodes the result was slightly lower (689000 iops),
+this is because all operations resulted in network roundtrips between the client and the primary OSD.
+When fio was colocated with OSDs (like in Ceph benchmarks above), 1/4 of the read workload actually
+used the loopback network.
+
+Vitastor was configured with: `--disable_data_fsync true --immediate_commit all --flusher_count 8
+  --disk_alignment 4096 --journal_block_size 4096 --meta_block_size 4096
+  --journal_no_same_sector_overwrites true --journal_sector_buffer_count 1024
+  --journal_size 16777216`.
+
+### NBD
+
+NBD is currently required to mount Vitastor via kernel, but it imposes additional overhead
+due to additional copying between the kernel and userspace. This mostly hurts linear
+bandwidth, not iops.
+
+Vitastor with single-thread NBD on the same hardware:
+- T1Q1 write: 6000 iops (0.166ms latency)
+- T1Q1 read: 5518 iops (0.18ms latency)
+- T1Q128 write: 94400 iops
+- T1Q128 read: 103000 iops
+- Linear write (4M T1Q128): 1266 MB/s (compared to 2600 MB/s via fio)
+- Linear read (4M T1Q128): 975 MB/s (compared to 1400 MB/s via fio)
+
+## Building
+
+- Install Linux kernel 5.4 or newer for io_uring support.
+- Install liburing 0.4 or newer and its headers.
+- Install lp_solve.
+- Install etcd.
+- Install node.js 12 or newer.
+- Install gcc and g++ 9.x.
+- Clone https://yourcmc.ru/git/vitalif/vitastor/ with submodules.
+- Install QEMU 4.x or 5.x, get its source, begin to build it, stop the build and copy headers:
+   - `<qemu>/include` &rarr; `<vitastor>/qemu/include`
+   - Debian:
+      * Use qemu packages from the main repository
+      * `<qemu>/b/qemu/config-host.h` &rarr; `<vitastor>/qemu/b/qemu/config-host.h`
+      * `<qemu>/b/qemu/qapi` &rarr; `<vitastor>/qemu/b/qemu/qapi`
+   - CentOS 8:
+      * Use qemu packages from the Advanced-Virtualization repository. To enable it, run
+        `yum install centos-release-advanced-virtualization.noarch` and then `yum install qemu`
+      * `<qemu>/config-host.h` &rarr; `<vitastor>/qemu/b/qemu/config-host.h`
+      * `<qemu>/qapi` &rarr; `<vitastor>/qemu/b/qemu/qapi`
+   - `config-host.h` and `qapi` are required because they contain generated headers
+- Install fio 3.16, get its source and symlink it into `<vitastor>/fio`. It doesn't currently
+  build with fio 3.20 or newer due to the conflicts between g++ and gcc's atomics. This will
+  be fixed in the future.
+- Build Vitastor with `make -j8`.
+- Copy binaries somewhere.
+
+## Running
+
+Please note that startup procedure isn't currently simple - you specify configuration
+and calculate disk offsets almost by hand. This will be fixed in near future.
+
+- Get some SATA or NVMe SSDs with capacitors (server-grade drives). You can use desktop SSDs
+  with lazy fsync, but prepare for inferior single-thread latency.
+- Get a fast network (at least 10 Gbit/s).
+- Disable CPU powersaving: `cpupower idle-set -D 0 && cpupower frequency-set -g performance`.
+- Install etcd with `--max-txn-ops=100000 --auto-compaction-retention=10 --auto-compaction-mode=revision` options.
+- Create global configuration in etcd: `etcdctl put /vitastor/config/global '{"immediate_commit":"all"}'`
+  (if all your drives have capacitors).
+- Create pool configuration in etcd: `etcdctl put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"replicated","pg_size":2,"pg_minsize":1,"pg_count":256,"failure_domain":"host"}}'`.
+- Calculate offsets for your drives with `node ./mon/simple-offsets.js /dev/sdX`.
+- Make systemd units for your OSDs. Look at `./mon/make-units.sh` for example.
+  Notable configuration variables from the example:
+  - `disable_data_fsync 1` - only safe with server-grade drives with capacitors.
+  - `immediate_commit all` - use this if all your drives are server-grade.
+  - `disable_device_lock 1` - only required if you run multiple OSDs on one block device.
+  - `flusher_count 16` - flusher is a micro-thread that removes old data from the journal.
+    More flushers mean more aggressive journal flushing which allows for more throughput
+    but slightly hurts latency under less load. Flushing will probably be improved in the future
+    because currently high queue depths sometimes lead to performance degradation.
+  - `disk_alignment`, `journal_block_size`, `meta_block_size` should be set to the internal
+    block size of your SSDs which is 4096 on most drives.
+  - `journal_no_same_sector_overwrites true` prevents multiple overwrites of the same journal sector.
+    Some SSDs (like Intel D3-4510) don't like such overwrites so they benefit from this setting.
+    When this setting is set, it is also required to raise `journal_sector_buffer_count` setting,
+    which is the number of dirty journal sectors that may be written to at the same time.
+- `systemctl start vitastor.target` everywhere.
+- Start any number of monitors: `cd mon; node mon-main.js --etcd_url 'http://10.115.0.10:2379,http://10.115.0.11:2379,http://10.115.0.12:2379,http://10.115.0.13:2379' --etcd_prefix '/vitastor' --etcd_start_timeout 5`.
+- At this point, one of the monitors will configure PGs and OSDs will start them.
+- You can check PG states with `etcdctl get --prefix /vitastor/pg/state`. All PGs should become 'active'.
+- Run tests with (for example): `fio -thread -ioengine=./libfio_cluster.so -name=test -bs=4M -direct=1 -iodepth=16 -rw=write -etcd=10.115.0.10:2379/v3 -pool=1 -inode=1 -size=400G`.
+- Upload VM disk image with qemu-img (for example):
+  ```
+  LD_PRELOAD=./qemu_driver.so qemu-img convert -f qcow2 debian10.qcow2 -p
+    -O raw 'vitastor:etcd_host=10.115.0.10\:2379/v3:pool=1:inode=1:size=2147483648'
+  ```
+- Run QEMU with (for example):
+  ```
+  LD_PRELOAD=./qemu_driver.so qemu-system-x86_64 -enable-kvm -m 1024
+    -drive 'file=vitastor:etcd_host=10.115.0.10\:2379/v3:pool=1:inode=1:size=2147483648',format=raw,if=none,id=drive-virtio-disk0,cache=none
+    -device virtio-blk-pci,scsi=off,bus=pci.0,addr=0x5,drive=drive-virtio-disk0,id=virtio-disk0,bootindex=1,write-cache=off,physical_block_size=4096,logical_block_size=512
+    -vnc 0.0.0.0:0
+  ```
+
+## Known Problems
+
+- OSDs may currently crash with "can't get SQE, will fall out of sync with EPOLLET"
+  if you try to load them with very long iodepths because io_uring queue (ring) is limited
+  and OSDs don't check if it fills up.
+- Object deletion requests may currently lead to unfound objects on crashes because
+  proper handling of deletions in a cluster requires a "three-phase cleanup process"
+  and it's currently not implemented. In fact, even though deletion requests are
+  implemented, there's no user tool to delete anything from the cluster yet :).
+  Of course I'll create such tool, but its first implementation will be vulnerable to this issue.
+  It's not a big deal though, because you'll be able to just repeat the deletion request
+  in this case.
+
+## Implementation Principles
+
+- I like simple and stupid solutions, so expect Vitastor to stay simple.
+- I also like reinventing the wheel to some extent, like writing my own HTTP client
+  for etcd interaction instead of using prebuilt libraries, because in this case
+  I'm confident about what my code does and what it doesn't do.
+- I don't care about C++ "best practices" like RAII or proper inheritance or usage of
+  smart pointers or whatever and I don't intend to change my mind, so if you're here
+  looking for ideal reference C++ code, this probably isn't the right place.
+- I like node.js better than any other dynamically-typed language interpreter
+  because it's faster than any other interpreter in the world, has neutral C-like
+  syntax and built-in event loop. That's why Monitor is implemented in node.js.
+
+## Author and License
+
+Copyright (c) Vitaliy Filippov (vitalif [at] yourcmc.ru), 2019+
+
+You can also find me in the Russian Telegram Ceph chat: https://t.me/ceph_ru
+
+All server-side code (OSD, Monitor and so on) is licensed under the terms of
+Vitastor Network Public License 1.0 (VNPL 1.0), a copyleft license based on
+GNU GPLv3.0 with the additional "Network Interaction" clause which requires
+opensourcing all programs directly or indirectly interacting with Vitastor
+through a computer network ("Proxy Programs"). Proxy Programs may be made public
+not only under the terms of the same license, but also under the terms of any
+GPL-Compatible Free Software License, as listed by the Free Software Foundation.
+This is a stricter copyleft license than the Affero GPL.
+
+Basically, you can't use the software in a proprietary environment to provide
+its functionality to users without opensourcing all intermediary components
+standing between the user and Vitastor or purchasing a commercial license
+from the author 😀.
+
+Client libraries (cluster_client and so on) are dual-licensed under the same
+VNPL 1.0 and also GNU GPL 2.0 or later to allow for compatibility with GPLed
+software like QEMU and fio.
+
+You can find the full text of VNPL-1.0 in the file [VNPL-1.0.txt](VNPL-1.0.txt).
+GPL 2.0 is also included in this repository as [GPL-2.0.txt](GPL-2.0.txt).

VNPL-1.0.txt

@@ -0,0 +1,645 @@
+                     VITASTOR NETWORK PUBLIC LICENSE
+                       Version 1, 17 September 2020
+
+ Copyright (C) 2020 Vitaliy Filippov <vitalif@yourcmc.ru>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The Vitastor Network Public License is a free, copyleft license for
+software and other kinds of works, specifically designed to ensure
+cooperation with the community in the case of network server software.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+GNU General Public Licenses and Vitastor Network Public License are
+intended to guarantee your freedom to share and change all versions
+of a program--to make sure it remains free software for all its users.
+
+  When we speak of free software, we are referring to freedom, not
+price.  GNU General Public Licenses and Vitastor Network Public License
+are designed to make sure that you have the freedom to distribute copies
+of free software (and charge for them if you wish), that you receive
+source code or can get it if you want it, that you can change the software
+or use pieces of it in new free programs, and that you know you can do these
+things.
+
+  Developers that use GNU General Public Licenses and Vitastor
+Network Public License protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer
+you this License which gives you legal permission to copy, distribute
+and/or modify the software.
+
+  A secondary benefit of defending all users' freedom is that
+improvements made in alternate versions of the program, if they
+receive widespread use, become available for other developers to
+incorporate.  Many developers of free software are heartened and
+encouraged by the resulting cooperation.  However, in the case of
+software used on network servers, this result may fail to come about.
+The GNU General Public License permits making a modified version and
+letting the public access it on a server without ever releasing its
+source code to the public. Even the GNU Affero General Public License
+permits running a modified version in a closed environment where
+public users only interact with it through a closed-source proxy, again,
+without making the program and the proxy available to the public
+for free.
+
+  The Vitastor Network Public License is designed specifically to
+ensure that, in such cases, the modified program and the proxy stays
+available to the community. It requires the operator of a network server to
+provide the source code of the original program and all other programs
+communicating with it running there to the users of that server.
+Therefore, public use of a modified version, on a server accessible
+directly or indirectly to the public, gives the public access to the source
+code of the modified version.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 1 of the Vitastor Network Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Remote Network Interaction.
+
+  Notwithstanding any other provision of this License, if you provide
+any user an opportunity to interact with the covered work directly
+or indirectly through a computer network, an imitation of such network,
+or an additional program (hereinafter referred to as a "Proxy Program")
+that, in turn, interacts with the covered work through a computer network,
+an imitation of such network, or another Proxy Program itself,
+you must prominently offer that user an opportunity to receive the
+Corresponding Source of the covered work and all Proxy Programs from a
+network server at no charge, through some standard or customary means of
+facilitating copying of software. The Corresponding Source for the covered
+work must be made available under the conditions of this License, and
+the Corresponding Source for all Proxy Programs must be made available
+under the conditions of either this License or any GPL-Compatible
+Free Software License, as described by the Free Software Foundation
+in their "GPL-Compatible License List".
+
+  14. Revised Versions of this License.
+
+  Vitastor Author may publish revised and/or new versions of
+the Vitastor Network Public License from time to time.  Such new versions
+will be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the Vitastor Network
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version. If the Program does not specify a version
+number of the Vitastor Network Public License, you may choose any version
+ever published.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the Vitastor Network Public License as published by
+    the Vitastor Author, either version 1 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Vitastor Network Public License for more details.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If your software can interact with users remotely through a computer
+network, you should also make sure that it provides a way for users to
+get its source.  For example, if your program is a web application, its
+interface could display a "Source" link that leads users to an archive
+of the code.  There are many ways you could offer source, and different
+solutions will be better for different programs; see section 13 for the
+specific requirements.

allocator.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <stdexcept>
 #include "allocator.h"
 

allocator.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #pragma once
 
 #include <stdint.h>

base64.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "base64.h"
 
 std::string base64_encode(const std::string &in)

base64.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #pragma once
 #include <string>
 

blockstore.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 blockstore_t::blockstore_t(blockstore_config_t & config, ring_loop_t *ringloop)

blockstore.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #pragma once
 
 #ifndef _LARGEFILE64_SOURCE
@@ -27,13 +30,14 @@
 #define BS_OP_MIN 1
 #define BS_OP_READ 1
 #define BS_OP_WRITE 2
-#define BS_OP_SYNC 3
+#define BS_OP_WRITE_STABLE 3
-#define BS_OP_STABLE 4
+#define BS_OP_SYNC 4
-#define BS_OP_DELETE 5
+#define BS_OP_STABLE 5
-#define BS_OP_LIST 6
+#define BS_OP_DELETE 6
-#define BS_OP_ROLLBACK 7
+#define BS_OP_LIST 7
-#define BS_OP_SYNC_STAB_ALL 8
+#define BS_OP_ROLLBACK 8
-#define BS_OP_MAX 8
+#define BS_OP_SYNC_STAB_ALL 9
+#define BS_OP_MAX 9
 
 #define BS_OP_PRIVATE_DATA_SIZE 256
 
@@ -41,9 +45,9 @@
 
 Blockstore opcode documentation:
 
-## BS_OP_READ / BS_OP_WRITE
+## BS_OP_READ / BS_OP_WRITE / BS_OP_WRITE_STABLE
 
-Read or write object data.
+Read or write object data. WRITE_STABLE writes a version that doesn't require marking as stable.
 
 Input:
 - oid = requested object
@@ -113,6 +117,8 @@ Input:
 - oid.stripe = PG alignment
 - len = PG count or 0 to list all objects
 - offset = PG number
+- oid.inode = min inode number or 0 to list all inodes
+- version = max inode number or 0 to list all inodes
 
 Output:
 - retval = total obj_ver_id count

blockstore_flush.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 journal_flusher_t::journal_flusher_t(int flusher_count, blockstore_impl_t *bs)
@@ -7,10 +10,10 @@ journal_flusher_t::journal_flusher_t(int flusher_count, blockstore_impl_t *bs)
     dequeuing = false;
     active_flushers = 0;
     syncing_flushers = 0;
-    sync_threshold = bs->journal_block_size / sizeof(journal_entry_stable);
+    flusher_start_threshold = bs->journal_block_size / sizeof(journal_entry_stable);
-    journal_trim_interval = sync_threshold;
+    journal_trim_interval = flusher_start_threshold;
     journal_trim_counter = 0;
-    journal_superblock = bs->journal.inmemory ? bs->journal.buffer : memalign(MEM_ALIGNMENT, bs->journal_block_size);
+    journal_superblock = bs->journal.inmemory ? bs->journal.buffer : memalign_or_die(MEM_ALIGNMENT, bs->journal_block_size);
     co = new journal_flusher_co[flusher_count];
     for (int i = 0; i < flusher_count; i++)
     {
@@ -81,7 +84,7 @@ void journal_flusher_t::enqueue_flush(obj_ver_id ov)
         flush_versions[ov.oid] = ov.version;
         flush_queue.push_back(ov.oid);
     }
-    if (!dequeuing && flush_queue.size() >= sync_threshold)
+    if (!dequeuing && flush_queue.size() >= flusher_start_threshold)
     {
         dequeuing = true;
         bs->ringloop->wakeup();
@@ -101,19 +104,25 @@ void journal_flusher_t::unshift_flush(obj_ver_id ov)
         flush_versions[ov.oid] = ov.version;
         flush_queue.push_front(ov.oid);
     }
-    if (!dequeuing && flush_queue.size() >= sync_threshold)
+    if (!dequeuing && flush_queue.size() >= flusher_start_threshold)
     {
         dequeuing = true;
         bs->ringloop->wakeup();
     }
 }
 
-void journal_flusher_t::force_start()
+void journal_flusher_t::request_trim()
 {
     dequeuing = true;
+    trim_wanted++;
     bs->ringloop->wakeup();
 }
 
+void journal_flusher_t::release_trim()
+{
+    trim_wanted--;
+}
+
 #define await_sqe(label) \
     resume_##label:\
         sqe = bs->get_sqe();\
@@ -177,21 +186,11 @@ resume_0:
     dirty_end = bs->dirty_db.find(cur);
     if (dirty_end != bs->dirty_db.end())
     {
-        if (dirty_end->second.journal_sector >= bs->journal.dirty_start &&
-            (bs->journal.dirty_start >= bs->journal.used_start ||
-            dirty_end->second.journal_sector < bs->journal.used_start))
-        {
-            // We can't flush journal sectors that are still written to
-            flusher->enqueue_flush(cur);
-            flusher->dequeuing = false;
-            wait_state = 0;
-            return true;
-        }
         repeat_it = flusher->sync_to_repeat.find(cur.oid);
         if (repeat_it != flusher->sync_to_repeat.end())
         {
 #ifdef BLOCKSTORE_DEBUG
-            printf("Postpone %lu:%lu v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
+            printf("Postpone %lx:%lx v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
 #endif
             // We don't flush different parts of history of the same object in parallel
             // So we check if someone is already flushing this object
@@ -204,42 +203,112 @@ resume_0:
         }
         else
             flusher->sync_to_repeat[cur.oid] = 0;
+        if (dirty_end->second.journal_sector >= bs->journal.dirty_start &&
+            (bs->journal.dirty_start >= bs->journal.used_start ||
+            dirty_end->second.journal_sector < bs->journal.used_start))
+        {
+            flusher->enqueue_flush(cur);
+            // We can't flush journal sectors that are still written to
+            // However, as we group flushes by oid, current oid may have older writes to flush!
+            // And it may even block writes if we don't flush the older version
+            // (if it's in the beginning of the journal)...
+            // So first try to find an older version of the same object to flush.
+            bool found = false;
+            while (dirty_end != bs->dirty_db.begin())
+            {
+                dirty_end--;
+                if (dirty_end->first.oid != cur.oid)
+                {
+                    break;
+                }
+                if (!(dirty_end->second.journal_sector >= bs->journal.dirty_start &&
+                    (bs->journal.dirty_start >= bs->journal.used_start ||
+                    dirty_end->second.journal_sector < bs->journal.used_start)))
+                {
+                    found = true;
+                    cur.version = dirty_end->first.version;
+                    break;
+                }
+            }
+            if (!found)
+            {
+                // Try other objects
+                flusher->sync_to_repeat.erase(cur.oid);
+                int search_left = flusher->flush_queue.size() - 1;
 #ifdef BLOCKSTORE_DEBUG
-        printf("Flushing %lu:%lu v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
+                printf("Flusher overran writers (dirty_start=%08lx) - searching for older flushes (%d left)\n", bs->journal.dirty_start, search_left);
+#endif
+                while (search_left > 0)
+                {
+                    cur.oid = flusher->flush_queue.front();
+                    cur.version = flusher->flush_versions[cur.oid];
+                    flusher->flush_queue.pop_front();
+                    flusher->flush_versions.erase(cur.oid);
+                    dirty_end = bs->dirty_db.find(cur);
+                    if (dirty_end != bs->dirty_db.end())
+                    {
+                        if (dirty_end->second.journal_sector >= bs->journal.dirty_start &&
+                            (bs->journal.dirty_start >= bs->journal.used_start ||
+                            dirty_end->second.journal_sector < bs->journal.used_start))
+                        {
+#ifdef BLOCKSTORE_DEBUG
+                            printf("Write %lx:%lx v%lu is too new: offset=%08lx\n", cur.oid.inode, cur.oid.stripe, cur.version, dirty_end->second.journal_sector);
+#endif
+                            flusher->enqueue_flush(cur);
+                        }
+                        else
+                        {
+                            repeat_it = flusher->sync_to_repeat.find(cur.oid);
+                            if (repeat_it == flusher->sync_to_repeat.end())
+                            {
+                                flusher->sync_to_repeat[cur.oid] = 0;
+                                break;
+                            }
+                        }
+                    }
+                    search_left--;
+                }
+                if (search_left <= 0)
+                {
+#ifdef BLOCKSTORE_DEBUG
+                    printf("No older flushes, stopping\n");
+#endif
+                    flusher->dequeuing = false;
+                    wait_state = 0;
+                    return true;
+                }
+            }
+        }
+#ifdef BLOCKSTORE_DEBUG
+        printf("Flushing %lx:%lx v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
 #endif
         flusher->active_flushers++;
 resume_1:
+        // Find it in clean_db
+        clean_it = bs->clean_db.find(cur.oid);
+        old_clean_loc = (clean_it != bs->clean_db.end() ? clean_it->second.location : UINT64_MAX);
         // Scan dirty versions of the object
         if (!scan_dirty(1))
         {
             wait_state += 1;
             return false;
         }
-        if (copy_count == 0 && clean_loc == UINT64_MAX && !has_delete && !has_empty)
+        // Writes and deletes shouldn't happen at the same time
+        assert(!(copy_count > 0 || has_writes) || !has_delete);
+        if (copy_count == 0 && !has_writes && !has_delete || has_delete && old_clean_loc == UINT64_MAX)
         {
             // Nothing to flush
-            flusher->active_flushers--;
+            bs->erase_dirty(dirty_start, std::next(dirty_end), clean_loc);
-            repeat_it = flusher->sync_to_repeat.find(cur.oid);
+            goto trim_journal;
-            if (repeat_it != flusher->sync_to_repeat.end() && repeat_it->second > cur.version)
-            {
-                // Requeue version
-                flusher->unshift_flush({ .oid = cur.oid, .version = repeat_it->second });
-            }
-            flusher->sync_to_repeat.erase(repeat_it);
-            wait_state = 0;
-            goto resume_0;
         }
-        // Find it in clean_db
-        clean_it = bs->clean_db.find(cur.oid);
-        old_clean_loc = (clean_it != bs->clean_db.end() ? clean_it->second.location : UINT64_MAX);
         if (clean_loc == UINT64_MAX)
         {
-            if (copy_count > 0 && has_delete || old_clean_loc == UINT64_MAX)
+            if (old_clean_loc == UINT64_MAX)
             {
                 // Object not allocated. This is a bug.
                 char err[1024];
                 snprintf(
-                    err, 1024, "BUG: Object %lu:%lu v%lu that we are trying to flush is not allocated on the data device",
+                    err, 1024, "BUG: Object %lx:%lx v%lu that we are trying to flush is not allocated on the data device",
                     cur.oid.inode, cur.oid.stripe, cur.version
                 );
                 throw std::runtime_error(err);
@@ -405,8 +474,9 @@ resume_1:
         }
         // Update clean_db and dirty_db, free old data locations
         update_clean_db();
+    trim_journal:
         // Clear unused part of the journal every <journal_trim_interval> flushes
-        if (!((++flusher->journal_trim_counter) % flusher->journal_trim_interval))
+        if (!((++flusher->journal_trim_counter) % flusher->journal_trim_interval) || flusher->trim_wanted > 0)
         {
             flusher->journal_trim_counter = 0;
             if (bs->journal.trim())
@@ -436,7 +506,8 @@ resume_1:
         }
         // All done
 #ifdef BLOCKSTORE_DEBUG
-        printf("Flushed %lu:%lu v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
+        printf("Flushed %lx:%lx v%lu (%d copies, wr:%d, del:%d), %ld left\n", cur.oid.inode, cur.oid.stripe, cur.version,
+            copy_count, has_writes, has_delete, flusher->flush_queue.size());
 #endif
         flusher->active_flushers--;
         repeat_it = flusher->sync_to_repeat.find(cur.oid);
@@ -464,19 +535,25 @@ bool journal_flusher_co::scan_dirty(int wait_base)
     copy_count = 0;
     clean_loc = UINT64_MAX;
     has_delete = false;
-    has_empty = false;
+    has_writes = false;
     skip_copy = false;
     clean_init_bitmap = false;
     while (1)
     {
-        if (dirty_it->second.state == ST_J_STABLE && !skip_copy)
+        if (!IS_STABLE(dirty_it->second.state))
+        {
+            char err[1024];
+            snprintf(
+                err, 1024, "BUG: Unexpected dirty_entry %lx:%lx v%lu state during flush: %d",
+                dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version, dirty_it->second.state
+            );
+            throw std::runtime_error(err);
+        }
+        else if (IS_JOURNAL(dirty_it->second.state) && !skip_copy)
         {
             // First we submit all reads
-            if (dirty_it->second.len == 0)
+            has_writes = true;
-            {
+            if (dirty_it->second.len != 0)
-                has_empty = true;
-            }
-            else
             {
                 offset = dirty_it->second.offset;
                 end_offset = dirty_it->second.offset + dirty_it->second.len;
@@ -490,18 +567,18 @@ bool journal_flusher_co::scan_dirty(int wait_base)
                     {
                         submit_offset = dirty_it->second.location + offset - dirty_it->second.offset;
                         submit_len = it == v.end() || it->offset >= end_offset ? end_offset-offset : it->offset-offset;
-                        it = v.insert(it, (copy_buffer_t){ .offset = offset, .len = submit_len, .buf = memalign(MEM_ALIGNMENT, submit_len) });
+                        it = v.insert(it, (copy_buffer_t){ .offset = offset, .len = submit_len, .buf = memalign_or_die(MEM_ALIGNMENT, submit_len) });
                         copy_count++;
                         if (bs->journal.inmemory)
                         {
                             // Take it from memory
-                            memcpy(v.back().buf, bs->journal.buffer + submit_offset, submit_len);
+                            memcpy(it->buf, bs->journal.buffer + submit_offset, submit_len);
                         }
                         else
                         {
                             // Read it from disk
                             await_sqe(0);
-                            data->iov = (struct iovec){ v.back().buf, (size_t)submit_len };
+                            data->iov = (struct iovec){ it->buf, (size_t)submit_len };
                             data->callback = simple_callback_r;
                             my_uring_prep_readv(
                                 sqe, bs->journal.fd, &data->iov, 1, bs->journal.offset + submit_offset
@@ -515,30 +592,22 @@ bool journal_flusher_co::scan_dirty(int wait_base)
                 }
             }
         }
-        else if (dirty_it->second.state == ST_D_STABLE && !skip_copy)
+        else if (IS_BIG_WRITE(dirty_it->second.state) && !skip_copy)
         {
             // There is an unflushed big write. Copy small writes in its position
+            has_writes = true;
             clean_loc = dirty_it->second.location;
             clean_init_bitmap = true;
             clean_bitmap_offset = dirty_it->second.offset;
             clean_bitmap_len = dirty_it->second.len;
             skip_copy = true;
         }
-        else if (dirty_it->second.state == ST_DEL_STABLE && !skip_copy)
+        else if (IS_DELETE(dirty_it->second.state) && !skip_copy)
         {
             // There is an unflushed delete
             has_delete = true;
             skip_copy = true;
         }
-        else if (!IS_STABLE(dirty_it->second.state))
-        {
-            char err[1024];
-            snprintf(
-                err, 1024, "BUG: Unexpected dirty_entry %lu:%lu v%lu state during flush: %d",
-                dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version, dirty_it->second.state
-            );
-            throw std::runtime_error(err);
-        }
         dirty_start = dirty_it;
         if (dirty_it == bs->dirty_db.begin())
         {
@@ -574,7 +643,7 @@ bool journal_flusher_co::modify_meta_read(uint64_t meta_loc, flusher_meta_write_
     if (wr.it == flusher->meta_sectors.end())
     {
         // Not in memory yet, read it
-        wr.buf = memalign(MEM_ALIGNMENT, bs->meta_block_size);
+        wr.buf = memalign_or_die(MEM_ALIGNMENT, bs->meta_block_size);
         wr.it = flusher->meta_sectors.emplace(wr.sector, (meta_sector_t){
             .offset = wr.sector,
             .len = bs->meta_block_size,
@@ -604,7 +673,7 @@ void journal_flusher_co::update_clean_db()
     if (old_clean_loc != UINT64_MAX && old_clean_loc != clean_loc)
     {
 #ifdef BLOCKSTORE_DEBUG
-        printf("Free block %lu\n", old_clean_loc >> bs->block_order);
+        printf("Free block %lu (new location is %lu)\n", old_clean_loc >> bs->block_order, clean_loc >> bs->block_order);
 #endif
         bs->data_alloc->set(old_clean_loc >> bs->block_order, false);
     }

blockstore_flush.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 struct copy_buffer_t
 {
     uint64_t offset, len;
@@ -45,7 +48,7 @@ class journal_flusher_co
     std::map<object_id, uint64_t>::iterator repeat_it;
     std::function<void(ring_data_t*)> simple_callback_r, simple_callback_w;
 
-    bool skip_copy, has_delete, has_empty;
+    bool skip_copy, has_delete, has_writes;
     blockstore_clean_db_t::iterator clean_it;
     std::vector<copy_buffer_t> v;
     std::vector<copy_buffer_t>::iterator it;
@@ -73,9 +76,10 @@ public:
 // Journal flusher itself
 class journal_flusher_t
 {
+    int trim_wanted = 0;
     bool dequeuing;
     int flusher_count;
-    int sync_threshold;
+    int flusher_start_threshold;
     journal_flusher_co *co;
     blockstore_impl_t *bs;
     friend class journal_flusher_co;
@@ -96,7 +100,8 @@ public:
     ~journal_flusher_t();
     void loop();
     bool is_active();
-    void force_start();
+    void request_trim();
+    void release_trim();
     void enqueue_flush(obj_ver_id oid);
     void unshift_flush(obj_ver_id oid);
 };

blockstore_impl.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 blockstore_impl_t::blockstore_impl_t(blockstore_config_t & config, ring_loop_t *ringloop)
@@ -7,7 +10,7 @@ blockstore_impl_t::blockstore_impl_t(blockstore_config_t & config, ring_loop_t *
     ring_consumer.loop = [this]() { loop(); };
     ringloop->register_consumer(&ring_consumer);
     initialized = 0;
-    zero_object = (uint8_t*)memalign(MEM_ALIGNMENT, block_size);
+    zero_object = (uint8_t*)memalign_or_die(MEM_ALIGNMENT, block_size);
     data_fd = meta_fd = journal.fd = -1;
     parse_config(config);
     try
@@ -124,19 +127,13 @@ void blockstore_impl_t::loop()
             if (PRIV(op)->wait_for)
             {
                 check_wait(op);
-#ifdef BLOCKSTORE_DEBUG
-                if (PRIV(op)->wait_for)
-                {
-                    printf("still waiting for %d\n", PRIV(op)->wait_for);
-                }
-#endif
                 if (PRIV(op)->wait_for == WAIT_SQE)
                 {
                     break;
                 }
                 else if (PRIV(op)->wait_for)
                 {
-                    if (op->opcode == BS_OP_WRITE || op->opcode == BS_OP_DELETE)
+                    if (op->opcode == BS_OP_WRITE || op->opcode == BS_OP_WRITE_STABLE || op->opcode == BS_OP_DELETE)
                     {
                         has_writes = 2;
                     }
@@ -150,7 +147,7 @@ void blockstore_impl_t::loop()
             {
                 dequeue_op = dequeue_read(op);
             }
-            else if (op->opcode == BS_OP_WRITE || op->opcode == BS_OP_DELETE)
+            else if (op->opcode == BS_OP_WRITE || op->opcode == BS_OP_WRITE_STABLE)
             {
                 if (has_writes == 2)
                 {
@@ -160,6 +157,16 @@ void blockstore_impl_t::loop()
                 dequeue_op = dequeue_write(op);
                 has_writes = dequeue_op ? 1 : 2;
             }
+            else if (op->opcode == BS_OP_DELETE)
+            {
+                if (has_writes == 2)
+                {
+                    // Some writes could not be submitted
+                    break;
+                }
+                dequeue_op = dequeue_del(op);
+                has_writes = dequeue_op ? 1 : 2;
+            }
             else if (op->opcode == BS_OP_SYNC)
             {
                 // wait for all small writes to be submitted
@@ -271,6 +278,9 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
         if (ringloop->space_left() < PRIV(op)->wait_detail)
         {
             // stop submission if there's still no free space
+#ifdef BLOCKSTORE_DEBUG
+            printf("Still waiting for %lu SQE(s)\n", PRIV(op)->wait_detail);
+#endif
             return;
         }
         PRIV(op)->wait_for = 0;
@@ -280,8 +290,12 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
         if (journal.used_start == PRIV(op)->wait_detail)
         {
             // do not submit
+#ifdef BLOCKSTORE_DEBUG
+            printf("Still waiting to flush journal offset %08lx\n", PRIV(op)->wait_detail);
+#endif
             return;
         }
+        flusher->release_trim();
         PRIV(op)->wait_for = 0;
     }
     else if (PRIV(op)->wait_for == WAIT_JOURNAL_BUFFER)
@@ -291,6 +305,9 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
             journal.sector_info[next].dirty)
         {
             // do not submit
+#ifdef BLOCKSTORE_DEBUG
+            printf("Still waiting for a journal buffer\n");
+#endif
             return;
         }
         PRIV(op)->wait_for = 0;
@@ -299,6 +316,9 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
     {
         if (!data_alloc->get_free_count() && !flusher->is_active())
         {
+#ifdef BLOCKSTORE_DEBUG
+            printf("Still waiting for free space on the data device\n");
+#endif
             return;
         }
         PRIV(op)->wait_for = 0;
@@ -312,13 +332,13 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
 void blockstore_impl_t::enqueue_op(blockstore_op_t *op, bool first)
 {
     if (op->opcode < BS_OP_MIN || op->opcode > BS_OP_MAX ||
-        ((op->opcode == BS_OP_READ || op->opcode == BS_OP_WRITE) && (
+        ((op->opcode == BS_OP_READ || op->opcode == BS_OP_WRITE || op->opcode == BS_OP_WRITE_STABLE) && (
             op->offset >= block_size ||
             op->len > block_size-op->offset ||
             (op->len % disk_alignment)
         )) ||
         readonly && op->opcode != BS_OP_READ && op->opcode != BS_OP_LIST ||
-        first && op->opcode == BS_OP_WRITE)
+        first && (op->opcode == BS_OP_WRITE || op->opcode == BS_OP_WRITE_STABLE))
     {
         // Basic verification not passed
         op->retval = -EINVAL;
@@ -363,7 +383,7 @@ void blockstore_impl_t::enqueue_op(blockstore_op_t *op, bool first)
             }
         };
     }
-    if (op->opcode == BS_OP_WRITE && !enqueue_write(op))
+    if ((op->opcode == BS_OP_WRITE || op->opcode == BS_OP_WRITE_STABLE || op->opcode == BS_OP_DELETE) && !enqueue_write(op))
     {
         std::function<void (blockstore_op_t*)>(op->callback)(op);
         return;
@@ -390,82 +410,201 @@ void blockstore_impl_t::enqueue_op(blockstore_op_t *op, bool first)
     ringloop->wakeup();
 }
 
+static bool replace_stable(object_id oid, uint64_t version, int search_start, int search_end, obj_ver_id* list)
+{
+    while (search_start < search_end)
+    {
+        int pos = search_start+(search_end-search_start)/2;
+        if (oid < list[pos].oid)
+        {
+            search_end = pos;
+        }
+        else if (list[pos].oid < oid)
+        {
+            search_start = pos+1;
+        }
+        else
+        {
+            list[pos].version = version;
+            return true;
+        }
+    }
+    return false;
+}
+
 void blockstore_impl_t::process_list(blockstore_op_t *op)
 {
-    // Count objects
     uint32_t list_pg = op->offset;
     uint32_t pg_count = op->len;
     uint64_t pg_stripe_size = op->oid.stripe;
+    uint64_t min_inode = op->oid.inode;
+    uint64_t max_inode = op->version;
+    // Check PG
     if (pg_count != 0 && (pg_stripe_size < MIN_BLOCK_SIZE || list_pg >= pg_count))
     {
         op->retval = -EINVAL;
         FINISH_OP(op);
         return;
     }
-    uint64_t stable_count = 0;
+    // Copy clean_db entries (sorted)
-    if (pg_count > 0)
+    int stable_count = 0, stable_alloc = clean_db.size() / (pg_count ? pg_count : 1);
-    {
+    obj_ver_id *stable = (obj_ver_id*)malloc(sizeof(obj_ver_id) * stable_alloc);
-        for (auto it = clean_db.begin(); it != clean_db.end(); it++)
+    if (!stable)
-        {
-            uint32_t pg = (it->first.inode + it->first.stripe / pg_stripe_size) % pg_count;
-            if (pg == list_pg)
-            {
-                stable_count++;
-            }
-        }
-    }
-    else
-    {
-        stable_count = clean_db.size();
-    }
-    uint64_t total_count = stable_count;
-    for (auto it = dirty_db.begin(); it != dirty_db.end(); it++)
-    {
-        if (!pg_count || ((it->first.oid.inode + it->first.oid.stripe / pg_stripe_size) % pg_count) == list_pg)
-        {
-            if (IS_STABLE(it->second.state))
-            {
-                stable_count++;
-            }
-            total_count++;
-        }
-    }
-    // Allocate memory
-    op->version = stable_count;
-    op->retval = total_count;
-    op->buf = malloc(sizeof(obj_ver_id) * total_count);
-    if (!op->buf)
     {
         op->retval = -ENOMEM;
         FINISH_OP(op);
         return;
     }
-    obj_ver_id *vers = (obj_ver_id*)op->buf;
-    int i = 0;
-    for (auto it = clean_db.begin(); it != clean_db.end(); it++)
     {
-        if (!pg_count || ((it->first.inode + it->first.stripe / pg_stripe_size) % pg_count) == list_pg)
+        auto clean_it = clean_db.begin(), clean_end = clean_db.end();
+        if ((min_inode != 0 || max_inode != 0) && min_inode <= max_inode)
         {
-            vers[i++] = {
+            clean_it = clean_db.lower_bound({
-                .oid = it->first,
+                .inode = min_inode,
-                .version = it->second.version,
+                .stripe = 0,
-            };
+            });
+            clean_end = clean_db.upper_bound({
+                .inode = max_inode,
+                .stripe = UINT64_MAX,
+            });
         }
-    }
+        for (; clean_it != clean_end; clean_it++)
-    int j = stable_count;
-    for (auto it = dirty_db.begin(); it != dirty_db.end(); it++)
-    {
-        if (!pg_count || ((it->first.oid.inode + it->first.oid.stripe / pg_stripe_size) % pg_count) == list_pg)
         {
-            if (IS_STABLE(it->second.state))
+            if (!pg_count || ((clean_it->first.inode + clean_it->first.stripe / pg_stripe_size) % pg_count) == list_pg)
             {
-                vers[i++] = it->first;
+                if (stable_count >= stable_alloc)
-            }
+                {
-            else
+                    stable_alloc += 32768;
-            {
+                    stable = (obj_ver_id*)realloc(stable, sizeof(obj_ver_id) * stable_alloc);
-                vers[j++] = it->first;
+                    if (!stable)
+                    {
+                        op->retval = -ENOMEM;
+                        FINISH_OP(op);
+                        return;
+                    }
+                }
+                stable[stable_count++] = {
+                    .oid = clean_it->first,
+                    .version = clean_it->second.version,
+                };
             }
         }
     }
+    int clean_stable_count = stable_count;
+    // Copy dirty_db entries (sorted, too)
+    int unstable_count = 0, unstable_alloc = 0;
+    obj_ver_id *unstable = NULL;
+    {
+        auto dirty_it = dirty_db.begin(), dirty_end = dirty_db.end();
+        if ((min_inode != 0 || max_inode != 0) && min_inode <= max_inode)
+        {
+            dirty_it = dirty_db.lower_bound({
+                .oid = {
+                    .inode = min_inode,
+                    .stripe = 0,
+                },
+                .version = 0,
+            });
+            dirty_end = dirty_db.upper_bound({
+                .oid = {
+                    .inode = max_inode,
+                    .stripe = UINT64_MAX,
+                },
+                .version = UINT64_MAX,
+            });
+        }
+        for (; dirty_it != dirty_end; dirty_it++)
+        {
+            if (!pg_count || ((dirty_it->first.oid.inode + dirty_it->first.oid.stripe / pg_stripe_size) % pg_count) == list_pg)
+            {
+                if (IS_DELETE(dirty_it->second.state))
+                {
+                    // Deletions are always stable, so try to zero out two possible entries
+                    if (!replace_stable(dirty_it->first.oid, 0, 0, clean_stable_count, stable))
+                    {
+                        replace_stable(dirty_it->first.oid, 0, clean_stable_count, stable_count, stable);
+                    }
+                }
+                else if (IS_STABLE(dirty_it->second.state))
+                {
+                    // First try to replace a clean stable version in the first part of the list
+                    if (!replace_stable(dirty_it->first.oid, dirty_it->first.version, 0, clean_stable_count, stable))
+                    {
+                        // Then try to replace the last dirty stable version in the second part of the list
+                        if (stable[stable_count-1].oid == dirty_it->first.oid)
+                        {
+                            stable[stable_count-1].version = dirty_it->first.version;
+                        }
+                        else
+                        {
+                            if (stable_count >= stable_alloc)
+                            {
+                                stable_alloc += 32768;
+                                stable = (obj_ver_id*)realloc(stable, sizeof(obj_ver_id) * stable_alloc);
+                                if (!stable)
+                                {
+                                    if (unstable)
+                                        free(unstable);
+                                    op->retval = -ENOMEM;
+                                    FINISH_OP(op);
+                                    return;
+                                }
+                            }
+                            stable[stable_count++] = dirty_it->first;
+                        }
+                    }
+                }
+                else
+                {
+                    if (unstable_count >= unstable_alloc)
+                    {
+                        unstable_alloc += 32768;
+                        unstable = (obj_ver_id*)realloc(unstable, sizeof(obj_ver_id) * unstable_alloc);
+                        if (!unstable)
+                        {
+                            if (stable)
+                                free(stable);
+                            op->retval = -ENOMEM;
+                            FINISH_OP(op);
+                            return;
+                        }
+                    }
+                    unstable[unstable_count++] = dirty_it->first;
+                }
+            }
+        }
+    }
+    // Remove zeroed out stable entries
+    int j = 0;
+    for (int i = 0; i < stable_count; i++)
+    {
+        if (stable[i].version != 0)
+        {
+            stable[j++] = stable[i];
+        }
+    }
+    stable_count = j;
+    if (stable_count+unstable_count > stable_alloc)
+    {
+        stable_alloc = stable_count+unstable_count;
+        stable = (obj_ver_id*)realloc(stable, sizeof(obj_ver_id) * stable_alloc);
+        if (!stable)
+        {
+            if (unstable)
+                free(unstable);
+            op->retval = -ENOMEM;
+            FINISH_OP(op);
+            return;
+        }
+    }
+    // Copy unstable entries
+    for (int i = 0; i < unstable_count; i++)
+    {
+        stable[j++] = unstable[i];
+    }
+    free(unstable);
+    op->version = stable_count;
+    op->retval = stable_count+unstable_count;
+    op->buf = stable;
     FINISH_OP(op);
 }

blockstore_impl.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #pragma once
 
 #include "blockstore.h"
@@ -7,7 +10,6 @@
 #include <sys/stat.h>
 #include <fcntl.h>
 #include <unistd.h>
-#include <malloc.h>
 #include <linux/fs.h>
 
 #include <vector>
@@ -17,45 +19,38 @@
 
 #include "cpp-btree/btree_map.h"
 
+#include "malloc_or_die.h"
 #include "allocator.h"
 
 //#define BLOCKSTORE_DEBUG
 
 // States are not stored on disk. Instead, they're deduced from the journal
-// FIXME: Rename to BS_ST_*
 
-#define ST_J_WAIT_BIG 1
+#define BS_ST_SMALL_WRITE 0x01
-#define ST_J_IN_FLIGHT 2
+#define BS_ST_BIG_WRITE 0x02
-#define ST_J_SUBMITTED 3
+#define BS_ST_DELETE 0x03
-#define ST_J_WRITTEN 4
-#define ST_J_SYNCED 5
-#define ST_J_STABLE 6
 
-#define ST_D_IN_FLIGHT 15
+#define BS_ST_WAIT_BIG 0x10
-#define ST_D_SUBMITTED 16
+#define BS_ST_IN_FLIGHT 0x20
-#define ST_D_WRITTEN 17
+#define BS_ST_SUBMITTED 0x30
-#define ST_D_SYNCED 20
+#define BS_ST_WRITTEN 0x40
-#define ST_D_STABLE 21
+#define BS_ST_SYNCED 0x50
+#define BS_ST_STABLE 0x60
 
-#define ST_DEL_IN_FLIGHT 31
+#define BS_ST_INSTANT 0x100
-#define ST_DEL_SUBMITTED 32
-#define ST_DEL_WRITTEN 33
-#define ST_DEL_SYNCED 34
-#define ST_DEL_STABLE 35
-
-#define ST_CURRENT 48
 
 #define IMMEDIATE_NONE 0
 #define IMMEDIATE_SMALL 1
 #define IMMEDIATE_ALL 2
 
-#define IS_IN_FLIGHT(st) (st == ST_J_WAIT_BIG || st == ST_J_IN_FLIGHT || st == ST_D_IN_FLIGHT || st == ST_DEL_IN_FLIGHT || st == ST_J_SUBMITTED || st == ST_D_SUBMITTED || st == ST_DEL_SUBMITTED)
+#define BS_ST_TYPE_MASK 0x0F
-#define IS_STABLE(st) (st == ST_J_STABLE || st == ST_D_STABLE || st == ST_DEL_STABLE || st == ST_CURRENT)
+#define BS_ST_WORKFLOW_MASK 0xF0
-#define IS_SYNCED(st) (IS_STABLE(st) || st == ST_J_SYNCED || st == ST_D_SYNCED || st == ST_DEL_SYNCED)
+#define IS_IN_FLIGHT(st) (((st) & 0xF0) <= BS_ST_SUBMITTED)
-#define IS_JOURNAL(st) (st >= ST_J_WAIT_BIG && st <= ST_J_STABLE)
+#define IS_STABLE(st) (((st) & 0xF0) == BS_ST_STABLE)
-#define IS_BIG_WRITE(st) (st >= ST_D_IN_FLIGHT && st <= ST_D_STABLE)
+#define IS_SYNCED(st) (((st) & 0xF0) >= BS_ST_SYNCED)
-#define IS_DELETE(st) (st >= ST_DEL_IN_FLIGHT && st <= ST_DEL_STABLE)
+#define IS_JOURNAL(st) (((st) & 0x0F) == BS_ST_SMALL_WRITE)
-#define IS_UNSYNCED(st) (st >= ST_J_WAIT_BIG && st <= ST_J_WRITTEN || st >= ST_D_IN_FLIGHT && st <= ST_D_WRITTEN|| st >= ST_DEL_IN_FLIGHT && st <= ST_DEL_WRITTEN)
+#define IS_BIG_WRITE(st) (((st) & 0x0F) == BS_ST_BIG_WRITE)
+#define IS_DELETE(st) (((st) & 0x0F) == BS_ST_DELETE)
 
 #define BS_SUBMIT_GET_SQE(sqe, data) \
     BS_SUBMIT_GET_ONLY_SQE(sqe); \
@@ -286,12 +281,14 @@ class blockstore_impl_t
     // Stabilize
     int dequeue_stable(blockstore_op_t *op);
     int continue_stable(blockstore_op_t *op);
+    void mark_stable(const obj_ver_id & ov);
     void handle_stable_event(ring_data_t *data, blockstore_op_t *op);
     void stabilize_object(object_id oid, uint64_t max_ver);
 
     // Rollback
     int dequeue_rollback(blockstore_op_t *op);
     int continue_rollback(blockstore_op_t *op);
+    void mark_rolled_back(const obj_ver_id & ov);
     void handle_rollback_event(ring_data_t *data, blockstore_op_t *op);
     void erase_dirty(blockstore_dirty_db_t::iterator dirty_start, blockstore_dirty_db_t::iterator dirty_end, uint64_t clean_loc);
 

blockstore_init.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 blockstore_init_meta::blockstore_init_meta(blockstore_impl_t *bs)
@@ -108,13 +111,13 @@ void blockstore_init_meta::handle_entries(void* entries, unsigned count, int blo
                 {
                     // free the previous block
 #ifdef BLOCKSTORE_DEBUG
-                    printf("Free block %lu\n", clean_it->second.location >> bs->block_order);
+                    printf("Free block %lu (new location is %lu)\n", clean_it->second.location >> block_order, done_cnt+i >> block_order);
 #endif
                     bs->data_alloc->set(clean_it->second.location >> block_order, false);
                 }
                 entries_loaded++;
 #ifdef BLOCKSTORE_DEBUG
-                printf("Allocate block (clean entry) %lu: %lu:%lu v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
+                printf("Allocate block (clean entry) %lu: %lx:%lx v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
 #endif
                 bs->data_alloc->set(done_cnt+i, true);
                 bs->clean_db[entry->oid] = (struct clean_entry){
@@ -125,7 +128,7 @@ void blockstore_init_meta::handle_entries(void* entries, unsigned count, int blo
             else
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("Old clean entry %lu: %lu:%lu v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
+                printf("Old clean entry %lu: %lx:%lx v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
 #endif
             }
         }
@@ -202,11 +205,7 @@ int blockstore_init_journal::loop()
         goto resume_7;
     printf("Reading blockstore journal\n");
     if (!bs->journal.inmemory)
-    {
+        submitted_buf = memalign_or_die(MEM_ALIGNMENT, 2*bs->journal.block_size);
-        submitted_buf = memalign(MEM_ALIGNMENT, 2*bs->journal.block_size);
-        if (!submitted_buf)
-            throw std::bad_alloc();
-    }
     else
         submitted_buf = bs->journal.buffer;
     // Read first block of the journal
@@ -317,7 +316,7 @@ resume_1:
                 if (journal_pos < bs->journal.used_start)
                     end = bs->journal.used_start;
                 if (!bs->journal.inmemory)
-                    submitted_buf = memalign(MEM_ALIGNMENT, JOURNAL_BUFFER_SIZE);
+                    submitted_buf = memalign_or_die(MEM_ALIGNMENT, JOURNAL_BUFFER_SIZE);
                 else
                     submitted_buf = bs->journal.buffer + journal_pos;
                 data->iov = {
@@ -404,7 +403,7 @@ resume_1:
     bs->journal.trim();
     bs->journal.dirty_start = bs->journal.next_free;
     printf(
-        "Journal entries loaded: %lu, free journal space: %lu bytes (%lu..%lu is used), free blocks: %lu / %lu\n",
+        "Journal entries loaded: %lu, free journal space: %lu bytes (%08lx..%08lx is used), free blocks: %lu / %lu\n",
         entries_loaded,
         (bs->journal.next_free >= bs->journal.used_start
             ? bs->journal.len-bs->journal.block_size - (bs->journal.next_free-bs->journal.used_start)
@@ -454,10 +453,15 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                     break;
                 }
             }
-            if (je->type == JE_SMALL_WRITE)
+            if (je->type == JE_SMALL_WRITE || je->type == JE_SMALL_WRITE_INSTANT)
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_small_write oid=%lu:%lu ver=%lu offset=%u len=%u\n", je->small_write.oid.inode, je->small_write.oid.stripe, je->small_write.version, je->small_write.offset, je->small_write.len);
+                printf(
+                    "je_small_write%s oid=%lx:%lx ver=%lu offset=%u len=%u\n",
+                    je->type == JE_SMALL_WRITE_INSTANT ? "_instant" : "",
+                    je->small_write.oid.inode, je->small_write.oid.stripe, je->small_write.version,
+                    je->small_write.offset, je->small_write.len
+                );
 #endif
                 // oid, version, offset, len
                 uint64_t prev_free = next_free;
@@ -475,7 +479,7 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                 if (location != je->small_write.data_offset)
                 {
                     char err[1024];
-                    snprintf(err, 1024, "BUG: calculated journal data offset (%lu) != stored journal data offset (%lu)", location, je->small_write.data_offset);
+                    snprintf(err, 1024, "BUG: calculated journal data offset (%08lx) != stored journal data offset (%08lx)", location, je->small_write.data_offset);
                     throw std::runtime_error(err);
                 }
                 uint32_t data_crc32 = 0;
@@ -528,7 +532,7 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                         .version = je->small_write.version,
                     };
                     bs->dirty_db.emplace(ov, (dirty_entry){
-                        .state = ST_J_SYNCED,
+                        .state = (BS_ST_SMALL_WRITE | BS_ST_SYNCED),
                         .flags = 0,
                         .location = location,
                         .offset = je->small_write.offset,
@@ -537,16 +541,27 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                     });
                     bs->journal.used_sectors[proc_pos]++;
 #ifdef BLOCKSTORE_DEBUG
-                    printf("journal offset %lu is used by %lu:%lu v%lu\n", proc_pos, ov.oid.inode, ov.oid.stripe, ov.version);
+                    printf(
+                        "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+                        proc_pos, ov.oid.inode, ov.oid.stripe, ov.version, bs->journal.used_sectors[proc_pos]
+                    );
 #endif
                     auto & unstab = bs->unstable_writes[ov.oid];
                     unstab = unstab < ov.version ? ov.version : unstab;
+                    if (je->type == JE_SMALL_WRITE_INSTANT)
+                    {
+                        bs->mark_stable(ov);
+                    }
                 }
             }
-            else if (je->type == JE_BIG_WRITE)
+            else if (je->type == JE_BIG_WRITE || je->type == JE_BIG_WRITE_INSTANT)
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_big_write oid=%lu:%lu ver=%lu loc=%lu\n", je->big_write.oid.inode, je->big_write.oid.stripe, je->big_write.version, je->big_write.location);
+                printf(
+                    "je_big_write%s oid=%lx:%lx ver=%lu loc=%lu\n",
+                    je->type == JE_BIG_WRITE_INSTANT ? "_instant" : "",
+                    je->big_write.oid.inode, je->big_write.oid.stripe, je->big_write.version, je->big_write.location
+                );
 #endif
                 auto clean_it = bs->clean_db.find(je->big_write.oid);
                 if (clean_it == bs->clean_db.end() ||
@@ -558,7 +573,7 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                         .version = je->big_write.version,
                     };
                     bs->dirty_db.emplace(ov, (dirty_entry){
-                        .state = ST_D_SYNCED,
+                        .state = (BS_ST_BIG_WRITE | BS_ST_SYNCED),
                         .flags = 0,
                         .location = je->big_write.location,
                         .offset = je->big_write.offset,
@@ -572,116 +587,63 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                     bs->journal.used_sectors[proc_pos]++;
                     auto & unstab = bs->unstable_writes[ov.oid];
                     unstab = unstab < ov.version ? ov.version : unstab;
+                    if (je->type == JE_BIG_WRITE_INSTANT)
+                    {
+                        bs->mark_stable(ov);
+                    }
                 }
             }
             else if (je->type == JE_STABLE)
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_stable oid=%lu:%lu ver=%lu\n", je->stable.oid.inode, je->stable.oid.stripe, je->stable.version);
+                printf("je_stable oid=%lx:%lx ver=%lu\n", je->stable.oid.inode, je->stable.oid.stripe, je->stable.version);
 #endif
                 // oid, version
                 obj_ver_id ov = {
                     .oid = je->stable.oid,
                     .version = je->stable.version,
                 };
-                auto it = bs->dirty_db.find(ov);
+                bs->mark_stable(ov);
-                if (it == bs->dirty_db.end())
-                {
-                    // journal contains a legitimate STABLE entry for a non-existing dirty write
-                    // this probably means that journal was trimmed between WRITE and STABLE entries
-                    // skip it
-                }
-                else
-                {
-                    while (1)
-                    {
-                        it->second.state = (it->second.state == ST_D_SYNCED
-                            ? ST_D_STABLE
-                            : (it->second.state == ST_DEL_SYNCED ? ST_DEL_STABLE : ST_J_STABLE));
-                        if (it == bs->dirty_db.begin())
-                            break;
-                        it--;
-                        if (it->first.oid != ov.oid || IS_STABLE(it->second.state))
-                            break;
-                    }
-                    bs->flusher->enqueue_flush(ov);
-                }
-                auto unstab_it = bs->unstable_writes.find(ov.oid);
-                if (unstab_it != bs->unstable_writes.end() && unstab_it->second <= ov.version)
-                {
-                    bs->unstable_writes.erase(unstab_it);
-                }
             }
             else if (je->type == JE_ROLLBACK)
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_rollback oid=%lu:%lu ver=%lu\n", je->rollback.oid.inode, je->rollback.oid.stripe, je->rollback.version);
+                printf("je_rollback oid=%lx:%lx ver=%lu\n", je->rollback.oid.inode, je->rollback.oid.stripe, je->rollback.version);
 #endif
                 // rollback dirty writes of <oid> up to <version>
-                auto it = bs->dirty_db.lower_bound((obj_ver_id){
+                obj_ver_id ov = {
                     .oid = je->rollback.oid,
-                    .version = UINT64_MAX,
+                    .version = je->rollback.version,
-                });
+                };
-                if (it != bs->dirty_db.begin())
+                bs->mark_rolled_back(ov);
-                {
-                    uint64_t max_unstable = 0;
-                    auto rm_start = it;
-                    auto rm_end = it;
-                    it--;
-                    while (it->first.oid == je->rollback.oid &&
-                        it->first.version > je->rollback.version &&
-                        !IS_IN_FLIGHT(it->second.state) &&
-                        !IS_STABLE(it->second.state))
-                    {
-                        if (it->first.oid != je->rollback.oid)
-                            break;
-                        else if (it->first.version <= je->rollback.version)
-                        {
-                            if (!IS_STABLE(it->second.state))
-                                max_unstable = it->first.version;
-                            break;
-                        }
-                        else if (IS_STABLE(it->second.state))
-                            break;
-                        // Remove entry
-                        rm_start = it;
-                        if (it == bs->dirty_db.begin())
-                            break;
-                        it--;
-                    }
-                    if (rm_start != rm_end)
-                    {
-                        bs->erase_dirty(rm_start, rm_end, UINT64_MAX);
-                    }
-                    auto unstab_it = bs->unstable_writes.find(je->rollback.oid);
-                    if (unstab_it != bs->unstable_writes.end())
-                    {
-                        if (max_unstable == 0)
-                            bs->unstable_writes.erase(unstab_it);
-                        else
-                            unstab_it->second = max_unstable;
-                    }
-                }
             }
             else if (je->type == JE_DELETE)
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_delete oid=%lu:%lu ver=%lu\n", je->del.oid.inode, je->del.oid.stripe, je->del.version);
+                printf("je_delete oid=%lx:%lx ver=%lu\n", je->del.oid.inode, je->del.oid.stripe, je->del.version);
 #endif
-                // oid, version
+                auto clean_it = bs->clean_db.find(je->del.oid);
-                obj_ver_id ov = {
+                if (clean_it == bs->clean_db.end() ||
-                    .oid = je->del.oid,
+                    clean_it->second.version < je->del.version)
-                    .version = je->del.version,
+                {
-                };
+                    // oid, version
-                bs->dirty_db.emplace(ov, (dirty_entry){
+                    obj_ver_id ov = {
-                    .state = ST_DEL_SYNCED,
+                        .oid = je->del.oid,
-                    .flags = 0,
+                        .version = je->del.version,
-                    .location = 0,
+                    };
-                    .offset = 0,
+                    bs->dirty_db.emplace(ov, (dirty_entry){
-                    .len = 0,
+                        .state = (BS_ST_DELETE | BS_ST_SYNCED),
-                    .journal_sector = proc_pos,
+                        .flags = 0,
-                });
+                        .location = 0,
-                bs->journal.used_sectors[proc_pos]++;
+                        .offset = 0,
+                        .len = 0,
+                        .journal_sector = proc_pos,
+                    });
+                    bs->journal.used_sectors[proc_pos]++;
+                    // Deletions are treated as immediately stable, because
+                    // "2-phase commit" (write->stabilize) isn't sufficient for them anyway
+                    bs->mark_stable(ov);
+                }
             }
             started = true;
             pos += je->size;

blockstore_init.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #pragma once
 
 class blockstore_init_meta

blockstore_journal.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 blockstore_journal_check_t::blockstore_journal_check_t(blockstore_impl_t *bs)
@@ -17,7 +20,9 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
     int required = entries_required;
     while (1)
     {
-        int fits = (bs->journal.block_size - next_in_pos) / size;
+        int fits = bs->journal.no_same_sector_overwrites && bs->journal.sector_info[next_sector].written
+            ? 0
+            : (bs->journal.block_size - next_in_pos) / size;
         if (fits > 0)
         {
             if (first_sector == -1)
@@ -101,7 +106,7 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
                 : bs->journal.used_start - bs->journal.next_free)
         );
         PRIV(op)->wait_for = WAIT_JOURNAL;
-        bs->flusher->force_start();
+        bs->flusher->request_trim();
         PRIV(op)->wait_detail = bs->journal.used_start;
         return 0;
     }
@@ -110,10 +115,12 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
 
 journal_entry* prefill_single_journal_entry(journal_t & journal, uint16_t type, uint32_t size)
 {
-    if (journal.block_size - journal.in_sector_pos < size)
+    if (journal.block_size - journal.in_sector_pos < size ||
+        journal.no_same_sector_overwrites && journal.sector_info[journal.cur_sector].written)
     {
         assert(!journal.sector_info[journal.cur_sector].dirty);
         // Move to the next journal sector
+        journal.sector_info[journal.cur_sector].written = false;
         if (journal.sector_info[journal.cur_sector].usage_count > 0)
         {
             // Also select next sector buffer in memory
@@ -148,6 +155,7 @@ journal_entry* prefill_single_journal_entry(journal_t & journal, uint16_t type,
 void prepare_journal_sector_write(journal_t & journal, int cur_sector, io_uring_sqe *sqe, std::function<void(ring_data_t*)> cb)
 {
     journal.sector_info[cur_sector].dirty = false;
+    journal.sector_info[cur_sector].written = true;
     journal.sector_info[cur_sector].usage_count++;
     ring_data_t *data = ((ring_data_t*)sqe->user_data);
     data->iov = (struct iovec){
@@ -180,8 +188,8 @@ bool journal_t::trim()
     auto journal_used_it = used_sectors.lower_bound(used_start);
 #ifdef BLOCKSTORE_DEBUG
     printf(
-        "Trimming journal (used_start=%08lx, next_free=%08lx, first_used=%08lx, usage_count=%08lx)\n",
+        "Trimming journal (used_start=%08lx, next_free=%08lx, dirty_start=%08lx, new_start=%08lx, new_refcount=%ld)\n",
-        used_start, next_free,
+        used_start, next_free, dirty_start,
         journal_used_it == used_sectors.end() ? 0 : journal_used_it->first,
         journal_used_it == used_sectors.end() ? 0 : journal_used_it->second
     );

blockstore_journal.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #pragma once
 
 #include "crc32c.h"
@@ -19,7 +22,9 @@
 #define JE_STABLE      0x04
 #define JE_DELETE      0x05
 #define JE_ROLLBACK    0x06
-#define JE_MAX         0x06
+#define JE_SMALL_WRITE_INSTANT 0x07
+#define JE_BIG_WRITE_INSTANT   0x08
+#define JE_MAX         0x08
 
 // crc32c comes first to ease calculation and is equal to crc32()
 struct __attribute__((__packed__)) journal_entry_start
@@ -127,6 +132,7 @@ struct journal_sector_info_t
 {
     uint64_t offset;
     uint64_t usage_count;
+    bool written;
     bool dirty;
 };
 
@@ -151,6 +157,7 @@ struct journal_t
     void *sector_buf = NULL;
     journal_sector_info_t *sector_info = NULL;
     uint64_t sector_count;
+    bool no_same_sector_overwrites = false;
     int cur_sector = 0;
     int in_sector_pos = 0;
 

blockstore_open.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <sys/file.h>
 #include "blockstore_impl.h"
 
@@ -59,6 +62,8 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
     journal_device = config["journal_device"];
     journal.offset = strtoull(config["journal_offset"].c_str(), NULL, 10);
     journal.sector_count = strtoull(config["journal_sector_buffer_count"].c_str(), NULL, 10);
+    journal.no_same_sector_overwrites = config["journal_no_same_sector_overwrites"] == "true" ||
+        config["journal_no_same_sector_overwrites"] == "1" || config["journal_no_same_sector_overwrites"] == "yes";
     journal.inmemory = config["inmemory_journal"] != "false";
     disk_alignment = strtoull(config["disk_alignment"].c_str(), NULL, 10);
     journal_block_size = strtoull(config["journal_block_size"].c_str(), NULL, 10);

blockstore_read.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 int blockstore_impl_t::fulfill_read_push(blockstore_op_t *op, void *buf, uint64_t offset, uint64_t len,
@@ -37,6 +40,7 @@ int blockstore_impl_t::fulfill_read_push(blockstore_op_t *op, void *buf, uint64_
     return 1;
 }
 
+// FIXME I've seen a bug here so I want some tests
 int blockstore_impl_t::fulfill_read(blockstore_op_t *read_op, uint64_t &fulfilled, uint32_t item_start, uint32_t item_end,
     uint32_t item_state, uint64_t item_version, uint64_t item_location)
 {
@@ -49,8 +53,20 @@ int blockstore_impl_t::fulfill_read(blockstore_op_t *read_op, uint64_t &fulfille
         while (1)
         {
             for (; it != PRIV(read_op)->read_vec.end(); it++)
+            {
                 if (it->offset >= cur_start)
+                {
                     break;
+                }
+                else if (it->offset + it->len > cur_start)
+                {
+                    cur_start = it->offset + it->len;
+                    if (cur_start >= item_end)
+                    {
+                        goto endwhile;
+                    }
+                }
+            }
             if (it == PRIV(read_op)->read_vec.end() || it->offset > cur_start)
             {
                 fulfill_read_t el = {
@@ -69,9 +85,12 @@ int blockstore_impl_t::fulfill_read(blockstore_op_t *read_op, uint64_t &fulfille
             }
             cur_start = it->offset + it->len;
             if (it == PRIV(read_op)->read_vec.end() || cur_start >= item_end)
+            {
                 break;
+            }
         }
     }
+endwhile:
     return 1;
 }
 
@@ -141,7 +160,7 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
         {
             if (!clean_entry_bitmap_size)
             {
-                if (!fulfill_read(read_op, fulfilled, 0, block_size, ST_CURRENT, 0, clean_it->second.location))
+                if (!fulfill_read(read_op, fulfilled, 0, block_size, (BS_ST_BIG_WRITE | BS_ST_STABLE), 0, clean_it->second.location))
                 {
                     // need to wait. undo added requests, don't dequeue op
                     PRIV(read_op)->read_vec.clear();
@@ -173,7 +192,7 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
                     {
                         // fill with zeroes
                         fulfill_read(read_op, fulfilled, bmp_start * bitmap_granularity,
-                            bmp_end * bitmap_granularity, ST_DEL_STABLE, 0, 0);
+                            bmp_end * bitmap_granularity, (BS_ST_DELETE | BS_ST_STABLE), 0, 0);
                     }
                     bmp_start = bmp_end;
                     while (clean_entry_bitmap[bmp_end >> 3] & (1 << (bmp_end & 0x7)) && bmp_end < bmp_size)
@@ -183,7 +202,8 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
                     if (bmp_end > bmp_start)
                     {
                         if (!fulfill_read(read_op, fulfilled, bmp_start * bitmap_granularity,
-                            bmp_end * bitmap_granularity, ST_CURRENT, 0, clean_it->second.location + bmp_start * bitmap_granularity))
+                            bmp_end * bitmap_granularity, (BS_ST_BIG_WRITE | BS_ST_STABLE), 0,
+                            clean_it->second.location + bmp_start * bitmap_granularity))
                         {
                             // need to wait. undo added requests, don't dequeue op
                             PRIV(read_op)->read_vec.clear();
@@ -198,7 +218,7 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
     else if (fulfilled < read_op->len)
     {
         // fill remaining parts with zeroes
-        fulfill_read(read_op, fulfilled, 0, block_size, ST_DEL_STABLE, 0, 0);
+        fulfill_read(read_op, fulfilled, 0, block_size, (BS_ST_DELETE | BS_ST_STABLE), 0, 0);
     }
     assert(fulfilled == read_op->len);
     read_op->version = result_version;

blockstore_rollback.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
@@ -77,33 +80,6 @@ int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
     }
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
     {
-        // FIXME This is here only for the purpose of tracking unstable_writes. Remove if not required
-        // FIXME ...aaaand this is similar to blockstore_init.cpp - maybe dedup it?
-        auto dirty_it = dirty_db.lower_bound((obj_ver_id){
-            .oid = v->oid,
-            .version = UINT64_MAX,
-        });
-        uint64_t max_unstable = 0;
-        while (dirty_it != dirty_db.begin())
-        {
-            dirty_it--;
-            if (dirty_it->first.oid != v->oid)
-                break;
-            else if (dirty_it->first.version <= v->version)
-            {
-                if (!IS_STABLE(dirty_it->second.state))
-                    max_unstable = dirty_it->first.version;
-                break;
-            }
-        }
-        auto unstab_it = unstable_writes.find(v->oid);
-        if (unstab_it != unstable_writes.end())
-        {
-            if (max_unstable == 0)
-                unstable_writes.erase(unstab_it);
-            else
-                unstab_it->second = max_unstable;
-        }
         journal_entry_rollback *je = (journal_entry_rollback*)
             prefill_single_journal_entry(journal, JE_ROLLBACK, sizeof(journal_entry_rollback));
         journal.sector_info[journal.cur_sector].dirty = false;
@@ -161,26 +137,7 @@ resume_5:
     int i;
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
     {
-        // Erase dirty_db entries
+        mark_rolled_back(*v);
-        auto rm_end = dirty_db.lower_bound((obj_ver_id){
-            .oid = v->oid,
-            .version = UINT64_MAX,
-        });
-        auto rm_start = rm_end;
-        assert(rm_start != dirty_db.begin());
-        rm_start--;
-        while (1)
-        {
-            if (rm_start->first.oid != v->oid || rm_start->first.version <= v->version)
-            {
-                rm_start++;
-                break;
-            }
-            if (rm_start == dirty_db.begin())
-                break;
-            rm_start--;
-        }
-        erase_dirty(rm_start, rm_end, UINT64_MAX);
     }
     journal.trim();
     inflight_writes--;
@@ -190,6 +147,54 @@ resume_5:
     return 1;
 }
 
+void blockstore_impl_t::mark_rolled_back(const obj_ver_id & ov)
+{
+    auto it = dirty_db.lower_bound((obj_ver_id){
+        .oid = ov.oid,
+        .version = UINT64_MAX,
+    });
+    if (it != dirty_db.begin())
+    {
+        uint64_t max_unstable = 0;
+        auto rm_start = it;
+        auto rm_end = it;
+        it--;
+        while (it->first.oid == ov.oid &&
+            it->first.version > ov.version &&
+            !IS_IN_FLIGHT(it->second.state) &&
+            !IS_STABLE(it->second.state))
+        {
+            if (it->first.oid != ov.oid)
+                break;
+            else if (it->first.version <= ov.version)
+            {
+                if (!IS_STABLE(it->second.state))
+                    max_unstable = it->first.version;
+                break;
+            }
+            else if (IS_STABLE(it->second.state))
+                break;
+            // Remove entry
+            rm_start = it;
+            if (it == dirty_db.begin())
+                break;
+            it--;
+        }
+        if (rm_start != rm_end)
+        {
+            erase_dirty(rm_start, rm_end, UINT64_MAX);
+        }
+        auto unstab_it = unstable_writes.find(ov.oid);
+        if (unstab_it != unstable_writes.end())
+        {
+            if (max_unstable == 0)
+                unstable_writes.erase(unstab_it);
+            else
+                unstab_it->second = max_unstable;
+        }
+    }
+}
+
 void blockstore_impl_t::handle_rollback_event(ring_data_t *data, blockstore_op_t *op)
 {
     live = true;
@@ -225,11 +230,13 @@ void blockstore_impl_t::erase_dirty(blockstore_dirty_db_t::iterator dirty_start,
 #endif
             data_alloc->set(dirty_it->second.location >> block_order, false);
         }
-#ifdef BLOCKSTORE_DEBUG
-        printf("remove usage of journal offset %lu by %lu:%lu v%lu\n", dirty_it->second.journal_sector,
-            dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version);
-#endif
         int used = --journal.used_sectors[dirty_it->second.journal_sector];
+#ifdef BLOCKSTORE_DEBUG
+        printf(
+            "remove usage of journal offset %08lx by %lx:%lx v%lu (%d refs)\n", dirty_it->second.journal_sector,
+            dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version, used
+        );
+#endif
         if (used == 0)
         {
             journal.used_sectors.erase(dirty_it->second.journal_sector);

blockstore_stable.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 // Stabilize small write:
@@ -64,7 +67,7 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
                 // Already stable
             }
         }
-        else if (IS_UNSYNCED(dirty_it->second.state))
+        else if (!IS_SYNCED(dirty_it->second.state))
         {
             // Object not synced yet. Caller must sync it first
             op->retval = -EBUSY;
@@ -109,12 +112,6 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
     {
         // FIXME: Only stabilize versions that aren't stable yet
-        auto unstab_it = unstable_writes.find(v->oid);
-        if (unstab_it != unstable_writes.end() &&
-            unstab_it->second <= v->version)
-        {
-            unstable_writes.erase(unstab_it);
-        }
         journal_entry_stable *je = (journal_entry_stable*)
             prefill_single_journal_entry(journal, JE_STABLE, sizeof(journal_entry_stable));
         journal.sector_info[journal.cur_sector].dirty = false;
@@ -174,42 +171,7 @@ resume_5:
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
     {
         // Mark all dirty_db entries up to op->version as stable
-        auto dirty_it = dirty_db.find(*v);
+        mark_stable(*v);
-        if (dirty_it != dirty_db.end())
-        {
-            while (1)
-            {
-                if (dirty_it->second.state == ST_J_SYNCED)
-                {
-                    dirty_it->second.state = ST_J_STABLE;
-                }
-                else if (dirty_it->second.state == ST_D_SYNCED)
-                {
-                    dirty_it->second.state = ST_D_STABLE;
-                }
-                else if (dirty_it->second.state == ST_DEL_SYNCED)
-                {
-                    dirty_it->second.state = ST_DEL_STABLE;
-                }
-                else if (IS_STABLE(dirty_it->second.state))
-                {
-                    break;
-                }
-                if (dirty_it == dirty_db.begin())
-                {
-                    break;
-                }
-                dirty_it--;
-                if (dirty_it->first.oid != v->oid)
-                {
-                    break;
-                }
-            }
-#ifdef BLOCKSTORE_DEBUG
-            printf("enqueue_flush %lu:%lu v%lu\n", v->oid.inode, v->oid.stripe, v->version);
-#endif
-            flusher->enqueue_flush(*v);
-        }
     }
     inflight_writes--;
     // Acknowledge op
@@ -218,6 +180,44 @@ resume_5:
     return 1;
 }
 
+void blockstore_impl_t::mark_stable(const obj_ver_id & v)
+{
+    auto dirty_it = dirty_db.find(v);
+    if (dirty_it != dirty_db.end())
+    {
+        while (1)
+        {
+            if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_SYNCED)
+            {
+                dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_STABLE;
+            }
+            else if (IS_STABLE(dirty_it->second.state))
+            {
+                break;
+            }
+            if (dirty_it == dirty_db.begin())
+            {
+                break;
+            }
+            dirty_it--;
+            if (dirty_it->first.oid != v.oid)
+            {
+                break;
+            }
+        }
+#ifdef BLOCKSTORE_DEBUG
+        printf("enqueue_flush %lx:%lx v%lu\n", v.oid.inode, v.oid.stripe, v.version);
+#endif
+        flusher->enqueue_flush(v);
+    }
+    auto unstab_it = unstable_writes.find(v.oid);
+    if (unstab_it != unstable_writes.end() &&
+        unstab_it->second <= v.version)
+    {
+        unstable_writes.erase(unstab_it);
+    }
+}
+
 void blockstore_impl_t::handle_stable_event(ring_data_t *data, blockstore_op_t *op)
 {
     live = true;

blockstore_sync.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 #define SYNC_HAS_SMALL 1
@@ -127,13 +130,19 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
         }
         while (it != PRIV(op)->sync_big_writes.end())
         {
-            journal_entry_big_write *je = (journal_entry_big_write*)
+            journal_entry_big_write *je = (journal_entry_big_write*)prefill_single_journal_entry(
-                prefill_single_journal_entry(journal, JE_BIG_WRITE, sizeof(journal_entry_big_write));
+                journal, (dirty_db[*it].state & BS_ST_INSTANT) ? JE_BIG_WRITE_INSTANT : JE_BIG_WRITE,
+                sizeof(journal_entry_big_write)
+            );
             dirty_db[*it].journal_sector = journal.sector_info[journal.cur_sector].offset;
             journal.sector_info[journal.cur_sector].dirty = false;
             journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
 #ifdef BLOCKSTORE_DEBUG
-            printf("journal offset %lu is used by %lu:%lu v%lu\n", dirty_db[*it].journal_sector, it->oid.inode, it->oid.stripe, it->version);
+            printf(
+                "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+                dirty_db[*it].journal_sector, it->oid.inode, it->oid.stripe, it->version,
+                journal.used_sectors[journal.sector_info[journal.cur_sector].offset]
+            );
 #endif
             je->oid = it->oid;
             je->version = it->version;
@@ -248,18 +257,22 @@ void blockstore_impl_t::ack_one_sync(blockstore_op_t *op)
     for (auto it = PRIV(op)->sync_big_writes.begin(); it != PRIV(op)->sync_big_writes.end(); it++)
     {
 #ifdef BLOCKSTORE_DEBUG
-        printf("Ack sync big %lu:%lu v%lu\n", it->oid.inode, it->oid.stripe, it->version);
+        printf("Ack sync big %lx:%lx v%lu\n", it->oid.inode, it->oid.stripe, it->version);
 #endif
         auto & unstab = unstable_writes[it->oid];
         unstab = unstab < it->version ? it->version : unstab;
         auto dirty_it = dirty_db.find(*it);
-        dirty_it->second.state = ST_D_SYNCED;
+        dirty_it->second.state = ((dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SYNCED);
+        if (dirty_it->second.state & BS_ST_INSTANT)
+        {
+            mark_stable(dirty_it->first);
+        }
         dirty_it++;
         while (dirty_it != dirty_db.end() && dirty_it->first.oid == it->oid)
         {
-            if (dirty_it->second.state == ST_J_WAIT_BIG)
+            if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG)
             {
-                dirty_it->second.state = ST_J_IN_FLIGHT;
+                dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_IN_FLIGHT;
             }
             dirty_it++;
         }
@@ -267,11 +280,24 @@ void blockstore_impl_t::ack_one_sync(blockstore_op_t *op)
     for (auto it = PRIV(op)->sync_small_writes.begin(); it != PRIV(op)->sync_small_writes.end(); it++)
     {
 #ifdef BLOCKSTORE_DEBUG
-        printf("Ack sync small %lu:%lu v%lu\n", it->oid.inode, it->oid.stripe, it->version);
+        printf("Ack sync small %lx:%lx v%lu\n", it->oid.inode, it->oid.stripe, it->version);
 #endif
         auto & unstab = unstable_writes[it->oid];
         unstab = unstab < it->version ? it->version : unstab;
-        dirty_db[*it].state = dirty_db[*it].state == ST_DEL_WRITTEN ? ST_DEL_SYNCED : ST_J_SYNCED;
+        if (dirty_db[*it].state == (BS_ST_DELETE | BS_ST_WRITTEN))
+        {
+            dirty_db[*it].state = (BS_ST_DELETE | BS_ST_SYNCED);
+            // Deletions are treated as immediately stable
+            mark_stable(*it);
+        }
+        else /* (BS_ST_INSTANT?) | BS_ST_SMALL_WRITE | BS_ST_WRITTEN */
+        {
+            dirty_db[*it].state = (dirty_db[*it].state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SYNCED;
+            if (dirty_db[*it].state & BS_ST_INSTANT)
+            {
+                mark_stable(*it);
+            }
+        }
     }
     in_progress_syncs.erase(PRIV(op)->in_progress_ptr);
     op->retval = 0;

blockstore_write.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
@@ -18,9 +21,9 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
             found = true;
             version = dirty_it->first.version + 1;
             deleted = IS_DELETE(dirty_it->second.state);
-            is_inflight_big = dirty_it->second.state >= ST_D_IN_FLIGHT &&
+            is_inflight_big = (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE
-                dirty_it->second.state < ST_D_SYNCED ||
+                ? !IS_SYNCED(dirty_it->second.state)
-                dirty_it->second.state == ST_J_WAIT_BIG;
+                : ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG);
         }
     }
     if (!found)
@@ -65,9 +68,9 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
     }
 #ifdef BLOCKSTORE_DEBUG
     if (is_del)
-        printf("Delete %lu:%lu v%lu\n", op->oid.inode, op->oid.stripe, op->version);
+        printf("Delete %lx:%lx v%lu\n", op->oid.inode, op->oid.stripe, op->version);
     else
-        printf("Write %lu:%lu v%lu offset=%u len=%u\n", op->oid.inode, op->oid.stripe, op->version, op->offset, op->len);
+        printf("Write %lx:%lx v%lu offset=%u len=%u\n", op->oid.inode, op->oid.stripe, op->version, op->offset, op->len);
 #endif
     // No strict need to add it into dirty_db here, it's just left
     // from the previous implementation where reads waited for writes
@@ -77,8 +80,10 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
     }, (dirty_entry){
         .state = (uint32_t)(
             is_del
-                ? ST_DEL_IN_FLIGHT
+                ? (BS_ST_DELETE | BS_ST_IN_FLIGHT)
-                : (op->len == block_size || deleted ? ST_D_IN_FLIGHT : (is_inflight_big ? ST_J_WAIT_BIG : ST_J_IN_FLIGHT))
+                : (op->opcode == BS_OP_WRITE_STABLE ? BS_ST_INSTANT : 0) | (op->len == block_size || deleted
+                    ? (BS_ST_BIG_WRITE | BS_ST_IN_FLIGHT)
+                    : (is_inflight_big ? (BS_ST_SMALL_WRITE | BS_ST_WAIT_BIG) : (BS_ST_SMALL_WRITE | BS_ST_IN_FLIGHT)))
         ),
         .flags = 0,
         .location = 0,
@@ -100,11 +105,13 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         .oid = op->oid,
         .version = op->version,
     });
-    if (dirty_it->second.state == ST_J_WAIT_BIG)
+    assert(dirty_it != dirty_db.end());
+    if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG)
     {
+        // Don't dequeue
         return 0;
     }
-    else if (dirty_it->second.state == ST_D_IN_FLIGHT)
+    else if ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE)
     {
         blockstore_journal_check_t space_check(this);
         if (!space_check.check_available(op, unsynced_big_writes.size() + 1, sizeof(journal_entry_big_write), JOURNAL_STABILIZE_RESERVATION))
@@ -128,7 +135,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         }
         BS_SUBMIT_GET_SQE(sqe, data);
         dirty_it->second.location = loc << block_order;
-        dirty_it->second.state = ST_D_SUBMITTED;
+        dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SUBMITTED;
 #ifdef BLOCKSTORE_DEBUG
         printf("Allocate block %lu\n", loc);
 #endif
@@ -168,7 +175,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
             PRIV(op)->op_state = 1;
         }
     }
-    else
+    else /* if ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_SMALL_WRITE) */
     {
         // Small (journaled) write
         // First check if the journal has sufficient space
@@ -208,12 +215,18 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
             }
         }
         // Then pre-fill journal entry
-        journal_entry_small_write *je = (journal_entry_small_write*)
+        journal_entry_small_write *je = (journal_entry_small_write*)prefill_single_journal_entry(
-            prefill_single_journal_entry(journal, JE_SMALL_WRITE, sizeof(journal_entry_small_write));
+            journal, op->opcode == BS_OP_WRITE_STABLE ? JE_SMALL_WRITE_INSTANT : JE_SMALL_WRITE,
+            sizeof(journal_entry_small_write)
+        );
         dirty_it->second.journal_sector = journal.sector_info[journal.cur_sector].offset;
         journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
 #ifdef BLOCKSTORE_DEBUG
-        printf("journal offset %lu is used by %lu:%lu v%lu\n", dirty_it->second.journal_sector, dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version);
+        printf(
+            "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+            dirty_it->second.journal_sector, dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version,
+            journal.used_sectors[journal.sector_info[journal.cur_sector].offset]
+        );
 #endif
         // Figure out where data will be
         journal.next_free = (journal.next_free + op->len) <= journal.len ? journal.next_free : journal_block_size;
@@ -252,7 +265,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
             // Zero-length overwrite. Allowed to bump object version in EC placement groups without actually writing data
         }
         dirty_it->second.location = journal.next_free;
-        dirty_it->second.state = ST_J_SUBMITTED;
+        dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SUBMITTED;
         journal.next_free += op->len;
         if (journal.next_free >= journal.len)
         {
@@ -288,6 +301,7 @@ int blockstore_impl_t::continue_write(blockstore_op_t *op)
         .oid = op->oid,
         .version = op->version,
     });
+    assert(dirty_it != dirty_db.end());
     if (PRIV(op)->op_state == 2)
         goto resume_2;
     else if (PRIV(op)->op_state == 4)
@@ -301,12 +315,19 @@ resume_2:
     {
         return 0;
     }
-    je = (journal_entry_big_write*)prefill_single_journal_entry(journal, JE_BIG_WRITE, sizeof(journal_entry_big_write));
+    je = (journal_entry_big_write*)prefill_single_journal_entry(
+        journal, op->opcode == BS_OP_WRITE_STABLE ? JE_BIG_WRITE_INSTANT : JE_BIG_WRITE,
+        sizeof(journal_entry_big_write)
+    );
     dirty_it->second.journal_sector = journal.sector_info[journal.cur_sector].offset;
     journal.sector_info[journal.cur_sector].dirty = false;
     journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
 #ifdef BLOCKSTORE_DEBUG
-    printf("journal offset %lu is used by %lu:%lu v%lu\n", journal.sector_info[journal.cur_sector].offset, op->oid.inode, op->oid.stripe, op->version);
+    printf(
+        "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+        journal.sector_info[journal.cur_sector].offset, op->oid.inode, op->oid.stripe, op->version,
+        journal.used_sectors[journal.sector_info[journal.cur_sector].offset]
+    );
 #endif
     je->oid = op->oid;
     je->version = op->version;
@@ -324,9 +345,9 @@ resume_2:
 resume_4:
     // Switch object state
 #ifdef BLOCKSTORE_DEBUG
-    printf("Ack write %lu:%lu v%lu = %d\n", op->oid.inode, op->oid.stripe, op->version, dirty_it->second.state);
+    printf("Ack write %lx:%lx v%lu = %d\n", op->oid.inode, op->oid.stripe, op->version, dirty_it->second.state);
 #endif
-    bool imm = dirty_it->second.state == ST_D_SUBMITTED
+    bool imm = (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE
         ? (immediate_commit == IMMEDIATE_ALL)
         : (immediate_commit != IMMEDIATE_NONE);
     if (imm)
@@ -334,26 +355,21 @@ resume_4:
         auto & unstab = unstable_writes[op->oid];
         unstab = unstab < op->version ? op->version : unstab;
     }
-    if (dirty_it->second.state == ST_J_SUBMITTED)
+    dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK)
+        | (imm ? BS_ST_SYNCED : BS_ST_WRITTEN);
+    if (imm && ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_DELETE || (dirty_it->second.state & BS_ST_INSTANT)))
     {
-        dirty_it->second.state = imm ? ST_J_SYNCED : ST_J_WRITTEN;
+        // Deletions are treated as immediately stable
-    }
+        mark_stable(dirty_it->first);
-    else if (dirty_it->second.state == ST_D_SUBMITTED)
-    {
-        dirty_it->second.state = imm ? ST_D_SYNCED : ST_D_WRITTEN;
-    }
-    else if (dirty_it->second.state == ST_DEL_SUBMITTED)
-    {
-        dirty_it->second.state = imm ? ST_DEL_SYNCED : ST_DEL_WRITTEN;
     }
     if (immediate_commit == IMMEDIATE_ALL)
     {
         dirty_it++;
         while (dirty_it != dirty_db.end() && dirty_it->first.oid == op->oid)
         {
-            if (dirty_it->second.state == ST_J_WAIT_BIG)
+            if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG)
             {
-                dirty_it->second.state = ST_J_IN_FLIGHT;
+                dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_IN_FLIGHT;
             }
             dirty_it++;
         }
@@ -403,6 +419,10 @@ void blockstore_impl_t::release_journal_sectors(blockstore_op_t *op)
             {
                 // We know for sure that we won't write into this sector anymore
                 uint64_t new_ds = journal.sector_info[s-1].offset + journal.block_size;
+                if (new_ds >= journal.len)
+                {
+                    new_ds = journal.block_size;
+                }
                 if ((journal.dirty_start + (journal.dirty_start >= journal.used_start ? 0 : journal.len)) <
                     (new_ds + (new_ds >= journal.used_start ? 0 : journal.len)))
                 {
@@ -423,6 +443,7 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
         .oid = op->oid,
         .version = op->version,
     });
+    assert(dirty_it != dirty_db.end());
     blockstore_journal_check_t space_check(this);
     if (!space_check.check_available(op, 1, sizeof(journal_entry_del), 0))
     {
@@ -452,18 +473,23 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
         }
     }
     // Pre-fill journal entry
-    journal_entry_del *je = (journal_entry_del*)
+    journal_entry_del *je = (journal_entry_del*)prefill_single_journal_entry(
-        prefill_single_journal_entry(journal, JE_DELETE, sizeof(struct journal_entry_del));
+        journal, JE_DELETE, sizeof(struct journal_entry_del)
+    );
     dirty_it->second.journal_sector = journal.sector_info[journal.cur_sector].offset;
     journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
 #ifdef BLOCKSTORE_DEBUG
-    printf("journal offset %lu is used by %lu:%lu v%lu\n", dirty_it->second.journal_sector, dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version);
+    printf(
+        "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+        dirty_it->second.journal_sector, dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version,
+        journal.used_sectors[journal.sector_info[journal.cur_sector].offset]
+    );
 #endif
     je->oid = op->oid;
     je->version = op->version;
     je->crc32 = je_crc32((journal_entry*)je);
     journal.crc32_last = je->crc32;
-    dirty_it->second.state = ST_DEL_SUBMITTED;
+    dirty_it->second.state = BS_ST_DELETE | BS_ST_SUBMITTED;
     if (immediate_commit != IMMEDIATE_NONE)
     {
         prepare_journal_sector_write(journal, journal.cur_sector, sqe, cb);

cluster_client.h

@@ -1,209 +1,102 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #pragma once
 
-#include <sys/types.h>
+#include "messenger.h"
-#include <stdint.h>
+#include "etcd_state_client.h"
-#include <arpa/inet.h>
-#include <malloc.h>
 
-#include <set>
+#define MIN_BLOCK_SIZE 4*1024
-#include <map>
+#define MAX_BLOCK_SIZE 128*1024*1024
-#include <deque>
+#define DEFAULT_BLOCK_SIZE 128*1024
-#include <vector>
+#define DEFAULT_DISK_ALIGNMENT 4096
+#define DEFAULT_BITMAP_GRANULARITY 4096
+#define DEFAULT_CLIENT_DIRTY_LIMIT 32*1024*1024
 
-#include "json11/json11.hpp"
+struct cluster_op_t;
-#include "osd_ops.h"
-#include "timerfd_manager.h"
-#include "ringloop.h"
 
-#define OSD_OP_IN 0
+struct cluster_op_part_t
-#define OSD_OP_OUT 1
-
-#define CL_READ_HDR 1
-#define CL_READ_DATA 2
-#define CL_READ_REPLY_DATA 3
-#define CL_WRITE_READY 1
-#define CL_WRITE_REPLY 2
-#define MAX_EPOLL_EVENTS 64
-#define OSD_OP_INLINE_BUF_COUNT 16
-
-#define PEER_CONNECTING 1
-#define PEER_CONNECTED 2
-
-struct osd_op_buf_list_t
 {
-    int count = 0, alloc = 0, sent = 0;
+    cluster_op_t *parent;
-    iovec *buf = NULL;
+    uint64_t offset;
-    iovec inline_buf[OSD_OP_INLINE_BUF_COUNT];
+    uint32_t len;
-
+    pg_num_t pg_num;
-    ~osd_op_buf_list_t()
+    osd_num_t osd_num;
-    {
+    osd_op_buf_list_t iov;
-        if (buf && buf != inline_buf)
+    bool sent;
-        {
+    bool done;
-            free(buf);
+    osd_op_t op;
-        }
-    }
-
-    inline iovec* get_iovec()
-    {
-        return (buf ? buf : inline_buf) + sent;
-    }
-
-    inline int get_size()
-    {
-        return count - sent;
-    }
-
-    inline void push_back(void *nbuf, size_t len)
-    {
-        if (count >= alloc)
-        {
-            if (!alloc)
-            {
-                alloc = OSD_OP_INLINE_BUF_COUNT;
-                buf = inline_buf;
-            }
-            else if (buf == inline_buf)
-            {
-                int old = alloc;
-                alloc = ((alloc/16)*16 + 1);
-                buf = (iovec*)malloc(sizeof(iovec) * alloc);
-                memcpy(buf, inline_buf, sizeof(iovec)*old);
-            }
-            else
-            {
-                alloc = ((alloc/16)*16 + 1);
-                buf = (iovec*)realloc(buf, sizeof(iovec) * alloc);
-            }
-        }
-        buf[count++] = { .iov_base = nbuf, .iov_len = len };
-    }
 };
 
-struct blockstore_op_t;
+struct cluster_op_t
-
-struct osd_primary_op_data_t;
-
-struct osd_op_t
 {
-    timespec tv_begin;
+    uint64_t opcode; // OSD_OP_READ, OSD_OP_WRITE, OSD_OP_SYNC
-    uint64_t op_type = OSD_OP_IN;
+    uint64_t inode;
-    int peer_fd;
+    uint64_t offset;
-    osd_any_op_t req;
+    uint64_t len;
-    osd_any_reply_t reply;
+    int retval;
-    blockstore_op_t *bs_op = NULL;
+    osd_op_buf_list_t iov;
+    std::function<void(cluster_op_t*)> callback;
+protected:
     void *buf = NULL;
-    void *rmw_buf = NULL;
+    cluster_op_t *orig_op = NULL;
-    osd_primary_op_data_t* op_data = NULL;
+    bool is_internal = false;
-    std::function<void(osd_op_t*)> callback;
+    bool needs_reslice = false;
-
+    bool up_wait = false;
-    osd_op_buf_list_t send_list;
+    int sent_count = 0, done_count = 0;
-
+    std::vector<cluster_op_part_t> parts;
-    ~osd_op_t();
+    friend class cluster_client_t;
 };
 
-struct osd_client_t
+class cluster_client_t
-{
-    sockaddr_in peer_addr;
-    int peer_port;
-    int peer_fd;
-    int peer_state;
-    int connect_timeout_id = -1;
-    osd_num_t osd_num = 0;
-
-    void *in_buf = NULL;
-
-    // Read state
-    int read_ready = 0;
-    osd_op_t *read_op = NULL;
-    int read_reply_id = 0;
-    iovec read_iov;
-    msghdr read_msg;
-    void *read_buf = NULL;
-    int read_remaining = 0;
-    int read_state = 0;
-
-    // Outbound operations sent to this peer
-    std::map<int, osd_op_t*> sent_ops;
-
-    // Outbound messages (replies or requests)
-    std::deque<osd_op_t*> outbox;
-
-    // PGs dirtied by this client's primary-writes (FIXME to drop the connection)
-    std::set<pg_num_t> dirty_pgs;
-
-    // Write state
-    osd_op_t *write_op = NULL;
-    msghdr write_msg;
-    int write_state = 0;
-};
-
-struct osd_wanted_peer_t
-{
-    json11::Json address_list;
-    int port;
-    time_t last_connect_attempt;
-    bool connecting, address_changed;
-    int address_index;
-    std::string cur_addr;
-    int cur_port;
-};
-
-struct osd_op_stats_t
-{
-    uint64_t op_stat_sum[OSD_OP_MAX+1] = { 0 };
-    uint64_t op_stat_count[OSD_OP_MAX+1] = { 0 };
-    uint64_t op_stat_bytes[OSD_OP_MAX+1] = { 0 };
-    uint64_t subop_stat_sum[OSD_OP_MAX+1] = { 0 };
-    uint64_t subop_stat_count[OSD_OP_MAX+1] = { 0 };
-};
-
-struct cluster_client_t
 {
     timerfd_manager_t *tfd;
     ring_loop_t *ringloop;
 
-    // osd_num_t is only for logging and asserts
+    uint64_t bs_block_size = 0;
-    osd_num_t osd_num;
+    uint64_t bs_disk_alignment = 0;
-    int receive_buffer_size = 9000;
+    uint64_t bs_bitmap_granularity = 0;
-    int peer_connect_interval = 5;
+    std::map<pool_id_t, uint64_t> pg_counts;
-    int peer_connect_timeout = 5;
+    bool immediate_commit = false;
-    int log_level = 0;
+    // FIXME: Implement inmemory_commit mode. Note that it requires to return overlapping reads from memory.
+    uint64_t client_dirty_limit = 0;
+    int log_level;
+    int up_wait_retry_interval = 500; // ms
 
-    std::map<osd_num_t, osd_wanted_peer_t> wanted_peers;
+    uint64_t op_id = 1;
-    std::map<uint64_t, int> osd_peer_fds;
+    etcd_state_client_t st_cli;
-    uint64_t next_subop_id = 1;
+    osd_messenger_t msgr;
+    ring_consumer_t consumer;
+    // operations currently in progress
+    std::set<cluster_op_t*> cur_ops;
+    int retry_timeout_id = 0;
+    // unsynced operations are copied in memory to allow replay when cluster isn't in the immediate_commit mode
+    // unsynced_writes are replayed in any order (because only the SYNC operation guarantees ordering)
+    std::vector<cluster_op_t*> unsynced_writes;
+    std::vector<cluster_op_t*> syncing_writes;
+    cluster_op_t* cur_sync = NULL;
+    std::vector<cluster_op_t*> next_writes;
+    std::vector<cluster_op_t*> offline_ops;
+    uint64_t queued_bytes = 0;
 
-    std::map<int, osd_client_t> clients;
+public:
-    std::vector<int> read_ready_clients;
+    cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json & config);
-    std::vector<int> write_ready_clients;
+    ~cluster_client_t();
+    void execute(cluster_op_t *op);
+    void stop();
 
-    // op statistics
+protected:
-    osd_op_stats_t stats;
+    void continue_ops(bool up_retry = false);
-
+    void on_load_config_hook(json11::Json::object & config);
-    // public
+    void on_load_pgs_hook(bool success);
-    void connect_peer(uint64_t osd_num, json11::Json address_list, int port);
+    void on_change_hook(json11::Json::object & changes);
-    void stop_client(int peer_fd);
+    void on_change_osd_state_hook(uint64_t peer_osd);
-    void outbox_push(osd_op_t *cur_op);
+    void continue_rw(cluster_op_t *op);
-    std::function<void(osd_op_t*)> exec_op;
+    void slice_rw(cluster_op_t *op);
-    std::function<void(osd_num_t)> repeer_pgs;
+    bool try_send(cluster_op_t *op, cluster_op_part_t *part);
-
+    void execute_sync(cluster_op_t *op);
-    // private
+    void continue_sync();
-    void try_connect_peer(uint64_t osd_num);
+    void finish_sync();
-    void try_connect_peer_addr(osd_num_t peer_osd, const char *peer_host, int peer_port);
+    void send_sync(cluster_op_t *op, cluster_op_part_t *part);
-    void handle_connect_epoll(int peer_fd);
+    void handle_op_part(cluster_op_part_t *part);
-    void handle_peer_epoll(int peer_fd, int epoll_events);
-    void on_connect_peer(osd_num_t peer_osd, int peer_fd);
-    void check_peer_config(osd_client_t & cl);
-    void cancel_osd_ops(osd_client_t & cl);
-    void cancel_out_op(osd_op_t *op);
-
-    bool try_send(osd_client_t & cl);
-    void send_replies();
-    void handle_send(ring_data_t *data, int peer_fd);
-
-    void read_requests();
-    void handle_read(ring_data_t *data, int peer_fd);
-    void handle_finished_read(osd_client_t & cl);
-    void handle_op_hdr(osd_client_t *cl);
-    void handle_reply_hdr(osd_client_t *cl);
 };

cpp-btree

@@ -0,0 +1 @@
+Subproject commit 5dc108754ad40d3b1d024f9bd7cca0595ef1a1db

dump_journal.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #define _LARGEFILE64_SOURCE
 #include <sys/types.h>
 #include <sys/ioctl.h>
@@ -94,7 +97,7 @@ void journal_dump_t::dump_block(void *buf)
     while (pos < journal_block)
     {
         journal_entry *je = (journal_entry*)(buf + pos);
-        if (je->magic != JOURNAL_MAGIC || je->type < JE_START || je->type > JE_DELETE)
+        if (je->magic != JOURNAL_MAGIC || je->type < JE_MIN || je->type > JE_MAX)
         {
             break;
         }
@@ -104,10 +107,11 @@ void journal_dump_t::dump_block(void *buf)
         {
             printf("je_start start=%08lx\n", je->start.journal_start);
         }
-        else if (je->type == JE_SMALL_WRITE)
+        else if (je->type == JE_SMALL_WRITE || je->type == JE_SMALL_WRITE_INSTANT)
         {
             printf(
-                "je_small_write oid=%lu:%lu ver=%lu offset=%u len=%u loc=%08lx",
+                "je_small_write%s oid=%lx:%lx ver=%lu offset=%u len=%u loc=%08lx",
+                je->type == JE_SMALL_WRITE_INSTANT ? "_instant" : "",
                 je->small_write.oid.inode, je->small_write.oid.stripe,
                 je->small_write.version, je->small_write.offset, je->small_write.len,
                 je->small_write.data_offset
@@ -139,21 +143,25 @@ void journal_dump_t::dump_block(void *buf)
             );
             printf("\n");
         }
-        else if (je->type == JE_BIG_WRITE)
+        else if (je->type == JE_BIG_WRITE || je->type == JE_BIG_WRITE_INSTANT)
         {
-            printf("je_big_write oid=%lu:%lu ver=%lu loc=%08lx\n", je->big_write.oid.inode, je->big_write.oid.stripe, je->big_write.version, je->big_write.location);
+            printf(
+                "je_big_write%s oid=%lx:%lx ver=%lu loc=%08lx\n",
+                je->type == JE_BIG_WRITE_INSTANT ? "_instant" : "",
+                je->big_write.oid.inode, je->big_write.oid.stripe, je->big_write.version, je->big_write.location
+            );
         }
         else if (je->type == JE_STABLE)
         {
-            printf("je_stable oid=%lu:%lu ver=%lu\n", je->stable.oid.inode, je->stable.oid.stripe, je->stable.version);
+            printf("je_stable oid=%lx:%lx ver=%lu\n", je->stable.oid.inode, je->stable.oid.stripe, je->stable.version);
         }
         else if (je->type == JE_ROLLBACK)
         {
-            printf("je_rollback oid=%lu:%lu ver=%lu\n", je->rollback.oid.inode, je->rollback.oid.stripe, je->rollback.version);
+            printf("je_rollback oid=%lx:%lx ver=%lu\n", je->rollback.oid.inode, je->rollback.oid.stripe, je->rollback.version);
         }
         else if (je->type == JE_DELETE)
         {
-            printf("je_delete oid=%lu:%lu ver=%lu\n", je->del.oid.inode, je->del.oid.stripe, je->del.version);
+            printf("je_delete oid=%lx:%lx ver=%lu\n", je->del.oid.inode, je->del.oid.stripe, je->del.version);
         }
         pos += je->size;
         entry++;

epoll_manager.cpp

@@ -0,0 +1,90 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+#include <sys/epoll.h>
+#include <sys/poll.h>
+#include <unistd.h>
+
+#include "epoll_manager.h"
+
+#define MAX_EPOLL_EVENTS 64
+
+epoll_manager_t::epoll_manager_t(ring_loop_t *ringloop)
+{
+    this->ringloop = ringloop;
+
+    epoll_fd = epoll_create(1);
+    if (epoll_fd < 0)
+    {
+        throw std::runtime_error(std::string("epoll_create: ") + strerror(errno));
+    }
+
+    tfd = new timerfd_manager_t([this](int fd, bool wr, std::function<void(int, int)> handler) { set_fd_handler(fd, wr, handler); });
+
+    handle_epoll_events();
+}
+
+epoll_manager_t::~epoll_manager_t()
+{
+    if (tfd)
+    {
+        delete tfd;
+        tfd = NULL;
+    }
+    close(epoll_fd);
+}
+
+void epoll_manager_t::set_fd_handler(int fd, bool wr, std::function<void(int, int)> handler)
+{
+    if (handler != NULL)
+    {
+        bool exists = epoll_handlers.find(fd) != epoll_handlers.end();
+        epoll_event ev;
+        ev.data.fd = fd;
+        ev.events = (wr ? EPOLLOUT : 0) | EPOLLIN | EPOLLRDHUP | EPOLLET;
+        if (epoll_ctl(epoll_fd, exists ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev) < 0)
+        {
+            throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
+        }
+        epoll_handlers[fd] = handler;
+    }
+    else
+    {
+        if (epoll_ctl(epoll_fd, EPOLL_CTL_DEL, fd, NULL) < 0 && errno != ENOENT)
+        {
+            throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
+        }
+        epoll_handlers.erase(fd);
+    }
+}
+
+void epoll_manager_t::handle_epoll_events()
+{
+    io_uring_sqe *sqe = ringloop->get_sqe();
+    if (!sqe)
+    {
+        throw std::runtime_error("can't get SQE, will fall out of sync with EPOLLET");
+    }
+    ring_data_t *data = ((ring_data_t*)sqe->user_data);
+    my_uring_prep_poll_add(sqe, epoll_fd, POLLIN);
+    data->callback = [this](ring_data_t *data)
+    {
+        if (data->res < 0)
+        {
+            throw std::runtime_error(std::string("epoll failed: ") + strerror(-data->res));
+        }
+        handle_epoll_events();
+    };
+    ringloop->submit();
+    int nfds;
+    epoll_event events[MAX_EPOLL_EVENTS];
+    do
+    {
+        nfds = epoll_wait(epoll_fd, events, MAX_EPOLL_EVENTS, 0);
+        for (int i = 0; i < nfds; i++)
+        {
+            auto & cb = epoll_handlers[events[i].data.fd];
+            cb(events[i].data.fd, events[i].events);
+        }
+    } while (nfds == MAX_EPOLL_EVENTS);
+}

epoll_manager.h

@@ -0,0 +1,23 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+#pragma once
+
+#include <map>
+
+#include "ringloop.h"
+#include "timerfd_manager.h"
+
+class epoll_manager_t
+{
+    int epoll_fd;
+    ring_loop_t *ringloop;
+    std::map<int, std::function<void(int, int)>> epoll_handlers;
+public:
+    epoll_manager_t(ring_loop_t *ringloop);
+    ~epoll_manager_t();
+    void set_fd_handler(int fd, bool wr, std::function<void(int, int)> handler);
+    void handle_epoll_events();
+
+    timerfd_manager_t *tfd;
+};

etcd_state_client.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #include "osd_ops.h"
 #include "pg_states.h"
 #include "etcd_state_client.h"
@@ -43,6 +46,53 @@ void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int t
     http_request_json(tfd, etcd_address, req, timeout, callback);
 }
 
+void etcd_state_client_t::parse_config(json11::Json & config)
+{
+    this->etcd_addresses.clear();
+    if (config["etcd_address"].is_string())
+    {
+        std::string ea = config["etcd_address"].string_value();
+        while (1)
+        {
+            int pos = ea.find(',');
+            std::string addr = pos >= 0 ? ea.substr(0, pos) : ea;
+            if (addr.length() > 0)
+            {
+                if (addr.find('/') < 0)
+                    addr += "/v3";
+                this->etcd_addresses.push_back(addr);
+            }
+            if (pos >= 0)
+                ea = ea.substr(pos+1);
+            else
+                break;
+        }
+    }
+    else if (config["etcd_address"].array_items().size())
+    {
+        for (auto & ea: config["etcd_address"].array_items())
+        {
+            std::string addr = ea.string_value();
+            if (addr != "")
+            {
+                if (addr.find('/') < 0)
+                    addr += "/v3";
+                this->etcd_addresses.push_back(addr);
+            }
+        }
+    }
+    this->etcd_prefix = config["etcd_prefix"].string_value();
+    if (this->etcd_prefix == "")
+    {
+        this->etcd_prefix = "/vitastor";
+    }
+    else if (this->etcd_prefix[0] != '/')
+    {
+        this->etcd_prefix = "/"+this->etcd_prefix;
+    }
+    this->log_level = config["log_level"].int64_value();
+}
+
 void etcd_state_client_t::start_etcd_watcher()
 {
     std::string etcd_address = etcd_addresses[rand() % etcd_addresses.size()];
@@ -93,7 +143,10 @@ void etcd_state_client_t::start_etcd_watcher()
                     parse_state(kv.first, kv.second);
                 }
                 // React to changes
-                on_change_hook(changes);
+                if (on_change_hook != NULL)
+                {
+                    on_change_hook(changes);
+                }
             }
         }
         if (msg->eof)
@@ -183,6 +236,11 @@ void etcd_state_client_t::load_global_config()
 void etcd_state_client_t::load_pgs()
 {
     json11::Json::array txn = {
+        json11::Json::object {
+            { "request_range", json11::Json::object {
+                { "key", base64_encode(etcd_prefix+"/config/pools") },
+            } }
+        },
         json11::Json::object {
             { "request_range", json11::Json::object {
                 { "key", base64_encode(etcd_prefix+"/config/pgs") },
@@ -208,7 +266,7 @@ void etcd_state_client_t::load_pgs()
         },
     };
     json11::Json::object req = { { "success", txn } };
-    json11::Json checks = load_pgs_checks_hook();
+    json11::Json checks = load_pgs_checks_hook != NULL ? load_pgs_checks_hook() : json11::Json();
     if (checks.array_items().size() > 0)
     {
         req["compare"] = checks;
@@ -243,47 +301,162 @@ void etcd_state_client_t::load_pgs()
 
 void etcd_state_client_t::parse_state(const std::string & key, const json11::Json & value)
 {
-    if (key == etcd_prefix+"/config/pgs")
+    if (key == etcd_prefix+"/config/pools")
     {
-        for (auto & pg_item: this->pg_config)
+        for (auto & pool_item: this->pool_config)
         {
-            pg_item.second.exists = false;
+            pool_item.second.exists = false;
         }
-        for (auto & pg_item: value["items"].object_items())
+        for (auto & pool_item: value.object_items())
         {
-            pg_num_t pg_num = stoull_full(pg_item.first);
+            pool_id_t pool_id = stoull_full(pool_item.first);
-            if (!pg_num)
+            if (!pool_id || pool_id >= POOL_ID_MAX)
             {
-                printf("Bad key in PG configuration: %s (must be a number), skipped\n", pg_item.first.c_str());
+                printf("Pool ID %s is invalid (must be a number less than 0x%x), skipping pool\n", pool_item.first.c_str(), POOL_ID_MAX);
                 continue;
             }
-            this->pg_config[pg_num].exists = true;
+            if (pool_item.second["pg_size"].uint64_value() < 1 ||
-            this->pg_config[pg_num].pause = pg_item.second["pause"].bool_value();
+                pool_item.second["scheme"] == "xor" && pool_item.second["pg_size"].uint64_value() < 3)
-            this->pg_config[pg_num].primary = pg_item.second["primary"].uint64_value();
-            this->pg_config[pg_num].target_set.clear();
-            for (auto pg_osd: pg_item.second["osd_set"].array_items())
             {
-                this->pg_config[pg_num].target_set.push_back(pg_osd.uint64_value());
+                printf("Pool %u has invalid pg_size, skipping pool\n", pool_id);
+                continue;
             }
-            if (this->pg_config[pg_num].target_set.size() != 3)
+            if (pool_item.second["pg_minsize"].uint64_value() < 1 ||
+                pool_item.second["pg_minsize"].uint64_value() > pool_item.second["pg_size"].uint64_value() ||
+                pool_item.second["pg_minsize"].uint64_value() < (pool_item.second["pg_size"].uint64_value() - 1))
             {
-                printf("Bad PG %u config format: incorrect osd_set = %s\n", pg_num, pg_item.second["osd_set"].dump().c_str());
+                printf("Pool %u has invalid pg_minsize, skipping pool\n", pool_id);
-                this->pg_config[pg_num].target_set.resize(3);
+                continue;
-                this->pg_config[pg_num].pause = true;
             }
+            if (pool_item.second["pg_count"].uint64_value() < 1)
+            {
+                printf("Pool %u has invalid pg_count, skipping pool\n", pool_id);
+                continue;
+            }
+            if (pool_item.second["name"].string_value() == "")
+            {
+                printf("Pool %u has empty name, skipping pool\n", pool_id);
+                continue;
+            }
+            if (pool_item.second["scheme"] != "replicated" && pool_item.second["scheme"] != "xor")
+            {
+                printf("Pool %u has invalid coding scheme (only \"xor\" and \"replicated\" are allowed), skipping pool\n", pool_id);
+                continue;
+            }
+            if (pool_item.second["max_osd_combinations"].uint64_value() > 0 &&
+                pool_item.second["max_osd_combinations"].uint64_value() < 100)
+            {
+                printf("Pool %u has invalid max_osd_combinations (must be at least 100), skipping pool\n", pool_id);
+                continue;
+            }
+            auto & parsed_cfg = this->pool_config[pool_id];
+            parsed_cfg.exists = true;
+            parsed_cfg.id = pool_id;
+            parsed_cfg.name = pool_item.second["name"].string_value();
+            parsed_cfg.scheme = pool_item.second["scheme"] == "replicated" ? POOL_SCHEME_REPLICATED : POOL_SCHEME_XOR;
+            parsed_cfg.pg_size = pool_item.second["pg_size"].uint64_value();
+            parsed_cfg.pg_minsize = pool_item.second["pg_minsize"].uint64_value();
+            parsed_cfg.pg_count = pool_item.second["pg_count"].uint64_value();
+            parsed_cfg.failure_domain = pool_item.second["failure_domain"].string_value();
+            parsed_cfg.pg_stripe_size = pool_item.second["pg_stripe_size"].uint64_value();
+            if (!parsed_cfg.pg_stripe_size)
+            {
+                parsed_cfg.pg_stripe_size = DEFAULT_PG_STRIPE_SIZE;
+            }
+            parsed_cfg.max_osd_combinations = pool_item.second["max_osd_combinations"].uint64_value();
+            if (!parsed_cfg.max_osd_combinations)
+            {
+                parsed_cfg.max_osd_combinations = 10000;
+            }
+            for (auto & pg_item: parsed_cfg.pg_config)
+            {
+                if (pg_item.second.target_set.size() != parsed_cfg.pg_size)
+                {
+                    printf("Pool %u PG %u configuration is invalid: osd_set size %lu != pool pg_size %lu\n",
+                        pool_id, pg_item.first, pg_item.second.target_set.size(), parsed_cfg.pg_size);
+                    pg_item.second.pause = true;
+                }
+            }
+        }
+    }
+    else if (key == etcd_prefix+"/config/pgs")
+    {
+        for (auto & pool_item: this->pool_config)
+        {
+            for (auto & pg_item: pool_item.second.pg_config)
+            {
+                pg_item.second.exists = false;
+            }
+        }
+        for (auto & pool_item: value["items"].object_items())
+        {
+            pool_id_t pool_id = stoull_full(pool_item.first);
+            if (!pool_id || pool_id >= POOL_ID_MAX)
+            {
+                printf("Pool ID %s is invalid in PG configuration (must be a number less than 0x%x), skipping pool\n", pool_item.first.c_str(), POOL_ID_MAX);
+                continue;
+            }
+            for (auto & pg_item: pool_item.second.object_items())
+            {
+                pg_num_t pg_num = stoull_full(pg_item.first);
+                if (!pg_num)
+                {
+                    printf("Bad key in pool %u PG configuration: %s (must be a number), skipped\n", pool_id, pg_item.first.c_str());
+                    continue;
+                }
+                auto & parsed_cfg = this->pool_config[pool_id].pg_config[pg_num];
+                parsed_cfg.exists = true;
+                parsed_cfg.pause = pg_item.second["pause"].bool_value();
+                parsed_cfg.primary = pg_item.second["primary"].uint64_value();
+                parsed_cfg.target_set.clear();
+                for (auto & pg_osd: pg_item.second["osd_set"].array_items())
+                {
+                    parsed_cfg.target_set.push_back(pg_osd.uint64_value());
+                }
+                if (parsed_cfg.target_set.size() != pool_config[pool_id].pg_size)
+                {
+                    printf("Pool %u PG %u configuration is invalid: osd_set size %lu != pool pg_size %lu\n",
+                        pool_id, pg_num, parsed_cfg.target_set.size(), pool_config[pool_id].pg_size);
+                    parsed_cfg.pause = true;
+                }
+            }
+        }
+        for (auto & pool_item: this->pool_config)
+        {
+            int n = 0;
+            for (auto pg_it = pool_item.second.pg_config.begin(); pg_it != pool_item.second.pg_config.end(); pg_it++)
+            {
+                if (pg_it->second.exists && pg_it->first != ++n)
+                {
+                    printf(
+                        "Invalid pool %u PG configuration: PG numbers don't cover whole 1..%lu range\n",
+                        pool_item.second.id, pool_item.second.pg_config.size()
+                    );
+                    for (pg_it = pool_item.second.pg_config.begin(); pg_it != pool_item.second.pg_config.end(); pg_it++)
+                    {
+                        pg_it->second.exists = false;
+                    }
+                    n = 0;
+                    break;
+                }
+            }
+            pool_item.second.real_pg_count = n;
         }
     }
     else if (key.substr(0, etcd_prefix.length()+12) == etcd_prefix+"/pg/history/")
     {
-        // <etcd_prefix>/pg/history/%d
+        // <etcd_prefix>/pg/history/%d/%d
-        pg_num_t pg_num = stoull_full(key.substr(etcd_prefix.length()+12));
+        pool_id_t pool_id = 0;
-        if (!pg_num)
+        pg_num_t pg_num = 0;
+        char null_byte = 0;
+        sscanf(key.c_str() + etcd_prefix.length()+12, "%u/%u%c", &pool_id, &pg_num, &null_byte);
+        if (!pool_id || pool_id >= POOL_ID_MAX || !pg_num || null_byte != 0)
         {
             printf("Bad etcd key %s, ignoring\n", key.c_str());
         }
         else
         {
-            auto & pg_cfg = this->pg_config[pg_num];
+            auto & pg_cfg = this->pool_config[pool_id].pg_config[pg_num];
             pg_cfg.target_history.clear();
             pg_cfg.all_peers.clear();
             // Refuse to start PG if any set of the <osd_sets> has no live OSDs
@@ -301,20 +474,29 @@ void etcd_state_client_t::parse_state(const std::string & key, const json11::Jso
             {
                 pg_cfg.all_peers.push_back(pg_osd.uint64_value());
             }
+            // Read epoch
+            pg_cfg.epoch = value["epoch"].uint64_value();
+            if (on_change_pg_history_hook != NULL)
+            {
+                on_change_pg_history_hook(pool_id, pg_num);
+            }
         }
     }
     else if (key.substr(0, etcd_prefix.length()+10) == etcd_prefix+"/pg/state/")
     {
-        // <etcd_prefix>/pg/state/%d
+        // <etcd_prefix>/pg/state/%d/%d
-        pg_num_t pg_num = stoull_full(key.substr(etcd_prefix.length()+10));
+        pool_id_t pool_id = 0;
-        if (!pg_num)
+        pg_num_t pg_num = 0;
+        char null_byte = 0;
+        sscanf(key.c_str() + etcd_prefix.length()+10, "%u/%u%c", &pool_id, &pg_num, &null_byte);
+        if (!pool_id || pool_id >= POOL_ID_MAX || !pg_num || null_byte != 0)
         {
             printf("Bad etcd key %s, ignoring\n", key.c_str());
         }
         else if (value.is_null())
         {
-            this->pg_config[pg_num].cur_primary = 0;
+            this->pool_config[pool_id].pg_config[pg_num].cur_primary = 0;
-            this->pg_config[pg_num].cur_state = 0;
+            this->pool_config[pool_id].pg_config[pg_num].cur_state = 0;
         }
         else
         {
@@ -333,7 +515,7 @@ void etcd_state_client_t::parse_state(const std::string & key, const json11::Jso
                 }
                 if (i >= pg_state_bit_count)
                 {
-                    printf("Unexpected PG %u state keyword in etcd: %s\n", pg_num, e.dump().c_str());
+                    printf("Unexpected pool %u PG %u state keyword in etcd: %s\n", pool_id, pg_num, e.dump().c_str());
                     return;
                 }
             }
@@ -342,11 +524,11 @@ void etcd_state_client_t::parse_state(const std::string & key, const json11::Jso
                 (state & PG_PEERING) && state != PG_PEERING ||
                 (state & PG_INCOMPLETE) && state != PG_INCOMPLETE)
             {
-                printf("Unexpected PG %u state in etcd: primary=%lu, state=%s\n", pg_num, cur_primary, value["state"].dump().c_str());
+                printf("Unexpected pool %u PG %u state in etcd: primary=%lu, state=%s\n", pool_id, pg_num, cur_primary, value["state"].dump().c_str());
                 return;
             }
-            this->pg_config[pg_num].cur_primary = cur_primary;
+            this->pool_config[pool_id].pg_config[pg_num].cur_primary = cur_primary;
-            this->pg_config[pg_num].cur_state = state;
+            this->pool_config[pool_id].pg_config[pg_num].cur_state = state;
         }
     }
     else if (key.substr(0, etcd_prefix.length()+11) == etcd_prefix+"/osd/state/")

etcd_state_client.h

@@ -1,5 +1,9 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #pragma once
 
+#include "osd_id.h"
 #include "http_client.h"
 #include "timerfd_manager.h"
 
@@ -12,6 +16,14 @@
 #define ETCD_SLOW_TIMEOUT 5000
 #define ETCD_QUICK_TIMEOUT 1000
 
+#define DEFAULT_PG_STRIPE_SIZE 4*1024*1024
+
+struct json_kv_t
+{
+    std::string key;
+    json11::Json value;
+};
+
 struct pg_config_t
 {
     bool exists;
@@ -22,12 +34,22 @@ struct pg_config_t
     bool pause;
     osd_num_t cur_primary;
     int cur_state;
+    uint64_t epoch;
 };
 
-struct json_kv_t
+struct pool_config_t
 {
-    std::string key;
+    bool exists;
-    json11::Json value;
+    pool_id_t id;
+    std::string name;
+    uint64_t scheme;
+    uint64_t pg_size, pg_minsize;
+    uint64_t pg_count;
+    uint64_t real_pg_count;
+    std::string failure_domain;
+    uint64_t max_osd_combinations;
+    uint64_t pg_stripe_size;
+    std::map<pg_num_t, pg_config_t> pg_config;
 };
 
 struct etcd_state_client_t
@@ -40,14 +62,15 @@ struct etcd_state_client_t
     int etcd_watches_initialised = 0;
     uint64_t etcd_watch_revision = 0;
     websocket_t *etcd_watch_ws = NULL;
-    std::map<pg_num_t, pg_config_t> pg_config;
+    std::map<pool_id_t, pool_config_t> pool_config;
     std::map<osd_num_t, json11::Json> peer_states;
 
     std::function<void(json11::Json::object &)> on_change_hook;
     std::function<void(json11::Json::object &)> on_load_config_hook;
     std::function<json11::Json()> load_pgs_checks_hook;
     std::function<void(bool)> on_load_pgs_hook;
-    std::function<void(uint64_t)> on_change_osd_state_hook;
+    std::function<void(pool_id_t, pg_num_t)> on_change_pg_history_hook;
+    std::function<void(osd_num_t)> on_change_osd_state_hook;
 
     json_kv_t parse_etcd_kv(const json11::Json & kv_json);
     void etcd_call(std::string api, json11::Json payload, int timeout, std::function<void(std::string, json11::Json)> callback);
@@ -56,4 +79,5 @@ struct etcd_state_client_t
     void load_global_config();
     void load_pgs();
     void parse_state(const std::string & key, const json11::Json & value);
+    void parse_config(json11::Json & config);
 };

fio_cluster.cpp

@@ -0,0 +1,338 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+// FIO engine to test cluster I/O
+//
+// Random write:
+//
+// fio -thread -ioengine=./libfio_cluster.so -name=test -bs=4k -direct=1 -fsync=16 -iodepth=16 -rw=randwrite \
+//     -etcd=127.0.0.1:2379 [-etcd_prefix=/vitastor] -pool=1 -inode=1 -size=1000M
+//
+// Linear write:
+//
+// fio -thread -ioengine=./libfio_cluster.so -name=test -bs=128k -direct=1 -fsync=32 -iodepth=32 -rw=write \
+//     -etcd=127.0.0.1:2379 [-etcd_prefix=/vitastor] -pool=1 -inode=1 -size=1000M
+//
+// Random read (run with -iodepth=32 or -iodepth=1):
+//
+// fio -thread -ioengine=./libfio_cluster.so -name=test -bs=4k -direct=1 -iodepth=32 -rw=randread \
+//     -etcd=127.0.0.1:2379 [-etcd_prefix=/vitastor] -pool=1 -inode=1 -size=1000M
+
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <netinet/tcp.h>
+
+#include <vector>
+#include <unordered_map>
+
+#include "epoll_manager.h"
+#include "cluster_client.h"
+extern "C" {
+#define CONFIG_HAVE_GETTID
+#define CONFIG_PWRITEV2
+#include "fio/fio.h"
+#include "fio/optgroup.h"
+}
+
+struct sec_data
+{
+    ring_loop_t *ringloop = NULL;
+    epoll_manager_t *epmgr = NULL;
+    cluster_client_t *cli = NULL;
+    bool last_sync = false;
+    /* The list of completed io_u structs. */
+    std::vector<io_u*> completed;
+    uint64_t op_n = 0, inflight = 0;
+    bool trace = false;
+};
+
+struct sec_options
+{
+    int __pad;
+    char *etcd_host = NULL;
+    char *etcd_prefix = NULL;
+    uint64_t pool = 0;
+    uint64_t inode = 0;
+    int cluster_log = 0;
+    int trace = 0;
+};
+
+static struct fio_option options[] = {
+    {
+        .name   = "etcd",
+        .lname  = "etcd address",
+        .type   = FIO_OPT_STR_STORE,
+        .off1   = offsetof(struct sec_options, etcd_host),
+        .help   = "etcd address in the form HOST:PORT[/PATH]",
+        .category = FIO_OPT_C_ENGINE,
+        .group  = FIO_OPT_G_FILENAME,
+    },
+    {
+        .name   = "etcd",
+        .lname  = "etcd key prefix",
+        .type   = FIO_OPT_STR_STORE,
+        .off1   = offsetof(struct sec_options, etcd_prefix),
+        .help   = "etcd key prefix, by default /vitastor",
+        .category = FIO_OPT_C_ENGINE,
+        .group  = FIO_OPT_G_FILENAME,
+    },
+    {
+        .name   = "pool",
+        .lname  = "pool number for the inode",
+        .type   = FIO_OPT_INT,
+        .off1   = offsetof(struct sec_options, pool),
+        .help   = "pool number for the inode to run tests on",
+        .category = FIO_OPT_C_ENGINE,
+        .group  = FIO_OPT_G_FILENAME,
+    },
+    {
+        .name   = "inode",
+        .lname  = "inode to run tests on",
+        .type   = FIO_OPT_INT,
+        .off1   = offsetof(struct sec_options, inode),
+        .help   = "inode to run tests on (1 by default)",
+        .category = FIO_OPT_C_ENGINE,
+        .group  = FIO_OPT_G_FILENAME,
+    },
+    {
+        .name   = "cluster_log_level",
+        .lname  = "cluster log level",
+        .type   = FIO_OPT_BOOL,
+        .off1   = offsetof(struct sec_options, cluster_log),
+        .help   = "Set log level for the Vitastor client",
+        .def    = "0",
+        .category = FIO_OPT_C_ENGINE,
+        .group  = FIO_OPT_G_FILENAME,
+    },
+    {
+        .name   = "osd_trace",
+        .lname  = "OSD trace",
+        .type   = FIO_OPT_BOOL,
+        .off1   = offsetof(struct sec_options, trace),
+        .help   = "Trace OSD operations",
+        .def    = "0",
+        .category = FIO_OPT_C_ENGINE,
+        .group  = FIO_OPT_G_FILENAME,
+    },
+    {
+        .name = NULL,
+    },
+};
+
+static int sec_setup(struct thread_data *td)
+{
+    sec_data *bsd;
+
+    bsd = new sec_data;
+    if (!bsd)
+    {
+        td_verror(td, errno, "calloc");
+        return 1;
+    }
+    td->io_ops_data = bsd;
+
+    if (!td->files_index)
+    {
+        add_file(td, "osd_cluster", 0, 0);
+        td->o.nr_files = td->o.nr_files ? : 1;
+        td->o.open_files++;
+    }
+
+    return 0;
+}
+
+static void sec_cleanup(struct thread_data *td)
+{
+    sec_data *bsd = (sec_data*)td->io_ops_data;
+    if (bsd)
+    {
+        delete bsd->cli;
+        delete bsd->epmgr;
+        delete bsd->ringloop;
+        bsd->cli = NULL;
+        bsd->epmgr = NULL;
+        bsd->ringloop = NULL;
+    }
+}
+
+/* Connect to the server from each thread. */
+static int sec_init(struct thread_data *td)
+{
+    sec_options *o = (sec_options*)td->eo;
+    sec_data *bsd = (sec_data*)td->io_ops_data;
+
+    json11::Json cfg = json11::Json::object {
+        { "etcd_address", std::string(o->etcd_host) },
+        { "etcd_prefix", std::string(o->etcd_prefix ? o->etcd_prefix : "/vitastor") },
+        { "log_level", o->cluster_log },
+    };
+
+    if (o->pool)
+        o->inode = (o->inode & ((1l << (64-POOL_ID_BITS)) - 1)) | (o->pool << (64-POOL_ID_BITS));
+    if (!(o->inode >> (64-POOL_ID_BITS)))
+    {
+        td_verror(td, EINVAL, "pool is missing");
+        return 1;
+    }
+    bsd->ringloop = new ring_loop_t(512);
+    bsd->epmgr = new epoll_manager_t(bsd->ringloop);
+    bsd->cli = new cluster_client_t(bsd->ringloop, bsd->epmgr->tfd, cfg);
+
+    bsd->trace = o->trace ? true : false;
+
+    return 0;
+}
+
+/* Begin read or write request. */
+static enum fio_q_status sec_queue(struct thread_data *td, struct io_u *io)
+{
+    sec_options *opt = (sec_options*)td->eo;
+    sec_data *bsd = (sec_data*)td->io_ops_data;
+    int n = bsd->op_n;
+
+    fio_ro_check(td, io);
+    if (io->ddir == DDIR_SYNC && bsd->last_sync)
+    {
+        return FIO_Q_COMPLETED;
+    }
+
+    io->engine_data = bsd;
+    cluster_op_t *op = new cluster_op_t;
+
+    switch (io->ddir)
+    {
+    case DDIR_READ:
+        op->opcode = OSD_OP_READ;
+        op->inode = opt->inode;
+        op->offset = io->offset;
+        op->len = io->xfer_buflen;
+        op->iov.push_back(io->xfer_buf, io->xfer_buflen);
+        bsd->last_sync = false;
+        break;
+    case DDIR_WRITE:
+        op->opcode = OSD_OP_WRITE;
+        op->inode = opt->inode;
+        op->offset = io->offset;
+        op->len = io->xfer_buflen;
+        op->iov.push_back(io->xfer_buf, io->xfer_buflen);
+        bsd->last_sync = false;
+        break;
+    case DDIR_SYNC:
+        op->opcode = OSD_OP_SYNC;
+        bsd->last_sync = true;
+        break;
+    default:
+        io->error = EINVAL;
+        return FIO_Q_COMPLETED;
+    }
+
+    op->callback = [io, n](cluster_op_t *op)
+    {
+        io->error = op->retval < 0 ? -op->retval : 0;
+        sec_data *bsd = (sec_data*)io->engine_data;
+        bsd->inflight--;
+        bsd->completed.push_back(io);
+        if (bsd->trace)
+        {
+            printf("--- %s n=%d retval=%d\n", io->ddir == DDIR_READ ? "READ" :
+                (io->ddir == DDIR_WRITE ? "WRITE" : "SYNC"), n, op->retval);
+        }
+        delete op;
+    };
+
+    if (opt->trace)
+    {
+        if (io->ddir == DDIR_SYNC)
+        {
+            printf("+++ SYNC # %d\n", n);
+        }
+        else
+        {
+            printf("+++ %s # %d 0x%llx+%llx\n",
+                io->ddir == DDIR_READ ? "READ" : "WRITE",
+                n, io->offset, io->xfer_buflen);
+        }
+    }
+
+    io->error = 0;
+    bsd->inflight++;
+    bsd->op_n++;
+    bsd->cli->execute(op);
+
+    if (io->error != 0)
+        return FIO_Q_COMPLETED;
+    return FIO_Q_QUEUED;
+}
+
+static int sec_getevents(struct thread_data *td, unsigned int min, unsigned int max, const struct timespec *t)
+{
+    sec_data *bsd = (sec_data*)td->io_ops_data;
+    while (true)
+    {
+        bsd->ringloop->loop();
+        if (bsd->completed.size() >= min)
+            break;
+        bsd->ringloop->wait();
+    }
+    return bsd->completed.size();
+}
+
+static struct io_u *sec_event(struct thread_data *td, int event)
+{
+    sec_data *bsd = (sec_data*)td->io_ops_data;
+    if (bsd->completed.size() == 0)
+        return NULL;
+    /* FIXME We ignore the event number and assume fio calls us exactly once for [0..nr_events-1] */
+    struct io_u *ev = bsd->completed.back();
+    bsd->completed.pop_back();
+    return ev;
+}
+
+static int sec_io_u_init(struct thread_data *td, struct io_u *io)
+{
+    io->engine_data = NULL;
+    return 0;
+}
+
+static void sec_io_u_free(struct thread_data *td, struct io_u *io)
+{
+}
+
+static int sec_open_file(struct thread_data *td, struct fio_file *f)
+{
+    return 0;
+}
+
+static int sec_invalidate(struct thread_data *td, struct fio_file *f)
+{
+    return 0;
+}
+
+struct ioengine_ops ioengine = {
+    .name               = "vitastor_cluster",
+    .version            = FIO_IOOPS_VERSION,
+    .flags              = FIO_MEMALIGN | FIO_DISKLESSIO | FIO_NOEXTEND,
+    .setup              = sec_setup,
+    .init               = sec_init,
+    .queue              = sec_queue,
+    .getevents          = sec_getevents,
+    .event              = sec_event,
+    .cleanup            = sec_cleanup,
+    .open_file          = sec_open_file,
+    .invalidate         = sec_invalidate,
+    .io_u_init          = sec_io_u_init,
+    .io_u_free          = sec_io_u_free,
+    .option_struct_size = sizeof(struct sec_options),
+    .options            = options,
+};
+
+static void fio_init fio_sec_register(void)
+{
+    register_ioengine(&ioengine);
+}
+
+static void fio_exit fio_sec_unregister(void)
+{
+    unregister_ioengine(&ioengine);
+}

fio_engine.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 // FIO engine to test Blockstore
 //
 // Initialize storage for tests:
@@ -290,7 +293,7 @@ static int bs_invalidate(struct thread_data *td, struct fio_file *f)
 }
 
 struct ioengine_ops ioengine = {
-    .name               = "microceph_blockstore",
+    .name               = "vitastor_blockstore",
     .version            = FIO_IOOPS_VERSION,
     .flags              = FIO_MEMALIGN | FIO_DISKLESSIO | FIO_NOEXTEND,
     .setup              = bs_setup,

fio_sec_osd.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 // FIO engine to test Blockstore through Secondary OSD interface
 //
 // Prepare storage like in fio_engine.cpp, then start OSD with ./osd, then test it
@@ -5,7 +8,7 @@
 // Random write:
 //
 // fio -thread -ioengine=./libfio_sec_osd.so -name=test -bs=4k -direct=1 -fsync=16 -iodepth=16 -rw=randwrite \
-//     -host=127.0.0.1 -port=11203 [-single_primary=1] -size=1000M
+//     -host=127.0.0.1 -port=11203 [-block_size_order=17] [-single_primary=1] -size=1000M
 //
 // Linear write:
 //
@@ -53,6 +56,7 @@ struct sec_options
     int port = 0;
     int single_primary = 0;
     int trace = 0;
+    int block_order = 17;
 };
 
 static struct fio_option options[] = {
@@ -74,6 +78,15 @@ static struct fio_option options[] = {
         .category = FIO_OPT_C_ENGINE,
         .group  = FIO_OPT_G_FILENAME,
     },
+    {
+        .name   = "block_size_order",
+        .lname  = "Blockstore block size order",
+        .type   = FIO_OPT_INT,
+        .off1   = offsetof(struct sec_options, block_order),
+        .help   = "Blockstore block size order (size = 2^order)",
+        .category = FIO_OPT_C_ENGINE,
+        .group  = FIO_OPT_G_FILENAME,
+    },
     {
         .name   = "single_primary",
         .lname  = "Single Primary",
@@ -140,6 +153,8 @@ static int sec_init(struct thread_data *td)
 {
     sec_options *o = (sec_options*)td->eo;
     sec_data *bsd = (sec_data*)td->io_ops_data;
+    bsd->block_order = o->block_order == 0 ? 17 : o->block_order;
+    bsd->block_size = 1 << o->block_order;
 
     struct sockaddr_in addr;
     int r;
@@ -193,7 +208,7 @@ static enum fio_q_status sec_queue(struct thread_data *td, struct io_u *io)
     case DDIR_READ:
         if (!opt->single_primary)
         {
-            op.hdr.opcode = OSD_OP_SECONDARY_READ;
+            op.hdr.opcode = OSD_OP_SEC_READ;
             op.sec_rw.oid = {
                 .inode = 1,
                 .stripe = io->offset >> bsd->block_order,
@@ -214,7 +229,7 @@ static enum fio_q_status sec_queue(struct thread_data *td, struct io_u *io)
     case DDIR_WRITE:
         if (!opt->single_primary)
         {
-            op.hdr.opcode = OSD_OP_SECONDARY_WRITE;
+            op.hdr.opcode = OSD_OP_SEC_WRITE;
             op.sec_rw.oid = {
                 .inode = 1,
                 .stripe = io->offset >> bsd->block_order,
@@ -369,7 +384,7 @@ static int sec_invalidate(struct thread_data *td, struct fio_file *f)
 }
 
 struct ioengine_ops ioengine = {
-    .name               = "microceph_secondary_osd",
+    .name               = "vitastor_secondary_osd",
     .version            = FIO_IOOPS_VERSION,
     .flags              = FIO_MEMALIGN | FIO_DISKLESSIO | FIO_NOEXTEND,
     .setup              = sec_setup,

http_client.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #include <netinet/tcp.h>
 #include <sys/epoll.h>
 
@@ -50,8 +53,15 @@ struct http_co_t
 
     websocket_t ws;
 
+    int onstack = 0;
+    bool ended = false;
+
     ~http_co_t();
+    inline void stackin() { onstack++; }
+    inline void stackout() { onstack--; if (!onstack && ended) end(); }
+    inline void end() { ended = true; if (!onstack) { delete this; } }
     void start_connection();
+    void handle_events();
     void handle_connect_result();
     void submit_read();
     void submit_send();
@@ -137,7 +147,7 @@ void websocket_t::post_message(int type, const std::string & msg)
 
 void websocket_t::close()
 {
-    delete co;
+    co->end();
 }
 
 http_co_t::~http_co_t()
@@ -149,7 +159,7 @@ http_co_t::~http_co_t()
     }
     if (peer_fd >= 0)
     {
-        tfd->set_fd_handler(peer_fd, NULL);
+        tfd->set_fd_handler(peer_fd, false, NULL);
         close(peer_fd);
         peer_fd = -1;
     }
@@ -173,14 +183,15 @@ http_co_t::~http_co_t()
 
 void http_co_t::start_connection()
 {
+    stackin();
     int port = extract_port(host);
     struct sockaddr_in addr;
     int r;
     if ((r = inet_pton(AF_INET, host.c_str(), &addr.sin_addr)) != 1)
     {
         parsed.error_code = ENXIO;
-        // FIXME 'delete this' is ugly...
+        stackout();
-        delete this;
+        end();
         return;
     }
     addr.sin_family = AF_INET;
@@ -189,7 +200,8 @@ void http_co_t::start_connection()
     if (peer_fd < 0)
     {
         parsed.error_code = errno;
-        delete this;
+        stackout();
+        end();
         return;
     }
     fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) | O_NONBLOCK);
@@ -201,71 +213,86 @@ void http_co_t::start_connection()
             {
                 parsed.error_code = ETIME;
             }
-            delete this;
+            end();
         });
     }
-    tfd->set_fd_handler(peer_fd, [this](int peer_fd, int epoll_events)
-    {
-        this->epoll_events |= epoll_events;
-        if (state == HTTP_CO_CONNECTING)
-        {
-            handle_connect_result();
-        }
-        else
-        {
-            if (this->epoll_events & EPOLLIN)
-            {
-                submit_read();
-            }
-            else if (this->epoll_events & (EPOLLRDHUP|EPOLLERR))
-            {
-                delete this;
-            }
-        }
-    });
     epoll_events = 0;
     // Finally call connect
     r = ::connect(peer_fd, (sockaddr*)&addr, sizeof(addr));
     if (r < 0 && errno != EINPROGRESS)
     {
         parsed.error_code = errno;
-        delete this;
+        stackout();
+        end();
         return;
     }
+    tfd->set_fd_handler(peer_fd, true, [this](int peer_fd, int epoll_events)
+    {
+        this->epoll_events |= epoll_events;
+        handle_events();
+    });
     state = HTTP_CO_CONNECTING;
+    stackout();
+}
+
+void http_co_t::handle_events()
+{
+    stackin();
+    while (epoll_events)
+    {
+        if (state == HTTP_CO_CONNECTING)
+        {
+            handle_connect_result();
+        }
+        else
+        {
+            epoll_events &= ~EPOLLOUT;
+            if (epoll_events & EPOLLIN)
+            {
+                submit_read();
+            }
+            else if (epoll_events & (EPOLLRDHUP|EPOLLERR))
+            {
+                end();
+                break;
+            }
+        }
+    }
+    stackout();
 }
 
 void http_co_t::handle_connect_result()
 {
-    if (epoll_events & (EPOLLOUT | EPOLLERR))
+    stackin();
+    int result = 0;
+    socklen_t result_len = sizeof(result);
+    if (getsockopt(peer_fd, SOL_SOCKET, SO_ERROR, &result, &result_len) < 0)
     {
-        int result = 0;
+        result = errno;
-        socklen_t result_len = sizeof(result);
-        if (getsockopt(peer_fd, SOL_SOCKET, SO_ERROR, &result, &result_len) < 0)
-        {
-            result = errno;
-        }
-        if (result != 0)
-        {
-            parsed.error_code = result;
-            delete this;
-            return;
-        }
-        int one = 1;
-        setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
-        state = HTTP_CO_SENDING_REQUEST;
-        submit_send();
     }
-    else
+    if (result != 0)
     {
-        delete this;
+        parsed.error_code = result;
+        stackout();
+        end();
+        return;
     }
+    int one = 1;
+    setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
+    tfd->set_fd_handler(peer_fd, false, [this](int peer_fd, int epoll_events)
+    {
+        this->epoll_events |= epoll_events;
+        handle_events();
+    });
+    state = HTTP_CO_SENDING_REQUEST;
+    submit_send();
+    stackout();
 }
 
 void http_co_t::submit_read()
 {
+    stackin();
     int res;
-again:
     if (rbuf.size() != READ_BUFFER_SIZE)
     {
         rbuf.resize(READ_BUFFER_SIZE);
@@ -273,7 +300,6 @@ again:
     read_iov = { .iov_base = rbuf.data(), .iov_len = READ_BUFFER_SIZE };
     read_msg.msg_iov = &read_iov;
     read_msg.msg_iovlen = 1;
-    epoll_events = epoll_events & ~EPOLLIN;
     res = recvmsg(peer_fd, &read_msg, 0);
     if (res < 0)
     {
@@ -281,31 +307,26 @@ again:
     }
     if (res == -EAGAIN)
     {
-        res = 0;
+        epoll_events = epoll_events & ~EPOLLIN;
     }
-    if (res < 0)
+    else if (res <= 0)
     {
-        delete this;
+        // < 0 means error, 0 means EOF
-        return;
+        if (!res)
+            epoll_events = epoll_events & ~EPOLLIN;
+        end();
     }
-    response += std::string(rbuf.data(), res);
+    else
-    if (res == READ_BUFFER_SIZE)
     {
-        goto again;
+        response += std::string(rbuf.data(), res);
-    }
+        handle_read();
-    if (!handle_read())
-    {
-        return;
-    }
-    if (res < READ_BUFFER_SIZE && (epoll_events & (EPOLLRDHUP|EPOLLERR)))
-    {
-        delete this;
-        return;
     }
+    stackout();
 }
 
 void http_co_t::submit_send()
 {
+    stackin();
     int res;
 again:
     if (sent < request.size())
@@ -313,7 +334,7 @@ again:
         send_iov = (iovec){ .iov_base = (void*)(request.c_str()+sent), .iov_len = request.size()-sent };
         send_msg.msg_iov = &send_iov;
         send_msg.msg_iovlen = 1;
-        res = sendmsg(peer_fd, &send_msg, 0);
+        res = sendmsg(peer_fd, &send_msg, MSG_NOSIGNAL);
         if (res < 0)
         {
             res = -errno;
@@ -324,14 +345,17 @@ again:
         }
         else if (res < 0)
         {
-            delete this;
+            stackout();
+            end();
             return;
         }
         sent += res;
         if (state == HTTP_CO_SENDING_REQUEST)
         {
             if (sent >= request.size())
+            {
                 state = HTTP_CO_REQUEST_SENT;
+            }
             else
                 goto again;
         }
@@ -342,10 +366,12 @@ again:
             goto again;
         }
     }
+    stackout();
 }
 
 bool http_co_t::handle_read()
 {
+    stackin();
     if (state == HTTP_CO_REQUEST_SENT)
     {
         int pos = response.find("\r\n\r\n");
@@ -380,7 +406,8 @@ bool http_co_t::handle_read()
                 if (!target_response_size)
                 {
                     // Sorry, unsupported response
-                    delete this;
+                    stackout();
+                    end();
                     return false;
                 }
             }
@@ -388,7 +415,8 @@ bool http_co_t::handle_read()
     }
     if (state == HTTP_CO_HEADERS_RECEIVED && target_response_size > 0 && response.size() >= target_response_size)
     {
-        delete this;
+        stackout();
+        end();
         return false;
     }
     if (state == HTTP_CO_CHUNKED && response.size() > 0)
@@ -416,7 +444,8 @@ bool http_co_t::handle_read()
         }
         if (parsed.eof)
         {
-            delete this;
+            stackout();
+            end();
             return false;
         }
         if (want_streaming && parsed.body.size() > 0)
@@ -433,11 +462,13 @@ bool http_co_t::handle_read()
             parsed.body = "";
         }
     }
+    stackout();
     return true;
 }
 
 void http_co_t::post_message(int type, const std::string & msg)
 {
+    stackin();
     if (state == HTTP_CO_WEBSOCKET)
     {
         request += ws_format_frame(type, msg.size());
@@ -449,6 +480,7 @@ void http_co_t::post_message(int type, const std::string & msg)
         ws_outbox += ws_format_frame(type, msg.size());
         ws_outbox += msg;
     }
+    stackout();
 }
 
 uint64_t stoull_full(const std::string & str, int base)

http_client.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #pragma once
 #include <string>
 #include <vector>

json11

@@ -0,0 +1 @@
+Subproject commit 97f06cb20c1e136fd37d58fb40f57dd8f8a3a4a7

lambda_size.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <iostream>
 #include <functional>
 #include <array>

lp/mon.js

@@ -1,858 +0,0 @@
-const http = require('http');
-const os = require('os');
-const WebSocket = require('ws');
-const LPOptimizer = require('./lp-optimizer.js');
-const stableStringify = require('./stable-stringify.js');
-
-class Mon
-{
-    static etcd_tree = {
-        config: {
-            global: null,
-            /* placement_tree = {
-                levels: { datacenter: 1, rack: 2, host: 3, osd: 4, ... },
-                nodes: { host1: { level: 'host', parent: 'rack1' }, ... },
-                failure_domain: 'host',
-            } */
-            placement_tree: null,
-            osd: {},
-            pgs: {},
-        },
-        osd: {
-            state: {},
-            stats: {},
-        },
-        mon: {
-            master: null,
-        },
-        pg: {
-            change_stamp: null,
-            state: {},
-            stats: {},
-            history: {},
-        },
-    }
-
-    constructor(config)
-    {
-        // FIXME: Maybe prefer local etcd
-        this.etcd_urls = [];
-        for (let url of config.etcd_url.split(/,/))
-        {
-            let scheme = 'http';
-            url = url.trim().replace(/^(https?):\/\//, (m, m1) => { scheme = m1; return ''; });
-            if (!/\/[^\/]/.exec(url))
-                url += '/v3';
-            this.etcd_urls.push(scheme+'://'+url);
-        }
-        this.etcd_prefix = config.etcd_prefix || '/rage';
-        this.etcd_prefix = this.etcd_prefix.replace(/\/\/+/g, '/').replace(/^\/?(.*[^\/])\/?$/, '/$1');
-        this.etcd_start_timeout = (config.etcd_start_timeout || 5) * 1000;
-        this.state = JSON.parse(JSON.stringify(Mon.etcd_tree));
-    }
-
-    async start()
-    {
-        await this.load_config();
-        await this.get_lease();
-        await this.become_master();
-        await this.load_cluster_state();
-        await this.start_watcher();
-        await this.recheck_pgs();
-    }
-
-    async load_config()
-    {
-        const res = await this.etcd_call('/txn', { success: [
-            { requestRange: { key: b64(this.etcd_prefix+'/config/global') } }
-        ] }, this.etcd_start_timeout, -1);
-        this.parse_kv(res.responses[0].response_range.kvs[0]);
-        this.check_config();
-    }
-
-    check_config()
-    {
-        this.config.etcd_mon_timeout = Number(this.config.etcd_mon_timeout) || 0;
-        if (this.config.etcd_mon_timeout <= 0)
-        {
-            this.config.etcd_mon_timeout = 1000;
-        }
-        this.config.etcd_mon_retries = Number(this.config.etcd_mon_retries) || 5;
-        if (this.config.etcd_mon_retries < 0)
-        {
-            this.config.etcd_mon_retries = 0;
-        }
-        this.config.mon_change_timeout = Number(this.config.mon_change_timeout) || 1000;
-        if (this.config.mon_change_timeout < 100)
-        {
-            this.config.mon_change_timeout = 100;
-        }
-        this.config.mon_stats_timeout = Number(this.config.mon_stats_timeout) || 1000;
-        if (this.config.mon_stats_timeout < 100)
-        {
-            this.config.mon_stats_timeout = 100;
-        }
-        // After this number of seconds, a dead OSD will be removed from PG distribution
-        this.config.osd_out_time = Number(this.config.osd_out_time) || 0;
-        if (!this.config.osd_out_time)
-        {
-            this.config.osd_out_time = 30*60; // 30 minutes by default
-        }
-        this.config.max_osd_combinations = Number(this.config.max_osd_combinations) || 10000;
-        if (this.config.max_osd_combinations < 100)
-        {
-            this.config.max_osd_combinations = 100;
-        }
-    }
-
-    async start_watcher(retries)
-    {
-        let retry = 0;
-        if (retries >= 0 && retries < 1)
-        {
-            retries = 1;
-        }
-        while (retries < 0 || retry < retries)
-        {
-            const base = 'ws'+this.etcd_urls[Math.floor(Math.random()*this.etcd_urls.length)].substr(4);
-            const ok = await new Promise((ok, no) =>
-            {
-                const timer_id = setTimeout(() =>
-                {
-                    this.ws.close();
-                    ok(false);
-                }, timeout);
-                this.ws = new WebSocket(base+'/watch');
-                this.ws.on('open', () =>
-                {
-                    if (timer_id)
-                        clearTimeout(timer_id);
-                    ok(true);
-                });
-            });
-            if (!ok)
-            {
-                this.ws = null;
-            }
-            retry++;
-        }
-        if (!this.ws)
-        {
-            this.die('Failed to open etcd watch websocket');
-        }
-        this.ws.send(JSON.stringify({
-            create_request: {
-                key: b64(this.etcd_prefix+'/'),
-                range_end: b64(this.etcd_prefix+'0'),
-                start_revision: ''+this.etcd_watch_revision,
-                watch_id: 1,
-            },
-        }));
-        this.ws.on('message', (msg) =>
-        {
-            let data;
-            try
-            {
-                data = JSON.parse(msg);
-            }
-            catch (e)
-            {
-            }
-            if (!data || !data.result || !data.result.events)
-            {
-                console.error('Garbage received from watch websocket: '+msg);
-            }
-            else
-            {
-                let stats_changed = false, changed = false;
-                console.log('Revision '+data.result.header.revision+' events: ');
-                for (const e of data.result.events)
-                {
-                    this.parse_kv(e.kv);
-                    const key = e.kv.key.substr(this.etcd_prefix.length);
-                    if (key.substr(0, 11) == '/osd/stats/' || key.substr(0, 10) == '/pg/stats/')
-                    {
-                        stats_changed = true;
-                    }
-                    else if (key != '/stats')
-                    {
-                        changed = true;
-                    }
-                    console.log(e);
-                }
-                if (stats_changed)
-                {
-                    this.schedule_update_stats();
-                }
-                if (changed)
-                {
-                    this.schedule_recheck();
-                }
-            }
-        });
-    }
-
-    async get_lease()
-    {
-        const max_ttl = this.config.etcd_mon_ttl + this.config.etcd_mon_timeout/1000*this.config.etcd_mon_retries;
-        const res = await this.etcd_call('/lease/grant', { TTL: max_ttl }, this.config.etcd_mon_timeout, this.config.etcd_mon_retries);
-        this.etcd_lease_id = res.ID;
-        setInterval(async () =>
-        {
-            const res = await this.etcd_call('/lease/keepalive', { ID: this.etcd_lease_id }, this.config.etcd_mon_timeout, this.config.etcd_mon_retries);
-            if (!res.result.TTL)
-            {
-                this.die('Lease expired');
-            }
-        }, config.etcd_mon_timeout);
-    }
-
-    async become_master()
-    {
-        const state = { ip: this.local_ips() };
-        while (1)
-        {
-            const res = await this.etcd_call('/txn', {
-                compare: [ { target: 'CREATE', create_revision: 0, key: b64(this.etcd_prefix+'/mon/master') } ],
-                success: [ { key: b64(this.etcd_prefix+'/mon/master'), value: b64(JSON.stringify(state)), lease: ''+this.etcd_lease_id } ],
-            }, this.etcd_start_timeout, 0);
-            if (!res.succeeded)
-            {
-                await new Promise(ok => setTimeout(ok, this.etcd_start_timeout));
-            }
-        }
-    }
-
-    async load_cluster_state()
-    {
-        const res = await this.etcd_call('/txn', { success: [
-            { requestRange: { key: b64(this.etcd_prefix+'/'), range_end: b64(this.etcd_prefix+'0') } },
-        ] }, this.etcd_start_timeout, -1);
-        this.etcd_watch_revision = BigInt(res.header.revision)+BigInt(1);
-        const data = JSON.parse(JSON.stringify(Mon.etcd_tree));
-        for (const response of res.responses)
-        {
-            for (const kv of response.response_range.kvs)
-            {
-                this.parse_kv(kv);
-            }
-        }
-        this.state = data;
-    }
-
-    all_osds()
-    {
-        return Object.keys(this.state.osd.stats);
-    }
-
-    get_osd_tree()
-    {
-        this.state.config.placement_tree = this.state.config.placement_tree||{};
-        const levels = this.state.config.placement_tree.levels||{};
-        levels.host = levels.host || 100;
-        levels.osd = levels.osd || 101;
-        const tree = { '': { children: [] } };
-        for (const node_id in this.state.config.placement_tree.nodes||{})
-        {
-            const node_cfg = this.state.config.placement_tree.nodes[node_id];
-            if (!node_id || /^\d/.exec(node_id) ||
-                !node_cfg.level || !levels[node_cfg.level])
-            {
-                // All nodes must have non-empty non-numeric IDs and valid levels
-                continue;
-            }
-            tree[node_id] = { id: node_id, level: node_cfg.level, parent: node_cfg.parent, children: [] };
-        }
-        // This requires monitor system time to be in sync with OSD system times (at least to some extent)
-        const down_time = Date.now()/1000 - this.config.osd_out_time;
-        for (const osd_num of this.all_osds().sort((a, b) => a - b))
-        {
-            const stat = this.state.osd.stats[osd_num];
-            if (stat.size && (this.state.osd.state[osd_num] || Number(stat.time) >= down_time))
-            {
-                // Numeric IDs are reserved for OSDs
-                const reweight = this.state.config.osd[osd_num] && Number(this.state.config.osd[osd_num].reweight) || 1;
-                tree[osd_num] = tree[osd_num] || { id: osd_num, parent: stat.host };
-                tree[osd_num].level = 'osd';
-                tree[osd_num].size = reweight * stat.size / 1024 / 1024 / 1024 / 1024; // terabytes
-                delete tree[osd_num].children;
-            }
-        }
-        for (const node_id in tree)
-        {
-            if (node_id === '')
-            {
-                continue;
-            }
-            const node_cfg = tree[node_id];
-            const node_level = levels[node_cfg.level] || node_cfg.level;
-            let parent_level = node_cfg.parent && tree[node_cfg.parent] && tree[node_cfg.parent].children
-                && tree[node_cfg.parent].level;
-            parent_level = parent_level ? (levels[parent_level] || parent_level) : null;
-            // Parent's level must be less than child's; OSDs must be leaves
-            const parent = parent_level && parent_level < node_level ? tree[node_cfg.parent] : '';
-            tree[parent].children.push(tree[node_id]);
-            delete node_cfg.parent;
-        }
-        return LPOptimizer.flatten_tree(tree[''].children, levels, this.state.config.failure_domain, 'osd');
-    }
-
-    async stop_all_pgs()
-    {
-        let has_online = false, paused = true;
-        for (const pg in this.state.config.pgs.items||{})
-        {
-            const cur_state = ((this.state.pg.state[pg]||{}).state||[]).join(',');
-            if (cur_state != '' && cur_state != 'offline')
-            {
-                has_online = true;
-            }
-            if (!this.state.config.pgs.items[pg].pause)
-            {
-                paused = false;
-            }
-        }
-        if (!paused)
-        {
-            console.log('Stopping all PGs before changing PG count');
-            const new_cfg = JSON.parse(JSON.stringify(this.state.config.pgs));
-            for (const pg in new_cfg.items)
-            {
-                new_cfg.items[pg].pause = true;
-            }
-            // Check that no OSDs change their state before we pause PGs
-            // Doing this we make sure that OSDs don't wake up in the middle of our "transaction"
-            // and can't see the old PG configuration
-            const checks = [];
-            for (const osd_num of this.all_osds())
-            {
-                const key = b64(this.etcd_prefix+'/osd/state/'+osd_num);
-                checks.push({ key, target: 'MOD', result: 'LESS', mod_revision: ''+this.etcd_watch_revision });
-            }
-            const res = await this.etcd_call('/txn', {
-                compare: [
-                    { key: b64(this.etcd_prefix+'/mon/master'), target: 'LEASE', lease: ''+this.etcd_lease_id },
-                    { key: b64(this.etcd_prefix+'/config/pgs'), target: 'MOD', mod_revision: ''+this.etcd_watch_revision, result: 'LESS' },
-                    ...checks,
-                ],
-                success: [
-                    { requestPut: { key: b64(this.etcd_prefix+'/config/pgs'), value: b64(JSON.stringify(new_cfg)) } },
-                ],
-            }, this.config.etcd_mon_timeout, 0);
-            if (!res.succeeded)
-            {
-                return false;
-            }
-            this.state.config.pgs = new_cfg;
-        }
-        return !has_online;
-    }
-
-    scale_pg_count(prev_pgs, pg_history, new_pg_count)
-    {
-        const old_pg_count = prev_pgs.length;
-        // Add all possibly intersecting PGs into the history of new PGs
-        if (!(new_pg_count % old_pg_count))
-        {
-            // New PG count is a multiple of the old PG count
-            const mul = (new_pg_count / old_pg_count);
-            for (let i = 0; i < new_pg_count; i++)
-            {
-                const old_i = Math.floor(new_pg_count / mul);
-                pg_history[i] = JSON.parse(JSON.stringify(this.state.pg.history[1+old_i]));
-            }
-        }
-        else if (!(old_pg_count % new_pg_count))
-        {
-            // Old PG count is a multiple of the new PG count
-            const mul = (old_pg_count / new_pg_count);
-            for (let i = 0; i < new_pg_count; i++)
-            {
-                pg_history[i] = {
-                    osd_sets: [],
-                    all_peers: [],
-                };
-                for (let j = 0; j < mul; j++)
-                {
-                    pg_history[i].osd_sets.push(prev_pgs[i*mul]);
-                    const hist = this.state.pg.history[1+i*mul+j];
-                    if (hist && hist.osd_sets && hist.osd_sets.length)
-                    {
-                        Array.prototype.push.apply(pg_history[i].osd_sets, hist.osd_sets);
-                    }
-                    if (hist && hist.all_peers && hist.all_peers.length)
-                    {
-                        Array.prototype.push.apply(pg_history[i].all_peers, hist.all_peers);
-                    }
-                }
-            }
-        }
-        else
-        {
-            // Any PG may intersect with any PG after non-multiple PG count change
-            // So, merge ALL PGs history
-            let all_sets = {};
-            let all_peers = {};
-            for (const pg of prev_pgs)
-            {
-                all_sets[pg.join(' ')] = pg;
-            }
-            for (const pg in this.state.pg.history)
-            {
-                const hist = this.state.pg.history[pg];
-                if (hist && hist.osd_sets)
-                {
-                    for (const pg of hist.osd_sets)
-                    {
-                        all_sets[pg.join(' ')] = pg;
-                    }
-                }
-                if (hist && hist.all_peers)
-                {
-                    for (const osd_num of hist.all_peers)
-                    {
-                        all_peers[osd_num] = Number(osd_num);
-                    }
-                }
-            }
-            all_sets = Object.values(all_sets);
-            all_peers = Object.values(all_peers);
-            for (let i = 0; i < new_pg_count; i++)
-            {
-                pg_history[i] = { osd_sets: all_sets, all_peers };
-            }
-        }
-        // Mark history keys for removed PGs as removed
-        for (let i = new_pg_count; i < old_pg_count; i++)
-        {
-            pg_history[i] = null;
-        }
-        if (old_pg_count < new_pg_count)
-        {
-            for (let i = new_pg_count-1; i >= 0; i--)
-            {
-                prev_pgs[i] = prev_pgs[Math.floor(i/new_pg_count*old_pg_count)];
-            }
-        }
-        else if (old_pg_count > new_pg_count)
-        {
-            for (let i = 0; i < new_pg_count; i++)
-            {
-                prev_pgs[i] = prev_pgs[Math.round(i/new_pg_count*old_pg_count)];
-            }
-            prev_pgs.splice(new_pg_count, old_pg_count-new_pg_count);
-        }
-    }
-
-    async save_new_pgs(prev_pgs, new_pgs, pg_history, tree_hash)
-    {
-        const txn = [], checks = [];
-        const pg_items = {};
-        new_pgs.map((osd_set, i) =>
-        {
-            osd_set = osd_set.map(osd_num => osd_num === LPOptimizer.NO_OSD ? 0 : osd_num);
-            const alive_set = osd_set.filter(osd_num => osd_num);
-            pg_items[i+1] = {
-                osd_set,
-                primary: alive_set.length ? alive_set[Math.floor(Math.random()*alive_set.length)] : 0,
-            };
-            if (prev_pgs[i] && prev_pgs[i].join(' ') != osd_set.join(' '))
-            {
-                pg_history[i] = pg_history[i] || {};
-                pg_history[i].osd_sets = pg_history[i].osd_sets || [];
-                pg_history[i].osd_sets.push(prev_pgs[i]);
-            }
-        });
-        for (let i = 0; i < new_pgs.length || i < prev_pgs.length; i++)
-        {
-            checks.push({
-                key: b64(this.etcd_prefix+'/pg/history/'+(i+1)),
-                target: 'MOD',
-                mod_revision: ''+this.etcd_watch_revision,
-                result: 'LESS',
-            });
-            if (pg_history[i])
-            {
-                txn.push({
-                    requestPut: {
-                        key: b64(this.etcd_prefix+'/pg/history/'+(i+1)),
-                        value: b64(JSON.stringify(pg_history[i])),
-                    },
-                });
-            }
-            else
-            {
-                txn.push({
-                    requestDeleteRange: {
-                        key: b64(this.etcd_prefix+'/pg/history/'+(i+1)),
-                    },
-                });
-            }
-        }
-        this.state.config.pgs = {
-            hash: tree_hash,
-            items: pg_items,
-        };
-        const res = await this.etcd_call('/txn', {
-            compare: [
-                { key: b64(this.etcd_prefix+'/mon/master'), target: 'LEASE', lease: ''+this.etcd_lease_id },
-                { key: b64(this.etcd_prefix+'/config/pgs'), target: 'MOD', mod_revision: ''+this.etcd_watch_revision, result: 'LESS' },
-                ...checks,
-            ],
-            success: [
-                { requestPut: { key: b64(this.etcd_prefix+'/config/pgs'), value: b64(JSON.stringify(this.state.config.pgs)) } },
-                ...txn,
-            ],
-        }, this.config.etcd_mon_timeout, 0);
-        return res.succeeded;
-    }
-
-    async recheck_pgs()
-    {
-        // Take configuration and state, check it against the stored configuration hash
-        // Recalculate PGs and save them to etcd if the configuration is changed
-        const tree_cfg = {
-            osd_tree: this.get_osd_tree(),
-            pg_count: this.config.pg_count || Object.keys(this.state.config.pgs.items||{}).length || 128,
-            max_osd_combinations: this.config.max_osd_combinations,
-        };
-        const tree_hash = sha1hex(stableStringify(tree_cfg));
-        if (this.state.config.pgs.hash != tree_hash)
-        {
-            // Something has changed
-            const prev_pgs = [];
-            for (const pg in this.state.config.pgs.items||{})
-            {
-                prev_pgs[pg-1] = this.state.config.pgs.items[pg].osd_set;
-            }
-            const pg_history = [];
-            const old_pg_count = prev_pgs.length;
-            let optimize_result;
-            if (old_pg_count > 0)
-            {
-                if (old_pg_count != tree_cfg.pg_count)
-                {
-                    // PG count changed. Need to bring all PGs down.
-                    if (!await this.stop_all_pgs())
-                    {
-                        this.schedule_recheck();
-                        return;
-                    }
-                    this.scale_pg_count(prev_pgs, pg_history, new_pg_count);
-                }
-                optimize_result = await LPOptimizer.optimize_change(prev_pgs, tree_cfg.osd_tree, tree_cfg.max_osd_combinations);
-            }
-            else
-            {
-                optimize_result = await LPOptimizer.optimize_initial(tree_cfg.osd_tree, tree_cfg.pg_count, tree_cfg.max_osd_combinations);
-            }
-            if (!await this.save_new_pgs(prev_pgs, optimize_result.int_pgs, pg_history, tree_hash))
-            {
-                console.log('Someone changed PG configuration while we also tried to change it. Retrying in '+this.config.mon_change_timeout+' ms');
-                this.schedule_recheck();
-                return;
-            }
-            console.log('PG configuration successfully changed');
-            if (old_pg_count != optimize_result.int_pgs.length)
-            {
-                console.log(`PG count changed from: ${old_pg_count} to ${optimize_result.int_pgs.length}`);
-            }
-            LPOptimizer.print_change_stats(optimize_result);
-        }
-    }
-
-    schedule_recheck()
-    {
-        if (this.recheck_timer)
-        {
-            clearTimeout(this.recheck_timer);
-            this.recheck_timer = null;
-        }
-        this.recheck_timer = setTimeout(() =>
-        {
-            this.recheck_timer = null;
-            this.recheck_pgs().catch(console.error);
-        }, this.config.mon_change_timeout || 1000);
-    }
-
-    sum_stats()
-    {
-        let overflow = false;
-        this.prev_stats = this.prev_stats || { op_stats: {}, subop_stats: {}, recovery_stats: {} };
-        const op_stats = {}, subop_stats = {}, recovery_stats = {};
-        for (const osd in this.state.osd.stats)
-        {
-            const st = this.state.osd.stats[osd];
-            for (const op in st.op_stats||{})
-            {
-                op_stats[op] = op_stats[op] || { count: 0n, usec: 0n, bytes: 0n };
-                op_stats[op].count += BigInt(st.op_stats.count||0);
-                op_stats[op].usec += BigInt(st.op_stats.usec||0);
-                op_stats[op].bytes += BigInt(st.op_stats.bytes||0);
-            }
-            for (const op in st.subop_stats||{})
-            {
-                subop_stats[op] = subop_stats[op] || { count: 0n, usec: 0n };
-                subop_stats[op].count += BigInt(st.subop_stats.count||0);
-                subop_stats[op].usec += BigInt(st.subop_stats.usec||0);
-            }
-            for (const op in st.recovery_stats||{})
-            {
-                recovery_stats[op] = recovery_stats[op] || { count: 0n, bytes: 0n };
-                recovery_stats[op].count += BigInt(st.recovery_stats.count||0);
-                recovery_stats[op].bytes += BigInt(st.recovery_stats.bytes||0);
-            }
-        }
-        for (const op in op_stats)
-        {
-            if (op_stats[op].count >= 0x10000000000000000n)
-            {
-                if (!this.prev_stats.op_stats[op])
-                {
-                    overflow = true;
-                }
-                else
-                {
-                    op_stats[op].count -= this.prev_stats.op_stats[op].count;
-                    op_stats[op].usec -= this.prev_stats.op_stats[op].usec;
-                    op_stats[op].bytes -= this.prev_stats.op_stats[op].bytes;
-                }
-            }
-        }
-        for (const op in subop_stats)
-        {
-            if (subop_stats[op].count >= 0x10000000000000000n)
-            {
-                if (!this.prev_stats.subop_stats[op])
-                {
-                    overflow = true;
-                }
-                else
-                {
-                    subop_stats[op].count -= this.prev_stats.subop_stats[op].count;
-                    subop_stats[op].usec -= this.prev_stats.subop_stats[op].usec;
-                }
-            }
-        }
-        for (const op in recovery_stats)
-        {
-            if (recovery_stats[op].count >= 0x10000000000000000n)
-            {
-                if (!this.prev_stats.recovery_stats[op])
-                {
-                    overflow = true;
-                }
-                else
-                {
-                    recovery_stats[op].count -= this.prev_stats.recovery_stats[op].count;
-                    recovery_stats[op].bytes -= this.prev_stats.recovery_stats[op].bytes;
-                }
-            }
-        }
-        const object_counts = { object: 0n, clean: 0n, misplaced: 0n, degraded: 0n, incomplete: 0n };
-        for (const pg_num in this.state.pg.stats)
-        {
-            const st = this.state.pg.stats[pg_num];
-            for (const k in object_counts)
-            {
-                if (st[k+'_count'])
-                {
-                    object_counts[k] += BigInt(st[k+'_count']);
-                }
-            }
-        }
-        return (this.prev_stats = { overflow, op_stats, subop_stats, recovery_stats, object_counts });
-    }
-
-    async update_total_stats()
-    {
-        const stats = this.sum_stats();
-        if (!stats.overflow)
-        {
-            // Convert to strings, serialize and save
-            const ser = {};
-            for (const st of [ 'op_stats', 'subop_stats', 'recovery_stats' ])
-            {
-                ser[st] = {};
-                for (const op in stats[st])
-                {
-                    ser[st][op] = {};
-                    for (const k in stats[st][op])
-                    {
-                        ser[st][op][k] = ''+stats[st][op][k];
-                    }
-                }
-            }
-            ser.object_counts = {};
-            for (const k in stats.object_counts)
-            {
-                ser.object_counts[k] = ''+stats.object_counts[k];
-            }
-            await this.etcd_call('/txn', {
-                success: [ { requestPut: { key: b64(this.etcd_prefix+'/stats'), value: b64(JSON.stringify(ser)) } } ],
-            }, this.config.etcd_mon_timeout, 0);
-        }
-    }
-
-    schedule_update_stats()
-    {
-        if (this.stats_timer)
-        {
-            clearTimeout(this.stats_timer);
-            this.stats_timer = null;
-        }
-        this.stats_timer = setTimeout(() =>
-        {
-            this.stats_timer = null;
-            this.update_total_stats().catch(console.error);
-        }, this.config.mon_stats_timeout || 1000);
-    }
-
-    parse_kv(kv)
-    {
-        if (!kv || !kv.key)
-        {
-            return;
-        }
-        kv.key = de64(kv.key);
-        kv.value = kv.value ? JSON.parse(de64(kv.value)) : null;
-        const key = kv.key.substr(this.etcd_prefix.length).replace(/^\/+/, '').split('/');
-        const cur = this.state, orig = Mon.etcd_tree;
-        for (let i = 0; i < key.length-1; i++)
-        {
-            if (!orig[key[i]])
-            {
-                console.log('Bad key in etcd: '+kv.key+' = '+kv.value);
-                return;
-            }
-            orig = orig[key[i]];
-            cur = (cur[key[i]] = cur[key[i]] || {});
-        }
-        if (orig[key.length-1])
-        {
-            console.log('Bad key in etcd: '+kv.key+' = '+kv.value);
-            return;
-        }
-        cur[key[key.length-1]] = kv.value;
-        if (key.join('/') === 'config/global')
-        {
-            this.state.config.global = this.state.config.global || {};
-            this.config = this.state.config.global;
-            this.check_config();
-        }
-    }
-
-    async etcd_call(path, body, timeout, retries)
-    {
-        let retry = 0;
-        if (retries >= 0 && retries < 1)
-        {
-            retries = 1;
-        }
-        while (retries < 0 || retry < retries)
-        {
-            const base = this.etcd_urls[Math.floor(Math.random()*this.etcd_urls.length)];
-            const res = await POST(base+path, body, timeout);
-            if (res.json)
-            {
-                if (res.json.error)
-                {
-                    console.log('etcd returned error: '+res.json.error);
-                    break;
-                }
-                return res.json;
-            }
-            retry++;
-        }
-        this.die();
-    }
-
-    die(err)
-    {
-        // In fact we can just try to rejoin
-        console.fatal(err || 'Cluster connection failed');
-        process.exit(1);
-    }
-
-    local_ips()
-    {
-        const ips = [];
-        const ifaces = os.networkInterfaces();
-        for (const ifname in ifaces)
-        {
-            for (const iface of ifaces[ifname])
-            {
-                if (iface.family == 'IPv4' && !iface.internal)
-                {
-                    ips.push(iface.address);
-                }
-            }
-        }
-        return ips;
-    }
-}
-
-function POST(url, body, timeout)
-{
-    return new Promise((ok, no) =>
-    {
-        const body_text = Buffer.from(JSON.stringify(body));
-        let timer_id = timeout > 0 ? setTimeout(() =>
-        {
-            if (req)
-                req.abort();
-            req = null;
-            ok({ error: 'timeout' });
-        }, timeout) : null;
-        let req = http.request(url, { method: 'POST', headers: {
-            'Content-Type': 'application/json',
-            'Content-Length': body_text,
-        } }, (res) =>
-        {
-            if (!req)
-            {
-                return;
-            }
-            clearTimeout(timer_id);
-            if (res.statusCode != 200)
-            {
-                ok({ error: res.statusCode, response: res });
-                return;
-            }
-            let res_body = '';
-            res.setEncoding('utf8');
-            res.on('data', chunk => { res_body += chunk });
-            res.on('end', () =>
-            {
-                try
-                {
-                    res_body = JSON.parse(res_body);
-                    ok({ response: res, json: res_body });
-                }
-                catch (e)
-                {
-                    ok({ error: e, response: res, body: res_body });
-                }
-            });
-        });
-        req.write(body_text);
-        req.end();
-    });
-}
-
-function b64(str)
-{
-    return Buffer.from(str).toString('base64');
-}
-
-function de64(str)
-{
-    return Buffer.from(str, 'base64').toString();
-}
-
-function sha1hex(str)
-{
-    const hash = crypto.createHash('sha1');
-    hash.update(str);
-    return hash.digest('hex');
-}

malloc_or_die.h

@@ -0,0 +1,50 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+#pragma once
+
+#include <malloc.h>
+
+inline void* memalign_or_die(size_t alignment, size_t size)
+{
+    void *buf = memalign(alignment, size);
+    if (!buf)
+    {
+        printf("Failed to allocate %lu bytes\n", size);
+        exit(1);
+    }
+    return buf;
+}
+
+inline void* malloc_or_die(size_t size)
+{
+    void *buf = malloc(size);
+    if (!buf)
+    {
+        printf("Failed to allocate %lu bytes\n", size);
+        exit(1);
+    }
+    return buf;
+}
+
+inline void* realloc_or_die(void *ptr, size_t size)
+{
+    void *buf = realloc(ptr, size);
+    if (!buf)
+    {
+        printf("Failed to allocate %lu bytes\n", size);
+        exit(1);
+    }
+    return buf;
+}
+
+inline void* calloc_or_die(size_t nmemb, size_t size)
+{
+    void *buf = calloc(nmemb, size);
+    if (!buf)
+    {
+        printf("Failed to allocate %lu bytes\n", size * nmemb);
+        exit(1);
+    }
+    return buf;
+}

messenger.cpp

@@ -0,0 +1,412 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+#include <unistd.h>
+#include <fcntl.h>
+#include <sys/socket.h>
+#include <sys/epoll.h>
+#include <netinet/tcp.h>
+
+#include "messenger.h"
+
+osd_op_t::~osd_op_t()
+{
+    assert(!bs_op);
+    assert(!op_data);
+    if (rmw_buf)
+    {
+        free(rmw_buf);
+    }
+    if (buf)
+    {
+        // Note: reusing osd_op_t WILL currently lead to memory leaks
+        // So we don't reuse it, but free it every time
+        free(buf);
+    }
+}
+
+osd_messenger_t::~osd_messenger_t()
+{
+    while (clients.size() > 0)
+    {
+        stop_client(clients.begin()->first);
+    }
+}
+
+void osd_messenger_t::connect_peer(uint64_t peer_osd, json11::Json peer_state)
+{
+    if (wanted_peers.find(peer_osd) == wanted_peers.end())
+    {
+        wanted_peers[peer_osd] = (osd_wanted_peer_t){
+            .address_list = peer_state["addresses"],
+            .port = (int)peer_state["port"].int64_value(),
+        };
+    }
+    else
+    {
+        wanted_peers[peer_osd].address_list = peer_state["addresses"];
+        wanted_peers[peer_osd].port = (int)peer_state["port"].int64_value();
+    }
+    wanted_peers[peer_osd].address_changed = true;
+    if (!wanted_peers[peer_osd].connecting &&
+        (time(NULL) - wanted_peers[peer_osd].last_connect_attempt) >= peer_connect_interval)
+    {
+        try_connect_peer(peer_osd);
+    }
+}
+
+void osd_messenger_t::try_connect_peer(uint64_t peer_osd)
+{
+    auto wp_it = wanted_peers.find(peer_osd);
+    if (wp_it == wanted_peers.end())
+    {
+        return;
+    }
+    if (osd_peer_fds.find(peer_osd) != osd_peer_fds.end())
+    {
+        wanted_peers.erase(peer_osd);
+        return;
+    }
+    auto & wp = wp_it->second;
+    if (wp.address_index >= wp.address_list.array_items().size())
+    {
+        return;
+    }
+    wp.cur_addr = wp.address_list[wp.address_index].string_value();
+    wp.cur_port = wp.port;
+    wp.connecting = true;
+    try_connect_peer_addr(peer_osd, wp.cur_addr.c_str(), wp.cur_port);
+}
+
+void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer_host, int peer_port)
+{
+    struct sockaddr_in addr;
+    int r;
+    if ((r = inet_pton(AF_INET, peer_host, &addr.sin_addr)) != 1)
+    {
+        on_connect_peer(peer_osd, -EINVAL);
+        return;
+    }
+    addr.sin_family = AF_INET;
+    addr.sin_port = htons(peer_port ? peer_port : 11203);
+    int peer_fd = socket(AF_INET, SOCK_STREAM, 0);
+    if (peer_fd < 0)
+    {
+        on_connect_peer(peer_osd, -errno);
+        return;
+    }
+    fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) | O_NONBLOCK);
+    int timeout_id = -1;
+    if (peer_connect_timeout > 0)
+    {
+        timeout_id = tfd->set_timer(1000*peer_connect_timeout, false, [this, peer_fd](int timer_id)
+        {
+            osd_num_t peer_osd = clients[peer_fd].osd_num;
+            stop_client(peer_fd);
+            on_connect_peer(peer_osd, -EIO);
+            return;
+        });
+    }
+    r = connect(peer_fd, (sockaddr*)&addr, sizeof(addr));
+    if (r < 0 && errno != EINPROGRESS)
+    {
+        close(peer_fd);
+        on_connect_peer(peer_osd, -errno);
+        return;
+    }
+    assert(peer_osd != this->osd_num);
+    clients[peer_fd] = (osd_client_t){
+        .peer_addr = addr,
+        .peer_port = peer_port,
+        .peer_fd = peer_fd,
+        .peer_state = PEER_CONNECTING,
+        .connect_timeout_id = timeout_id,
+        .osd_num = peer_osd,
+        .in_buf = malloc_or_die(receive_buffer_size),
+    };
+    tfd->set_fd_handler(peer_fd, true, [this](int peer_fd, int epoll_events)
+    {
+        // Either OUT (connected) or HUP
+        handle_connect_epoll(peer_fd);
+    });
+}
+
+void osd_messenger_t::handle_connect_epoll(int peer_fd)
+{
+    auto & cl = clients[peer_fd];
+    if (cl.connect_timeout_id >= 0)
+    {
+        tfd->clear_timer(cl.connect_timeout_id);
+        cl.connect_timeout_id = -1;
+    }
+    osd_num_t peer_osd = cl.osd_num;
+    int result = 0;
+    socklen_t result_len = sizeof(result);
+    if (getsockopt(peer_fd, SOL_SOCKET, SO_ERROR, &result, &result_len) < 0)
+    {
+        result = errno;
+    }
+    if (result != 0)
+    {
+        stop_client(peer_fd);
+        on_connect_peer(peer_osd, -result);
+        return;
+    }
+    int one = 1;
+    setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
+    cl.peer_state = PEER_CONNECTED;
+    tfd->set_fd_handler(peer_fd, false, [this](int peer_fd, int epoll_events)
+    {
+        handle_peer_epoll(peer_fd, epoll_events);
+    });
+    // Check OSD number
+    check_peer_config(cl);
+}
+
+void osd_messenger_t::handle_peer_epoll(int peer_fd, int epoll_events)
+{
+    // Mark client as ready (i.e. some data is available)
+    if (epoll_events & EPOLLRDHUP)
+    {
+        // Stop client
+        printf("[OSD %lu] client %d disconnected\n", this->osd_num, peer_fd);
+        stop_client(peer_fd);
+    }
+    else if (epoll_events & EPOLLIN)
+    {
+        // Mark client as ready (i.e. some data is available)
+        auto & cl = clients[peer_fd];
+        cl.read_ready++;
+        if (cl.read_ready == 1)
+        {
+            read_ready_clients.push_back(cl.peer_fd);
+            if (ringloop)
+                ringloop->wakeup();
+            else
+                read_requests();
+        }
+    }
+}
+
+void osd_messenger_t::on_connect_peer(osd_num_t peer_osd, int peer_fd)
+{
+    auto & wp = wanted_peers.at(peer_osd);
+    wp.connecting = false;
+    if (peer_fd < 0)
+    {
+        printf("Failed to connect to peer OSD %lu address %s port %d: %s\n", peer_osd, wp.cur_addr.c_str(), wp.cur_port, strerror(-peer_fd));
+        if (wp.address_changed)
+        {
+            wp.address_changed = false;
+            wp.address_index = 0;
+            try_connect_peer(peer_osd);
+        }
+        else if (wp.address_index < wp.address_list.array_items().size()-1)
+        {
+            // Try other addresses
+            wp.address_index++;
+            try_connect_peer(peer_osd);
+        }
+        else
+        {
+            // Retry again in <peer_connect_interval> seconds
+            wp.last_connect_attempt = time(NULL);
+            wp.address_index = 0;
+            tfd->set_timer(1000*peer_connect_interval, false, [this, peer_osd](int)
+            {
+                try_connect_peer(peer_osd);
+            });
+        }
+        return;
+    }
+    printf("Connected with peer OSD %lu (fd %d)\n", peer_osd, peer_fd);
+    wanted_peers.erase(peer_osd);
+    repeer_pgs(peer_osd);
+}
+
+void osd_messenger_t::check_peer_config(osd_client_t & cl)
+{
+    osd_op_t *op = new osd_op_t();
+    op->op_type = OSD_OP_OUT;
+    op->peer_fd = cl.peer_fd;
+    op->req = {
+        .show_conf = {
+            .header = {
+                .magic = SECONDARY_OSD_OP_MAGIC,
+                .id = this->next_subop_id++,
+                .opcode = OSD_OP_SHOW_CONFIG,
+            },
+        },
+    };
+    op->callback = [this](osd_op_t *op)
+    {
+        osd_client_t & cl = clients[op->peer_fd];
+        std::string json_err;
+        json11::Json config;
+        bool err = false;
+        if (op->reply.hdr.retval < 0)
+        {
+            err = true;
+            printf("Failed to get config from OSD %lu (retval=%ld), disconnecting peer\n", cl.osd_num, op->reply.hdr.retval);
+        }
+        else
+        {
+            config = json11::Json::parse(std::string((char*)op->buf), json_err);
+            if (json_err != "")
+            {
+                err = true;
+                printf("Failed to get config from OSD %lu: bad JSON: %s, disconnecting peer\n", cl.osd_num, json_err.c_str());
+            }
+            else if (config["osd_num"].uint64_value() != cl.osd_num)
+            {
+                err = true;
+                printf("Connected to OSD %lu instead of OSD %lu, peer state is outdated, disconnecting peer\n", config["osd_num"].uint64_value(), cl.osd_num);
+            }
+        }
+        if (err)
+        {
+            osd_num_t osd_num = cl.osd_num;
+            stop_client(op->peer_fd);
+            on_connect_peer(osd_num, -1);
+            delete op;
+            return;
+        }
+        osd_peer_fds[cl.osd_num] = cl.peer_fd;
+        on_connect_peer(cl.osd_num, cl.peer_fd);
+        delete op;
+    };
+    outbox_push(op);
+}
+
+void osd_messenger_t::cancel_osd_ops(osd_client_t & cl)
+{
+    for (auto p: cl.sent_ops)
+    {
+        cancel_op(p.second);
+    }
+    cl.sent_ops.clear();
+    for (auto op: cl.outbox)
+    {
+        cancel_op(op);
+    }
+    cl.outbox.clear();
+    if (cl.write_op)
+    {
+        cancel_op(cl.write_op);
+        cl.write_op = NULL;
+    }
+}
+
+void osd_messenger_t::cancel_op(osd_op_t *op)
+{
+    if (op->op_type == OSD_OP_OUT)
+    {
+        op->reply.hdr.magic = SECONDARY_OSD_REPLY_MAGIC;
+        op->reply.hdr.id = op->req.hdr.id;
+        op->reply.hdr.opcode = op->req.hdr.opcode;
+        op->reply.hdr.retval = -EPIPE;
+        // Copy lambda to be unaffected by `delete op`
+        std::function<void(osd_op_t*)>(op->callback)(op);
+    }
+    else
+    {
+        // This function is only called in stop_client(), so it's fine to destroy the operation
+        delete op;
+    }
+}
+
+void osd_messenger_t::stop_client(int peer_fd)
+{
+    assert(peer_fd != 0);
+    auto it = clients.find(peer_fd);
+    if (it == clients.end())
+    {
+        return;
+    }
+    uint64_t repeer_osd = 0;
+    osd_client_t cl = it->second;
+    if (cl.peer_state == PEER_CONNECTED)
+    {
+        if (cl.osd_num)
+        {
+            // Reload configuration from etcd when the connection is dropped
+            printf("[OSD %lu] Stopping client %d (OSD peer %lu)\n", osd_num, peer_fd, cl.osd_num);
+            repeer_osd = cl.osd_num;
+        }
+        else
+        {
+            printf("[OSD %lu] Stopping client %d (regular client)\n", osd_num, peer_fd);
+        }
+    }
+    clients.erase(it);
+    tfd->set_fd_handler(peer_fd, false, NULL);
+    if (cl.osd_num)
+    {
+        osd_peer_fds.erase(cl.osd_num);
+        // Cancel outbound operations
+        cancel_osd_ops(cl);
+    }
+    if (cl.read_op)
+    {
+        delete cl.read_op;
+        cl.read_op = NULL;
+    }
+    for (auto rit = read_ready_clients.begin(); rit != read_ready_clients.end(); rit++)
+    {
+        if (*rit == peer_fd)
+        {
+            read_ready_clients.erase(rit);
+            break;
+        }
+    }
+    for (auto wit = write_ready_clients.begin(); wit != write_ready_clients.end(); wit++)
+    {
+        if (*wit == peer_fd)
+        {
+            write_ready_clients.erase(wit);
+            break;
+        }
+    }
+    free(cl.in_buf);
+    close(peer_fd);
+    if (repeer_osd)
+    {
+        repeer_pgs(repeer_osd);
+    }
+}
+
+void osd_messenger_t::accept_connections(int listen_fd)
+{
+    // Accept new connections
+    sockaddr_in addr;
+    socklen_t peer_addr_size = sizeof(addr);
+    int peer_fd;
+    while ((peer_fd = accept(listen_fd, (sockaddr*)&addr, &peer_addr_size)) >= 0)
+    {
+        assert(peer_fd != 0);
+        char peer_str[256];
+        printf("[OSD %lu] new client %d: connection from %s port %d\n", this->osd_num, peer_fd,
+            inet_ntop(AF_INET, &addr.sin_addr, peer_str, 256), ntohs(addr.sin_port));
+        fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) | O_NONBLOCK);
+        int one = 1;
+        setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
+        clients[peer_fd] = {
+            .peer_addr = addr,
+            .peer_port = ntohs(addr.sin_port),
+            .peer_fd = peer_fd,
+            .peer_state = PEER_CONNECTED,
+            .in_buf = malloc_or_die(receive_buffer_size),
+        };
+        // Add FD to epoll
+        tfd->set_fd_handler(peer_fd, false, [this](int peer_fd, int epoll_events)
+        {
+            handle_peer_epoll(peer_fd, epoll_events);
+        });
+        // Try to accept next connection
+        peer_addr_size = sizeof(addr);
+    }
+    if (peer_fd == -1 && errno != EAGAIN)
+    {
+        throw std::runtime_error(std::string("accept: ") + strerror(errno));
+    }
+}

messenger.h

@@ -0,0 +1,303 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+#pragma once
+
+#include <sys/types.h>
+#include <stdint.h>
+#include <arpa/inet.h>
+
+#include <set>
+#include <map>
+#include <deque>
+#include <vector>
+
+#include "malloc_or_die.h"
+#include "json11/json11.hpp"
+#include "osd_ops.h"
+#include "timerfd_manager.h"
+#include "ringloop.h"
+
+#define OSD_OP_IN 0
+#define OSD_OP_OUT 1
+
+#define CL_READ_HDR 1
+#define CL_READ_DATA 2
+#define CL_READ_REPLY_DATA 3
+#define CL_WRITE_READY 1
+#define CL_WRITE_REPLY 2
+#define OSD_OP_INLINE_BUF_COUNT 16
+
+#define PEER_CONNECTING 1
+#define PEER_CONNECTED 2
+
+#define DEFAULT_PEER_CONNECT_INTERVAL 5
+#define DEFAULT_PEER_CONNECT_TIMEOUT 5
+
+// Kind of a vector with small-list-optimisation
+struct osd_op_buf_list_t
+{
+    int count = 0, alloc = OSD_OP_INLINE_BUF_COUNT, done = 0;
+    iovec *buf = NULL;
+    iovec inline_buf[OSD_OP_INLINE_BUF_COUNT];
+
+    inline osd_op_buf_list_t()
+    {
+        buf = inline_buf;
+    }
+
+    inline osd_op_buf_list_t(const osd_op_buf_list_t & other)
+    {
+        buf = inline_buf;
+        append(other);
+    }
+
+    inline osd_op_buf_list_t & operator = (const osd_op_buf_list_t & other)
+    {
+        reset();
+        append(other);
+        return *this;
+    }
+
+    inline ~osd_op_buf_list_t()
+    {
+        if (buf && buf != inline_buf)
+        {
+            free(buf);
+        }
+    }
+
+    inline void reset()
+    {
+        count = 0;
+        done = 0;
+    }
+
+    inline iovec* get_iovec()
+    {
+        return buf + done;
+    }
+
+    inline int get_size()
+    {
+        return count - done;
+    }
+
+    inline void append(const osd_op_buf_list_t & other)
+    {
+        if (count+other.count > alloc)
+        {
+            if (buf == inline_buf)
+            {
+                int old = alloc;
+                alloc = (((count+other.count+15)/16)*16);
+                buf = (iovec*)malloc(sizeof(iovec) * alloc);
+                if (!buf)
+                {
+                    printf("Failed to allocate %lu bytes\n", sizeof(iovec) * alloc);
+                    exit(1);
+                }
+                memcpy(buf, inline_buf, sizeof(iovec) * old);
+            }
+            else
+            {
+                alloc = (((count+other.count+15)/16)*16);
+                buf = (iovec*)realloc(buf, sizeof(iovec) * alloc);
+                if (!buf)
+                {
+                    printf("Failed to allocate %lu bytes\n", sizeof(iovec) * alloc);
+                    exit(1);
+                }
+            }
+        }
+        for (int i = 0; i < other.count; i++)
+        {
+            buf[count++] = other.buf[i];
+        }
+    }
+
+    inline void push_back(void *nbuf, size_t len)
+    {
+        if (count >= alloc)
+        {
+            if (buf == inline_buf)
+            {
+                int old = alloc;
+                alloc = ((alloc/16)*16 + 1);
+                buf = (iovec*)malloc(sizeof(iovec) * alloc);
+                if (!buf)
+                {
+                    printf("Failed to allocate %lu bytes\n", sizeof(iovec) * alloc);
+                    exit(1);
+                }
+                memcpy(buf, inline_buf, sizeof(iovec)*old);
+            }
+            else
+            {
+                alloc = alloc < 16 ? 16 : (alloc+16);
+                buf = (iovec*)realloc(buf, sizeof(iovec) * alloc);
+                if (!buf)
+                {
+                    printf("Failed to allocate %lu bytes\n", sizeof(iovec) * alloc);
+                    exit(1);
+                }
+            }
+        }
+        buf[count++] = { .iov_base = nbuf, .iov_len = len };
+    }
+
+    inline void eat(int result)
+    {
+        while (result > 0 && done < count)
+        {
+            iovec & iov = buf[done];
+            if (iov.iov_len <= result)
+            {
+                result -= iov.iov_len;
+                done++;
+            }
+            else
+            {
+                iov.iov_len -= result;
+                iov.iov_base += result;
+                break;
+            }
+        }
+    }
+};
+
+struct blockstore_op_t;
+
+struct osd_primary_op_data_t;
+
+struct osd_op_t
+{
+    timespec tv_begin;
+    uint64_t op_type = OSD_OP_IN;
+    int peer_fd;
+    osd_any_op_t req;
+    osd_any_reply_t reply;
+    blockstore_op_t *bs_op = NULL;
+    void *buf = NULL;
+    void *rmw_buf = NULL;
+    osd_primary_op_data_t* op_data = NULL;
+    std::function<void(osd_op_t*)> callback;
+
+    osd_op_buf_list_t iov;
+
+    ~osd_op_t();
+};
+
+struct osd_client_t
+{
+    sockaddr_in peer_addr;
+    int peer_port;
+    int peer_fd;
+    int peer_state;
+    int connect_timeout_id = -1;
+    osd_num_t osd_num = 0;
+
+    void *in_buf = NULL;
+
+    // Read state
+    int read_ready = 0;
+    osd_op_t *read_op = NULL;
+    iovec read_iov;
+    msghdr read_msg;
+    int read_remaining = 0;
+    int read_state = 0;
+    osd_op_buf_list_t recv_list;
+
+    // Incoming operations
+    std::vector<osd_op_t*> received_ops;
+
+    // Outbound operations
+    std::deque<osd_op_t*> outbox;
+    std::map<int, osd_op_t*> sent_ops;
+
+    // PGs dirtied by this client's primary-writes
+    std::set<pool_pg_num_t> dirty_pgs;
+
+    // Write state
+    osd_op_t *write_op = NULL;
+    msghdr write_msg;
+    int write_state = 0;
+    osd_op_buf_list_t send_list;
+};
+
+struct osd_wanted_peer_t
+{
+    json11::Json address_list;
+    int port;
+    time_t last_connect_attempt;
+    bool connecting, address_changed;
+    int address_index;
+    std::string cur_addr;
+    int cur_port;
+};
+
+struct osd_op_stats_t
+{
+    uint64_t op_stat_sum[OSD_OP_MAX+1] = { 0 };
+    uint64_t op_stat_count[OSD_OP_MAX+1] = { 0 };
+    uint64_t op_stat_bytes[OSD_OP_MAX+1] = { 0 };
+    uint64_t subop_stat_sum[OSD_OP_MAX+1] = { 0 };
+    uint64_t subop_stat_count[OSD_OP_MAX+1] = { 0 };
+};
+
+struct osd_messenger_t
+{
+    timerfd_manager_t *tfd;
+    ring_loop_t *ringloop;
+
+    // osd_num_t is only for logging and asserts
+    osd_num_t osd_num;
+    // FIXME: make receive_buffer_size configurable
+    int receive_buffer_size = 64*1024;
+    int peer_connect_interval = DEFAULT_PEER_CONNECT_INTERVAL;
+    int peer_connect_timeout = DEFAULT_PEER_CONNECT_TIMEOUT;
+    int log_level = 0;
+    bool use_sync_send_recv = false;
+
+    std::map<osd_num_t, osd_wanted_peer_t> wanted_peers;
+    std::map<uint64_t, int> osd_peer_fds;
+    uint64_t next_subop_id = 1;
+
+    std::map<int, osd_client_t> clients;
+    std::vector<int> read_ready_clients;
+    std::vector<int> write_ready_clients;
+    std::vector<std::function<void()>> set_immediate;
+
+    // op statistics
+    osd_op_stats_t stats;
+
+public:
+    void connect_peer(uint64_t osd_num, json11::Json peer_state);
+    void stop_client(int peer_fd);
+    void outbox_push(osd_op_t *cur_op);
+    std::function<void(osd_op_t*)> exec_op;
+    std::function<void(osd_num_t)> repeer_pgs;
+    void handle_peer_epoll(int peer_fd, int epoll_events);
+    void read_requests();
+    void send_replies();
+    void accept_connections(int listen_fd);
+    ~osd_messenger_t();
+
+protected:
+    void try_connect_peer(uint64_t osd_num);
+    void try_connect_peer_addr(osd_num_t peer_osd, const char *peer_host, int peer_port);
+    void handle_connect_epoll(int peer_fd);
+    void on_connect_peer(osd_num_t peer_osd, int peer_fd);
+    void check_peer_config(osd_client_t & cl);
+    void cancel_osd_ops(osd_client_t & cl);
+    void cancel_op(osd_op_t *op);
+
+    bool try_send(osd_client_t & cl);
+    void handle_send(int result, int peer_fd);
+
+    bool handle_read(int result, int peer_fd);
+    bool handle_finished_read(osd_client_t & cl);
+    void handle_op_hdr(osd_client_t *cl);
+    bool handle_reply_hdr(osd_client_t *cl);
+    void handle_reply_ready(osd_op_t *op);
+};

mon/PGUtil.js

@@ -0,0 +1,112 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
+module.exports = {
+    scale_pg_count,
+};
+
+function scale_pg_count(prev_pgs, prev_pg_history, new_pg_history, new_pg_count)
+{
+    const old_pg_count = prev_pgs.length;
+    // Add all possibly intersecting PGs into the history of new PGs
+    if (!(new_pg_count % old_pg_count))
+    {
+        // New PG count is a multiple of the old PG count
+        const mul = (new_pg_count / old_pg_count);
+        for (let i = 0; i < new_pg_count; i++)
+        {
+            const old_i = Math.floor(new_pg_count / mul);
+            new_pg_history[i] = JSON.parse(JSON.stringify(prev_pg_history[1+old_i]));
+        }
+    }
+    else if (!(old_pg_count % new_pg_count))
+    {
+        // Old PG count is a multiple of the new PG count
+        const mul = (old_pg_count / new_pg_count);
+        for (let i = 0; i < new_pg_count; i++)
+        {
+            new_pg_history[i] = {
+                osd_sets: [],
+                all_peers: [],
+                epoch: 0,
+            };
+            for (let j = 0; j < mul; j++)
+            {
+                new_pg_history[i].osd_sets.push(prev_pgs[i*mul]);
+                const hist = prev_pg_history[1+i*mul+j];
+                if (hist && hist.osd_sets && hist.osd_sets.length)
+                {
+                    Array.prototype.push.apply(new_pg_history[i].osd_sets, hist.osd_sets);
+                }
+                if (hist && hist.all_peers && hist.all_peers.length)
+                {
+                    Array.prototype.push.apply(new_pg_history[i].all_peers, hist.all_peers);
+                }
+                if (hist && hist.epoch)
+                {
+                    new_pg_history[i].epoch = new_pg_history[i].epoch < hist.epoch ? hist.epoch : new_pg_history[i].epoch;
+                }
+            }
+        }
+    }
+    else
+    {
+        // Any PG may intersect with any PG after non-multiple PG count change
+        // So, merge ALL PGs history
+        let all_sets = {};
+        let all_peers = {};
+        let max_epoch = 0;
+        for (const pg of prev_pgs)
+        {
+            all_sets[pg.join(' ')] = pg;
+        }
+        for (const pg in prev_pg_history)
+        {
+            const hist = prev_pg_history[pg];
+            if (hist && hist.osd_sets)
+            {
+                for (const pg of hist.osd_sets)
+                {
+                    all_sets[pg.join(' ')] = pg;
+                }
+            }
+            if (hist && hist.all_peers)
+            {
+                for (const osd_num of hist.all_peers)
+                {
+                    all_peers[osd_num] = Number(osd_num);
+                }
+            }
+            if (hist && hist.epoch)
+            {
+                max_epoch = max_epoch < hist.epoch ? hist.epoch : max_epoch;
+            }
+        }
+        all_sets = Object.values(all_sets);
+        all_peers = Object.values(all_peers);
+        for (let i = 0; i < new_pg_count; i++)
+        {
+            new_pg_history[i] = { osd_sets: all_sets, all_peers, epoch: max_epoch };
+        }
+    }
+    // Mark history keys for removed PGs as removed
+    for (let i = new_pg_count; i < old_pg_count; i++)
+    {
+        new_pg_history[i] = null;
+    }
+    if (old_pg_count < new_pg_count)
+    {
+        for (let i = new_pg_count-1; i >= 0; i--)
+        {
+            prev_pgs[i] = prev_pgs[Math.floor(i/new_pg_count*old_pg_count)];
+        }
+    }
+    else if (old_pg_count > new_pg_count)
+    {
+        for (let i = 0; i < new_pg_count; i++)
+        {
+            prev_pgs[i] = prev_pgs[Math.round(i/new_pg_count*old_pg_count)];
+        }
+        prev_pgs.splice(new_pg_count, old_pg_count-new_pg_count);
+    }
+}

mon/afr.js

@@ -0,0 +1,159 @@
+// Functions to calculate Annualized Failure Rate of your cluster
+// if you know AFR of your drives, number of drives, expected rebalance time
+// and replication factor
+// License: VNPL-1.0 (see README.md for details)
+
+const { sprintf } = require('sprintf-js');
+
+module.exports = {
+    cluster_afr_fullmesh,
+    failure_rate_fullmesh,
+    cluster_afr,
+    print_cluster_afr,
+    c_n_k,
+};
+
+print_cluster_afr({ n_hosts: 4, n_drives: 6, afr_drive: 0.03, afr_host: 0.05, capacity: 4000, speed: 0.1, replicas: 2 });
+print_cluster_afr({ n_hosts: 4, n_drives: 3, afr_drive: 0.03, capacity: 4000, speed: 0.1, replicas: 2 });
+print_cluster_afr({ n_hosts: 4, n_drives: 3, afr_drive: 0.03, afr_host: 0.05, capacity: 4000, speed: 0.1, replicas: 2 });
+print_cluster_afr({ n_hosts: 4, n_drives: 3, afr_drive: 0.03, capacity: 4000, speed: 0.1, ec: [ 2, 1 ] });
+print_cluster_afr({ n_hosts: 4, n_drives: 3, afr_drive: 0.03, afr_host: 0.05, capacity: 4000, speed: 0.1, ec: [ 2, 1 ] });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, capacity: 8000, speed: 0.02, replicas: 2 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0.05, capacity: 8000, speed: 0.02, replicas: 2 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, capacity: 8000, speed: 0.02, replicas: 3 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0.05, capacity: 8000, speed: 0.02, replicas: 3 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, capacity: 8000, speed: 0.02, replicas: 3, pgs: 100 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0.05, capacity: 8000, speed: 0.02, replicas: 3, pgs: 100 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0.05, capacity: 8000, speed: 0.02, replicas: 3, pgs: 100, degraded_replacement: 1 });
+
+/******** "FULL MESH": ASSUME EACH OSD COMMUNICATES WITH ALL OTHER OSDS ********/
+
+// Estimate AFR of the cluster
+// n - number of drives
+// afr - annualized failure rate of a single drive
+// l - expected rebalance time in days after a single drive failure
+// k - replication factor / number of drives that must fail at the same time for the cluster to fail
+function cluster_afr_fullmesh(n, afr, l, k)
+{
+    return 1 - (1 - afr * failure_rate_fullmesh(n-(k-1), afr*l/365, k-1)) ** (n-(k-1));
+}
+
+// Probability of at least <f> failures in a cluster with <n> drives with AFR=<a>
+function failure_rate_fullmesh(n, a, f)
+{
+    if (f <= 0)
+    {
+        return (1-a)**n;
+    }
+    let p = 1;
+    for (let i = 0; i < f; i++)
+    {
+        p -= c_n_k(n, i) * (1-a)**(n-i) * a**i;
+    }
+    return p;
+}
+
+/******** PGS: EACH OSD ONLY COMMUNICATES WITH <pgs> OTHER OSDs ********/
+
+// <n> hosts of <m> drives of <capacity> GB, each able to backfill at <speed> GB/s,
+// <k> replicas, <pgs> unique peer PGs per OSD
+//
+// For each of n*m drives: P(drive fails in a year) * P(any of its peers fail in <l*365> next days).
+// More peers per OSD increase rebalance speed (more drives work together to resilver) if you
+// let them finish rebalance BEFORE replacing the failed drive.
+// At the same time, more peers per OSD increase probability of any of them to fail!
+//
+// Probability of all except one drives in a replica group to fail is (AFR^(k-1)).
+// So with <x> PGs it becomes ~ (x * (AFR*L/365)^(k-1)). Interesting but reasonable consequence
+// is that, with k=2, total failure rate doesn't depend on number of peers per OSD,
+// because it gets increased linearly by increased number of peers to fail
+// and decreased linearly by reduced rebalance time.
+function cluster_afr_pgs({ n_hosts, n_drives, afr_drive, capacity, speed, replicas, pgs = 1, degraded_replacement })
+{
+    pgs = Math.min(pgs, (n_hosts-1)*n_drives/(replicas-1));
+    const l = capacity/(degraded_replacement ? 1 : pgs)/speed/86400/365;
+    return 1 - (1 - afr_drive * (1-(1-(afr_drive*l)**(replicas-1))**pgs)) ** (n_hosts*n_drives);
+}
+
+function cluster_afr_pgs_ec({ n_hosts, n_drives, afr_drive, capacity, speed, ec: [ ec_data, ec_parity ], pgs = 1, degraded_replacement })
+{
+    const ec_total = ec_data+ec_parity;
+    pgs = Math.min(pgs, (n_hosts-1)*n_drives/(ec_total-1));
+    const l = capacity/(degraded_replacement ? 1 : pgs)/speed/86400/365;
+    return 1 - (1 - afr_drive * (1-(1-failure_rate_fullmesh(ec_total-1, afr_drive*l, ec_parity))**pgs)) ** (n_hosts*n_drives);
+}
+
+// Same as above, but also take server failures into account
+function cluster_afr_pgs_hosts({ n_hosts, n_drives, afr_drive, afr_host, capacity, speed, replicas, pgs = 1, degraded_replacement })
+{
+    let otherhosts = Math.min(pgs, (n_hosts-1)/(replicas-1));
+    pgs = Math.min(pgs, (n_hosts-1)*n_drives/(replicas-1));
+    let pgh = Math.min(pgs*n_drives, (n_hosts-1)*n_drives/(replicas-1));
+    const ld = capacity/(degraded_replacement ? 1 : pgs)/speed/86400/365;
+    const lh = n_drives*capacity/pgs/speed/86400/365;
+    const p1 = ((afr_drive+afr_host*pgs/otherhosts)*lh);
+    const p2 = ((afr_drive+afr_host*pgs/otherhosts)*ld);
+    return 1 - ((1 - afr_host * (1-(1-p1**(replicas-1))**pgh)) ** n_hosts) *
+        ((1 - afr_drive * (1-(1-p2**(replicas-1))**pgs)) ** (n_hosts*n_drives));
+}
+
+function cluster_afr_pgs_ec_hosts({ n_hosts, n_drives, afr_drive, afr_host, capacity, speed, ec: [ ec_data, ec_parity ], pgs = 1, degraded_replacement })
+{
+    const ec_total = ec_data+ec_parity;
+    const otherhosts = Math.min(pgs, (n_hosts-1)/(ec_total-1));
+    pgs = Math.min(pgs, (n_hosts-1)*n_drives/(ec_total-1));
+    const pgh = Math.min(pgs*n_drives, (n_hosts-1)*n_drives/(ec_total-1));
+    const ld = capacity/(degraded_replacement ? 1 : pgs)/speed/86400/365;
+    const lh = n_drives*capacity/pgs/speed/86400/365;
+    const p1 = ((afr_drive+afr_host*pgs/otherhosts)*lh);
+    const p2 = ((afr_drive+afr_host*pgs/otherhosts)*ld);
+    return 1 - ((1 - afr_host * (1-(1-failure_rate_fullmesh(ec_total-1, p1, ec_parity))**pgh)) ** n_hosts) *
+        ((1 - afr_drive * (1-(1-failure_rate_fullmesh(ec_total-1, p2, ec_parity))**pgs)) ** (n_hosts*n_drives));
+}
+
+// Wrapper for 4 above functions
+function cluster_afr(config)
+{
+    if (config.ec && config.afr_host)
+    {
+        return cluster_afr_pgs_ec_hosts(config);
+    }
+    else if (config.ec)
+    {
+        return cluster_afr_pgs_ec(config);
+    }
+    else if (config.afr_host)
+    {
+        return cluster_afr_pgs_hosts(config);
+    }
+    else
+    {
+        return cluster_afr_pgs(config);
+    }
+}
+
+function print_cluster_afr(config)
+{
+    console.log(
+        `${config.n_hosts} nodes with ${config.n_drives} ${sprintf("%.1f", config.capacity/1000)}TB drives`+
+        `, capable to backfill at ${sprintf("%.1f", config.speed*1000)} MB/s, drive AFR ${sprintf("%.1f", config.afr_drive*100)}%`+
+        (config.afr_host ? `, host AFR ${sprintf("%.1f", config.afr_host*100)}%` : '')+
+        (config.ec ? `, EC ${config.ec[0]}+${config.ec[1]}` : `, ${config.replicas} replicas`)+
+        `, ${config.pgs||1} PG per OSD`+
+        (config.degraded_replacement ? `\n...and you don't let the rebalance finish before replacing drives` : '')
+    );
+    console.log('-> '+sprintf("%.7f%%", 100*cluster_afr(config))+'\n');
+}
+
+/******** UTILITY ********/
+
+// Combination count
+function c_n_k(n, k)
+{
+    let r = 1;
+    for (let i = 0; i < k; i++)
+    {
+        r *= (n-i) / (i+1);
+    }
+    return r;
+}

mon/lp-optimizer.js

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 // Data distribution optimizer using linear programming (lp_solve)
 
 const child_process = require('child_process');
@@ -25,7 +28,7 @@ async function lp_solve(text)
     let vars = {};
     for (const line of stdout.split(/\n/))
     {
-        let m = /^(^Value of objective function: ([\d\.]+)|Actual values of the variables:)\s*$/.exec(line);
+        let m = /^(^Value of objective function: (-?[\d\.]+)|Actual values of the variables:)\s*$/.exec(line);
         if (m)
         {
             if (m[2])
@@ -47,34 +50,34 @@ async function lp_solve(text)
     return { score, vars };
 }
 
-async function optimize_initial(osd_tree, pg_count, max_combinations)
+async function optimize_initial({ osd_tree, pg_count, pg_size = 3, pg_minsize = 2, max_combinations = 10000, parity_space = 1 })
 {
-    max_combinations = max_combinations || 10000;
+    if (!pg_count || !osd_tree)
+    {
+        return null;
+    }
     const all_weights = Object.assign({}, ...Object.values(osd_tree));
     const total_weight = Object.values(all_weights).reduce((a, c) => Number(a) + Number(c), 0);
-    let all_pgs = all_combinations(osd_tree, null, true);
+    const all_pgs = Object.values(random_combinations(osd_tree, pg_size, max_combinations));
-    if (all_pgs.length > max_combinations)
-    {
-        const prob = max_combinations/all_pgs.length;
-        all_pgs = all_pgs.filter(pg => Math.random() < prob);
-    }
     const pg_per_osd = {};
     for (const pg of all_pgs)
     {
-        for (const osd of pg)
+        for (let i = 0; i < pg.length; i++)
         {
+            const osd = pg[i];
             pg_per_osd[osd] = pg_per_osd[osd] || [];
-            pg_per_osd[osd].push("pg_"+pg.join("_"));
+            pg_per_osd[osd].push((i >= pg_minsize ? parity_space+'*' : '')+"pg_"+pg.join("_"));
         }
     }
-    const pg_size = Math.min(Object.keys(osd_tree).length, 3);
+    const pg_effsize = Math.min(pg_minsize, Object.keys(osd_tree).length)
+        + Math.max(0, Math.min(pg_size, Object.keys(osd_tree).length) - pg_minsize) * parity_space;
     let lp = '';
     lp += "max: "+all_pgs.map(pg => 'pg_'+pg.join('_')).join(' + ')+";\n";
     for (const osd in pg_per_osd)
     {
         if (osd !== NO_OSD)
         {
-            let osd_pg_count = all_weights[osd]/total_weight*pg_size*pg_count;
+            let osd_pg_count = all_weights[osd]/total_weight*pg_effsize*pg_count;
             lp += pg_per_osd[osd].join(' + ')+' <= '+osd_pg_count+';\n';
         }
     }
@@ -86,11 +89,19 @@ async function optimize_initial(osd_tree, pg_count, max_combinations)
     const lp_result = await lp_solve(lp);
     if (!lp_result)
     {
+        console.log(lp);
         throw new Error('Problem is infeasible or unbounded - is it a bug?');
     }
     const int_pgs = make_int_pgs(lp_result.vars, pg_count);
-    const eff = pg_list_space_efficiency(int_pgs, all_weights);
+    const eff = pg_list_space_efficiency(int_pgs, all_weights, pg_minsize, parity_space);
-    return { score: lp_result.score, weights: lp_result.vars, int_pgs, space: eff*pg_size, total_space: total_weight };
+    const res = {
+        score: lp_result.score,
+        weights: lp_result.vars,
+        int_pgs,
+        space: eff * pg_effsize,
+        total_space: total_weight,
+    };
+    return res;
 }
 
 function make_int_pgs(weights, pg_count)
@@ -112,11 +123,117 @@ function make_int_pgs(weights, pg_count)
     return int_pgs;
 }
 
-// Try to minimize data movement
+function calc_intersect_weights(pg_size, pg_count, prev_weights, all_pgs)
-async function optimize_change(prev_int_pgs, osd_tree, max_combinations)
 {
-    max_combinations = max_combinations || 10000;
+    const move_weights = {};
-    const pg_size = Math.min(Object.keys(osd_tree).length, 3);
+    if ((1 << pg_size) < pg_count)
+    {
+        const intersect = {};
+        for (const pg_name in prev_weights)
+        {
+            const pg = pg_name.substr(3).split(/_/);
+            for (let omit = 1; omit < (1 << pg_size); omit++)
+            {
+                let pg_omit = [ ...pg ];
+                let intersect_count = pg_size;
+                for (let i = 0; i < pg_size; i++)
+                {
+                    if (omit & (1 << i))
+                    {
+                        pg_omit[i] = '';
+                        intersect_count--;
+                    }
+                }
+                pg_omit = pg_omit.join(':');
+                intersect[pg_omit] = Math.max(intersect[pg_omit] || 0, intersect_count);
+            }
+        }
+        for (const pg of all_pgs)
+        {
+            let max_int = 0;
+            for (let omit = 1; omit < (1 << pg_size); omit++)
+            {
+                let pg_omit = [ ...pg ];
+                for (let i = 0; i < pg_size; i++)
+                {
+                    if (omit & (1 << i))
+                    {
+                        pg_omit[i] = '';
+                    }
+                }
+                pg_omit = pg_omit.join(':');
+                max_int = Math.max(max_int, intersect[pg_omit] || 0);
+            }
+            move_weights['pg_'+pg.join('_')] = pg_size-max_int;
+        }
+    }
+    else
+    {
+        const prev_pg_hashed = Object.keys(prev_weights).map(pg_name => pg_name.substr(3).split(/_/).reduce((a, c) => { a[c] = 1; return a; }, {}));
+        for (const pg of all_pgs)
+        {
+            if (!prev_weights['pg_'+pg.join('_')])
+            {
+                let max_int = 0;
+                for (const prev_hash in prev_pg_hashed)
+                {
+                    const intersect_count = pg.reduce((a, osd) => a + (prev_hash[osd] ? 1 : 0), 0);
+                    if (max_int < intersect_count)
+                    {
+                        max_int = intersect_count;
+                        if (max_int >= pg_size)
+                        {
+                            break;
+                        }
+                    }
+                }
+                move_weights['pg_'+pg.join('_')] = pg_size-max_int;
+            }
+        }
+    }
+    return move_weights;
+}
+
+function add_valid_previous(osd_tree, prev_weights, all_pgs)
+{
+    // Add previous combinations that are still valid
+    const hosts = Object.keys(osd_tree).sort();
+    const host_per_osd = {};
+    for (const host in osd_tree)
+    {
+        for (const osd in osd_tree[host])
+        {
+            host_per_osd[osd] = host;
+        }
+    }
+    skip_pg: for (const pg_name in prev_weights)
+    {
+        const seen_hosts = {};
+        const pg = pg_name.substr(3).split(/_/);
+        for (const osd of pg)
+        {
+            if (!host_per_osd[osd] || seen_hosts[host_per_osd[osd]])
+            {
+                continue skip_pg;
+            }
+            seen_hosts[host_per_osd[osd]] = true;
+        }
+        if (!all_pgs[pg_name])
+        {
+            all_pgs[pg_name] = pg;
+        }
+    }
+}
+
+// Try to minimize data movement
+async function optimize_change({ prev_pgs: prev_int_pgs, osd_tree, pg_size = 3, pg_minsize = 2, max_combinations = 10000, parity_space = 1 })
+{
+    if (!osd_tree)
+    {
+        return null;
+    }
+    const pg_effsize = Math.min(pg_minsize, Object.keys(osd_tree).length)
+        + Math.max(0, Math.min(pg_size, Object.keys(osd_tree).length) - pg_minsize) * parity_space;
     const pg_count = prev_int_pgs.length;
     const prev_weights = {};
     const prev_pg_per_osd = {};
@@ -124,70 +241,50 @@ async function optimize_change(prev_int_pgs, osd_tree, max_combinations)
     {
         const pg_name = 'pg_'+pg.join('_');
         prev_weights[pg_name] = (prev_weights[pg_name]||0) + 1;
-        for (const osd of pg)
+        for (let i = 0; i < pg.length; i++)
         {
+            const osd = pg[i];
             prev_pg_per_osd[osd] = prev_pg_per_osd[osd] || [];
-            prev_pg_per_osd[osd].push(pg_name);
+            prev_pg_per_osd[osd].push([ pg_name, (i >= pg_minsize ? parity_space : 1) ]);
         }
     }
     // Get all combinations
-    let all_pgs = all_combinations(osd_tree, null, true);
+    let all_pgs = random_combinations(osd_tree, pg_size, max_combinations);
-    if (all_pgs.length > max_combinations)
+    add_valid_previous(osd_tree, prev_weights, all_pgs);
-    {
+    all_pgs = Object.values(all_pgs);
-        const intersecting = all_pgs.filter(pg => prev_weights['pg_'+pg.join('_')]);
-        if (intersecting.length > max_combinations)
-        {
-            const prob = max_combinations/intersecting.length;
-            all_pgs = intersecting.filter(pg => Math.random() < prob);
-        }
-        else
-        {
-            const prob = (max_combinations-intersecting.length)/all_pgs.length;
-            all_pgs = all_pgs.filter(pg => Math.random() < prob || prev_weights['pg_'+pg.join('_')]);
-        }
-    }
     const pg_per_osd = {};
     for (const pg of all_pgs)
     {
         const pg_name = 'pg_'+pg.join('_');
-        for (const osd of pg)
+        for (let i = 0; i < pg.length; i++)
         {
+            const osd = pg[i];
             pg_per_osd[osd] = pg_per_osd[osd] || [];
-            pg_per_osd[osd].push(pg_name);
+            pg_per_osd[osd].push([ pg_name, (i >= pg_minsize ? parity_space : 1) ]);
         }
     }
     // Penalize PGs based on their similarity to old PGs
-    const intersect = {};
+    const move_weights = calc_intersect_weights(pg_size, pg_count, prev_weights, all_pgs);
-    for (const pg_name in prev_weights)
-    {
-        const pg = pg_name.substr(3).split(/_/);
-        intersect[pg[0]+'::'] = intersect[':'+pg[1]+':'] = intersect['::'+pg[2]] = 2;
-        intersect[pg[0]+'::'+pg[2]] = intersect[':'+pg[1]+':'+pg[2]] = intersect[pg[0]+':'+pg[1]+':'] = 1;
-    }
-    const move_weights = {};
-    for (const pg of all_pgs)
-    {
-        move_weights['pg_'+pg.join('_')] =
-            intersect[pg[0]+'::'+pg[2]] || intersect[':'+pg[1]+':'+pg[2]] || intersect[pg[0]+':'+pg[1]+':'] ||
-            intersect[pg[0]+'::'] || intersect[':'+pg[1]+':'] || intersect['::'+pg[2]] ||
-            3;
-    }
     // Calculate total weight - old PG weights
     const all_pg_names = all_pgs.map(pg => 'pg_'+pg.join('_'));
+    const all_pgs_hash = all_pg_names.reduce((a, c) => { a[c] = true; return a; }, {});
     const all_weights = Object.assign({}, ...Object.values(osd_tree));
     const total_weight = Object.values(all_weights).reduce((a, c) => Number(a) + Number(c), 0);
     // Generate the LP problem
     let lp = '';
     lp += 'max: '+all_pg_names.map(pg_name => (
-        prev_weights[pg_name] ? `${4-move_weights[pg_name]}*add_${pg_name} - 4*del_${pg_name}` : `${4-move_weights[pg_name]}*${pg_name}`
+        prev_weights[pg_name] ? `${pg_size+1}*add_${pg_name} - ${pg_size+1}*del_${pg_name}` : `${pg_size+1-move_weights[pg_name]}*${pg_name}`
     )).join(' + ')+';\n';
     for (const osd in pg_per_osd)
     {
         if (osd !== NO_OSD)
         {
-            const osd_sum = (pg_per_osd[osd]||[]).map(pg_name => prev_weights[pg_name] ? `add_${pg_name} - del_${pg_name}` : pg_name).join(' + ');
+            const osd_sum = (pg_per_osd[osd]||[]).map(([ pg_name, space ]) => (
-            const rm_osd_pg_count = (prev_pg_per_osd[osd]||[]).filter(old_pg_name => move_weights[old_pg_name]).length;
+                prev_weights[pg_name] ? `${space} * add_${pg_name} - ${space} * del_${pg_name}` : `${space} * ${pg_name}`
-            let osd_pg_count = all_weights[osd]*3/total_weight*pg_count - rm_osd_pg_count;
+            )).join(' + ');
+            const rm_osd_pg_count = (prev_pg_per_osd[osd]||[])
+                .reduce((a, [ old_pg_name, space ]) => (a + (all_pgs_hash[old_pg_name] ? space : 0)), 0);
+            const osd_pg_count = all_weights[osd]*pg_effsize/total_weight*pg_count - rm_osd_pg_count;
             lp += osd_sum + ' <= ' + osd_pg_count + ';\n';
         }
     }
@@ -221,7 +318,7 @@ async function optimize_change(prev_int_pgs, osd_tree, max_combinations)
     const weights = { ...prev_weights };
     for (const k in prev_weights)
     {
-        if (!move_weights[k])
+        if (!all_pgs_hash[k])
         {
             delete weights[k];
         }
@@ -236,7 +333,7 @@ async function optimize_change(prev_int_pgs, osd_tree, max_combinations)
         {
             weights[k.substr(4)] = (weights[k.substr(4)] || 0) - Number(lp_result.vars[k]);
         }
-        else
+        else if (k.substr(0, 3) === 'pg_')
         {
             weights[k] = Number(lp_result.vars[k]);
         }
@@ -258,7 +355,7 @@ async function optimize_change(prev_int_pgs, osd_tree, max_combinations)
         {
             differs++;
         }
-        for (let j = 0; j < 3; j++)
+        for (let j = 0; j < pg_size; j++)
         {
             if (new_pgs[i][j] != prev_int_pgs[i][j])
             {
@@ -273,7 +370,7 @@ async function optimize_change(prev_int_pgs, osd_tree, max_combinations)
         int_pgs: new_pgs,
         differs,
         osd_differs,
-        space: pg_size * pg_list_space_efficiency(new_pgs, all_weights),
+        space: pg_effsize * pg_list_space_efficiency(new_pgs, all_weights, pg_minsize, parity_space),
         total_space: total_weight,
     };
 }
@@ -391,64 +488,132 @@ function extract_osds(osd_tree, levels, osd_level, osds = {})
     return osds;
 }
 
-// FIXME: support different pg_sizes, not just 3
+function random_combinations(osd_tree, pg_size, count)
+{
+    let seed = 0x5f020e43;
+    let rng = () =>
+    {
+        seed ^= seed << 13;
+        seed ^= seed >> 17;
+        seed ^= seed << 5;
+        return seed + 2147483648;
+    };
+    const hosts = Object.keys(osd_tree).sort();
+    const osds = Object.keys(osd_tree).reduce((a, c) => { a[c] = Object.keys(osd_tree[c]).sort(); return a; }, {});
+    const r = {};
+    // Generate random combinations including each OSD at least once
+    for (let h = 0; h < hosts.length; h++)
+    {
+        for (let o = 0; o < osds[hosts[h]].length; o++)
+        {
+            const pg = [ osds[hosts[h]][o] ];
+            const cur_hosts = [ ...hosts ];
+            cur_hosts.splice(h, 1);
+            for (let i = 1; i < pg_size && i < hosts.length; i++)
+            {
+                const next_host = rng() % cur_hosts.length;
+                const next_osd = rng() % osds[cur_hosts[next_host]].length;
+                pg.push(osds[cur_hosts[next_host]][next_osd]);
+                cur_hosts.splice(next_host, 1);
+            }
+            while (pg.length < pg_size)
+            {
+                pg.push(NO_OSD);
+            }
+            r['pg_'+pg.join('_')] = pg;
+        }
+    }
+    // Generate purely random combinations
+    restart: while (count > 0)
+    {
+        let host_idx = [];
+        for (let i = 0; i < pg_size && i < hosts.length; i++)
+        {
+            let start = i > 0 ? host_idx[i-1]+1 : 0;
+            if (start >= hosts.length)
+            {
+                continue restart;
+            }
+            host_idx[i] = start + rng() % (hosts.length-start);
+        }
+        let pg = host_idx.map(h => osds[hosts[h]][rng() % osds[hosts[h]].length]);
+        while (pg.length < pg_size)
+        {
+            pg.push(NO_OSD);
+        }
+        r['pg_'+pg.join('_')] = pg;
+        count--;
+    }
+    return r;
+}
+
+// Super-stupid algorithm. Given the current OSD tree, generate all possible OSD combinations
 // osd_tree = { failure_domain1: { osd1: size1, ... }, ... }
-function all_combinations(osd_tree, count, ordered)
+// ordered = return combinations without duplicates having different order
+function all_combinations(osd_tree, pg_size, ordered, count)
 {
     const hosts = Object.keys(osd_tree).sort();
     const osds = Object.keys(osd_tree).reduce((a, c) => { a[c] = Object.keys(osd_tree[c]).sort(); return a; }, {});
-    while (hosts.length < 3)
+    while (hosts.length < pg_size)
     {
         osds[NO_OSD] = [ NO_OSD ];
         hosts.push(NO_OSD);
     }
-    let host_idx = [ 0, 1, 2 ];
+    let host_idx = [];
-    let osd_idx = [ 0, 0, 0 ];
+    let osd_idx = [];
+    for (let i = 0; i < pg_size; i++)
+    {
+        host_idx.push(i);
+        osd_idx.push(0);
+    }
     const r = [];
     while (!count || count < 0 || r.length < count)
     {
-        let inc;
+        r.push(host_idx.map((hi, i) => osds[hosts[hi]][osd_idx[i]]));
-        if (host_idx[2] != host_idx[1] && host_idx[2] != host_idx[0] && host_idx[1] != host_idx[0])
+        let inc = pg_size-1;
+        while (inc >= 0)
         {
-            r.push(host_idx.map((hi, i) => osds[hosts[hi]][osd_idx[i]]));
+            osd_idx[inc]++;
-            inc = 2;
+            if (osd_idx[inc] >= osds[hosts[host_idx[inc]]].length)
-            while (inc >= 0)
             {
-                osd_idx[inc]++;
+                osd_idx[inc] = 0;
-                if (osd_idx[inc] >= osds[hosts[host_idx[inc]]].length)
+                inc--;
-                {
+            }
-                    osd_idx[inc] = 0;
+            else
-                    inc--;
+            {
-                }
+                break;
-                else
-                {
-                    break;
-                }
             }
-        }
-        else
-        {
-            inc = -1;
         }
         if (inc < 0)
         {
-            // no osds left in current host combination, select the next one
+            // no osds left in the current host combination, select the next one
-            osd_idx = [ 0, 0, 0 ];
+            inc = pg_size-1;
-            host_idx[2]++;
+            same_again: while (inc >= 0)
-            if (host_idx[2] >= hosts.length)
             {
-                host_idx[1]++;
+                host_idx[inc]++;
-                host_idx[2] = ordered ? host_idx[1]+1 : 0;
+                for (let prev_host = 0; prev_host < inc; prev_host++)
-                if ((ordered ? host_idx[2] : host_idx[1]) >= hosts.length)
                 {
-                    host_idx[0]++;
+                    if (host_idx[prev_host] == host_idx[inc])
-                    host_idx[1] = ordered ? host_idx[0]+1 : 0;
-                    host_idx[2] = ordered ? host_idx[1]+1 : 0;
-                    if ((ordered ? host_idx[2] : host_idx[0]) >= hosts.length)
                     {
-                        break;
+                        continue same_again;
                     }
                 }
+                if (host_idx[inc] < (ordered ? hosts.length-(pg_size-1-inc) : hosts.length))
+                {
+                    while ((++inc) < pg_size)
+                    {
+                        host_idx[inc] = (ordered ? host_idx[inc-1]+1 : 0);
+                    }
+                    break;
+                }
+                else
+                {
+                    inc--;
+                }
+            }
+            if (inc < 0)
+            {
+                break;
             }
         }
     }
@@ -468,14 +633,15 @@ function pg_weights_space_efficiency(weights, pg_count, osd_sizes)
     return pg_per_osd_space_efficiency(per_osd, pg_count, osd_sizes);
 }
 
-function pg_list_space_efficiency(pgs, osd_sizes)
+function pg_list_space_efficiency(pgs, osd_sizes, pg_minsize, parity_space)
 {
     const per_osd = {};
     for (const pg of pgs)
     {
-        for (const osd of pg)
+        for (let i = 0; i < pg.length; i++)
         {
-            per_osd[osd] = (per_osd[osd]||0) + 1;
+            const osd = pg[i];
+            per_osd[osd] = (per_osd[osd]||0) + (i >= pg_minsize ? (parity_space||1) : 1);
         }
     }
     return pg_per_osd_space_efficiency(per_osd, pgs.length, osd_sizes);
@@ -517,5 +683,6 @@ module.exports = {
     lp_solve,
     make_int_pgs,
     align_pgs,
+    random_combinations,
     all_combinations,
 };

mon/make-units.sh

@@ -0,0 +1,136 @@
+#!/bin/bash
+# Example startup script generator
+# Of course this isn't a production solution yet, this is just for tests
+# Copyright (c) Vitaliy Filippov, 2019+
+# License: MIT
+
+IP=`ip -json a s | jq -r '.[].addr_info[] | select(.broadcast == "10.115.0.255") | .local'`
+
+[ "$IP" != "" ] || exit 1
+
+useradd vitastor
+chmod 755 /root
+
+BASE=${IP/*./}
+BASE=$(((BASE-10)*12))
+
+cat >/etc/systemd/system/vitastor.target <<EOF
+[Unit]
+Description=vitastor target
+[Install]
+WantedBy=multi-user.target
+EOF
+
+i=1
+for DEV in `ls /dev/disk/by-id/ | grep ata-INTEL_SSDSC2KB`; do
+    dd if=/dev/zero of=/dev/disk/by-id/$DEV bs=1048576 count=$(((427814912+1048575)/1048576+2))
+    dd if=/dev/zero of=/dev/disk/by-id/$DEV bs=1048576 count=$(((427814912+1048575)/1048576+2)) seek=$((1920377991168/1048576))
+cat >/etc/systemd/system/vitastor-osd$((BASE+i)).service <<EOF
+[Unit]
+Description=Vitastor object storage daemon osd.$((BASE+i))
+After=network-online.target local-fs.target time-sync.target
+Wants=network-online.target local-fs.target time-sync.target
+PartOf=vitastor.target
+
+[Service]
+LimitNOFILE=1048576
+LimitNPROC=1048576
+LimitMEMLOCK=infinity
+ExecStart=/root/vitastor/osd \\
+    --etcd_address $IP:2379/v3 \\
+    --bind_address $IP \\
+    --osd_num $((BASE+i)) \\
+    --disable_data_fsync 1 \\
+    --disable_device_lock 1 \\
+    --immediate_commit all \\
+    --flusher_count 8 \\
+    --disk_alignment 4096 --journal_block_size 4096 --meta_block_size 4096 \\
+    --journal_no_same_sector_overwrites true \\
+    --journal_sector_buffer_count 1024 \\
+    --journal_offset 0 \\
+    --meta_offset 16777216 \\
+    --data_offset 427814912 \\
+    --data_size $((1920377991168-427814912)) \\
+    --data_device /dev/disk/by-id/$DEV
+WorkingDirectory=/root/vitastor
+ExecStartPre=+chown vitastor:vitastor /dev/disk/by-id/$DEV
+User=vitastor
+PrivateTmp=false
+TasksMax=infinity
+Restart=always
+StartLimitInterval=0
+StartLimitIntervalSec=0
+RestartSec=10
+
+[Install]
+WantedBy=vitastor.target
+EOF
+    systemctl enable vitastor-osd$((BASE+i))
+    i=$((i+1))
+cat >/etc/systemd/system/vitastor-osd$((BASE+i)).service <<EOF
+[Unit]
+Description=Vitastor object storage daemon osd.$((BASE+i))
+After=network-online.target local-fs.target time-sync.target
+Wants=network-online.target local-fs.target time-sync.target
+PartOf=vitastor.target
+
+[Service]
+LimitNOFILE=1048576
+LimitNPROC=1048576
+LimitMEMLOCK=infinity
+ExecStart=/root/vitastor/osd \\
+    --etcd_address $IP:2379/v3 \\
+    --bind_address $IP \\
+    --osd_num $((BASE+i)) \\
+    --disable_data_fsync 1 \\
+    --immediate_commit all \\
+    --flusher_count 8 \\
+    --disk_alignment 4096 --journal_block_size 4096 --meta_block_size 4096 \\
+    --journal_no_same_sector_overwrites true \\
+    --journal_sector_buffer_count 1024 \\
+    --journal_offset 1920377991168 \\
+    --meta_offset $((1920377991168+16777216)) \\
+    --data_offset $((1920377991168+427814912)) \\
+    --data_size $((1920377991168-427814912)) \\
+    --data_device /dev/disk/by-id/$DEV
+WorkingDirectory=/root/vitastor
+ExecStartPre=+chown vitastor:vitastor /dev/disk/by-id/$DEV
+User=vitastor
+PrivateTmp=false
+TasksMax=infinity
+Restart=always
+StartLimitInterval=0
+StartLimitIntervalSec=0
+RestartSec=10
+
+[Install]
+WantedBy=vitastor.target
+EOF
+    systemctl enable vitastor-osd$((BASE+i))
+    i=$((i+1))
+done
+
+exit
+
+node mon-main.js --etcd_url 'http://10.115.0.10:2379,http://10.115.0.11:2379,http://10.115.0.12:2379,http://10.115.0.13:2379' --etcd_prefix '/vitastor' --etcd_start_timeout 5
+
+podman run -d --network host --restart always -v /var/lib/etcd0.etcd:/etcd0.etcd --name etcd quay.io/coreos/etcd:v3.4.13 etcd -name etcd0 \
+    -advertise-client-urls http://10.115.0.10:2379 -listen-client-urls http://10.115.0.10:2379 \
+    -initial-advertise-peer-urls http://10.115.0.10:2380 -listen-peer-urls http://10.115.0.10:2380 \
+    -initial-cluster-token vitastor-etcd-1 -initial-cluster etcd0=http://10.115.0.10:2380,etcd1=http://10.115.0.11:2380,etcd2=http://10.115.0.12:2380,etcd3=http://10.115.0.13:2380 \
+    -initial-cluster-state new --max-txn-ops=100000 --auto-compaction-retention=10 --auto-compaction-mode=revision
+
+etcdctl --endpoints http://10.115.0.10:2379 put /vitastor/config/global '{"immediate_commit":"all"}'
+
+etcdctl --endpoints http://10.115.0.10:2379 put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"replicated","pg_size":2,"pg_minsize":1,"pg_count":48,"failure_domain":"host"}}'
+
+#let pgs = {};
+#for (let n = 0; n < 48; n++) { let i = n/2 | 0; pgs[1+n] = { osd_set: [ (1+i%12+(i/12 | 0)*24), (1+12+i%12+(i/12 | 0)*24) ], primary: (1+(n%2)*12+i%12+(i/12 | 0)*24) }; };
+#console.log(JSON.stringify({ items: { 1: pgs } }));
+#etcdctl --endpoints http://10.115.0.10:2379 put /vitastor/config/pgs ...
+
+#    --disk_alignment 4096 --journal_block_size 4096 --meta_block_size 4096 \\
+#    --data_offset 427814912 \\
+
+#    --disk_alignment 4096 --journal_block_size 512 --meta_block_size 512 \\
+#    --data_offset 433434624 \\

mon/mon-main.js

@@ -1,5 +1,8 @@
 #!/usr/bin/node
 
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 const Mon = require('./mon.js');
 
 const options = {};
@@ -15,8 +18,8 @@ for (let i = 2; i < process.argv.length; i++)
 
 if (!options.etcd_url)
 {
-    console.error('USAGE: '+process.argv[0]+' '+process.argv[1]+' --etcd_url "http://127.0.0.1:2379,..." --etcd_prefix "/rage" --etcd_start_timeout 5');
+    console.error('USAGE: '+process.argv[0]+' '+process.argv[1]+' --etcd_url "http://127.0.0.1:2379,..." --etcd_prefix "/vitastor" --etcd_start_timeout 5 [--verbose 1]');
     process.exit();
 }
 
-new Mon(options).start();
+new Mon(options).start().catch(e => { console.error(e); process.exit(); });

mon/package.json

@@ -1,14 +1,15 @@
 {
-  "name": "rage-mon",
+  "name": "vitastor-mon",
   "version": "1.0.0",
-  "description": "RAGE storage monitor service",
+  "description": "Vitastor SDS monitor service",
-  "main": "mon.js",
+  "main": "mon-main.js",
   "scripts": {
     "test": "echo \"Error: no test specified\" && exit 1"
   },
   "author": "Vitaliy Filippov",
   "license": "UNLICENSED",
   "dependencies": {
+    "sprintf-js": "^1.1.2",
     "ws": "^7.2.5"
   }
 }

mon/simple-offsets.js

@@ -0,0 +1,56 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: MIT
+
+// Simple tool to calculate journal and metadata offsets for a single device
+// Will be replaced by smarter tools in the future
+
+const child_process = require('child_process');
+
+async function run()
+{
+    const options = {
+        object_size: 128*1024,
+        bitmap_granularity: 4096,
+        journal_size: 16*1024*1024,
+        device_block_size: 4096,
+        journal_offset: 0,
+        device_size: 0,
+    };
+    for (let i = 2; i < process.argv.length; i++)
+    {
+        if (process.argv[i].substr(0, 2) == '--')
+        {
+            options[process.argv[i].substr(2)] = process.argv[i+1];
+            i++;
+        }
+    }
+    const device_size = Number(options.device_size || await system("blockdev --getsize64 "+options.device));
+    if (!device_size)
+    {
+        process.stderr.write('Failed to get device size\n');
+        process.exit(1);
+    }
+    options.journal_offset = Math.ceil(options.journal_offset/options.device_block_size)*options.device_block_size;
+    const meta_offset = options.journal_offset + Math.ceil(options.journal_size/options.device_block_size)*options.device_block_size;
+    const entries_per_block = Math.floor(options.device_block_size / (24 + options.object_size/options.bitmap_granularity/8));
+    const object_count = Math.floor((device_size-meta_offset)/options.object_size);
+    const meta_size = Math.ceil(object_count / entries_per_block) * options.device_block_size;
+    const data_offset = meta_offset + meta_size;
+    const meta_size_fmt = (meta_size > 1024*1024*1024 ? Math.round(meta_size/1024/1024/1024*100)/100+" GB"
+        : Math.round(meta_size/1024/1024*100)/100+" MB");
+    process.stdout.write(
+        `Metadata size: ${meta_size_fmt}\n`+
+        `Options for the OSD:\n`+
+        `    --journal_offset ${options.journal_offset}\n`+
+        `    --meta_offset ${meta_offset}\n`+
+        `    --data_offset ${data_offset}\n`+
+        (options.device_size ? `    --data_size ${device_size-data_offset}\n` : '')
+    );
+}
+
+function system(cmd)
+{
+    return new Promise((ok, no) => child_process.exec(cmd, { maxBuffer: 64*1024*1024 }, (err, stdout, stderr) => (err ? no(err) : ok(stdout))));
+}
+
+run().catch(console.error);

mon/stable-stringify.js

@@ -0,0 +1,78 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: MIT
+
+function stableStringify(obj, opts)
+{
+    if (!opts)
+        opts = {};
+    if (typeof opts === 'function')
+        opts = { cmp: opts };
+    let space = opts.space || '';
+    if (typeof space === 'number')
+        space = Array(space+1).join(' ');
+    const cycles = (typeof opts.cycles === 'boolean') ? opts.cycles : false;
+    const cmp = opts.cmp && (function (f)
+    {
+        return function (node)
+        {
+            return function (a, b)
+            {
+                let aobj = { key: a, value: node[a] };
+                let bobj = { key: b, value: node[b] };
+                return f(aobj, bobj);
+            };
+        };
+    })(opts.cmp);
+    const seen = new Map();
+    return (function stringify (parent, key, node, level)
+    {
+        const indent = space ? ('\n' + new Array(level + 1).join(space)) : '';
+        const colonSeparator = space ? ': ' : ':';
+        if (node === undefined)
+        {
+            return;
+        }
+        if (typeof node !== 'object' || node === null)
+        {
+            return JSON.stringify(node);
+        }
+        if (node instanceof Array)
+        {
+            const out = [];
+            for (let i = 0; i < node.length; i++)
+            {
+                const item = stringify(node, i, node[i], level+1) || JSON.stringify(null);
+                out.push(indent + space + item);
+            }
+            return '[' + out.join(',') + indent + ']';
+        }
+        else
+        {
+            if (seen.has(node))
+            {
+                if (cycles)
+                    return JSON.stringify('__cycle__');
+                throw new TypeError('Converting circular structure to JSON');
+            }
+            else
+                seen.set(node, true);
+            const keys = Object.keys(node).sort(cmp && cmp(node));
+            const out = [];
+            for (let i = 0; i < keys.length; i++)
+            {
+                const key = keys[i];
+                const value = stringify(node, key, node[key], level+1);
+                if (!value)
+                    continue;
+                const keyValue = JSON.stringify(key)
+                    + colonSeparator
+                    + value;
+                out.push(indent + space + keyValue);
+            }
+            seen.delete(node);
+            return '{' + out.join(',') + indent + '}';
+        }
+    })({ '': obj }, '', obj, 0);
+}
+
+module.exports = stableStringify;

mon/test-nonuniform.js

@@ -0,0 +1,130 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
+// Interesting real-world example coming from Ceph with EC and compression enabled.
+// EC parity chunks can't be compressed as efficiently as data chunks,
+// thus they occupy more space (2.26x more space) in OSD object stores.
+// This leads to really uneven OSD fill ratio in Ceph even when PGs are perfectly balanced.
+// But we support this case with the "parity_space" parameter in optimize_initial()/optimize_change().
+
+const LPOptimizer = require('./lp-optimizer.js');
+
+const osd_tree = {
+    ripper5: {
+        osd0: 3.493144989013672,
+        osd1: 3.493144989013672,
+        osd2: 3.454082489013672,
+        osd12: 3.461894989013672,
+    },
+    ripper7: {
+        osd4: 3.638690948486328,
+        osd5: 3.638690948486328,
+        osd6: 3.638690948486328,
+    },
+    ripper4: {
+        osd9: 3.4609375,
+        osd10: 3.4609375,
+        osd11: 3.4609375,
+    },
+    ripper6: {
+        osd3: 3.5849609375,
+        osd7: 3.5859336853027344,
+        osd8: 3.638690948486328,
+        osd13: 3.461894989013672
+    },
+};
+
+const prev_pgs = [[12,7,5],[6,11,12],[3,6,9],[10,0,5],[2,5,13],[9,8,6],[3,4,12],[7,4,12],[12,11,13],[13,6,0],[4,13,10],[9,7,6],[7,10,0],[10,8,0],[3,10,2],[3,0,4],[6,13,0],[13,10,0],[13,10,5],[8,11,6],[3,9,2],[2,8,5],[8,9,5],[3,12,11],[0,7,4],[13,11,1],[11,3,12],[12,8,10],[7,5,12],[2,13,5],[7,11,0],[13,2,6],[0,6,8],[13,1,6],[0,13,4],[0,8,10],[4,10,0],[8,12,4],[8,12,9],[12,7,4],[13,9,5],[3,2,11],[1,9,7],[1,8,5],[5,12,9],[3,5,12],[2,8,10],[0,8,4],[1,4,11],[7,10,2],[12,13,5],[3,1,11],[7,1,4],[4,12,8],[7,0,9],[11,1,8],[3,0,5],[11,13,0],[1,13,5],[12,7,10],[12,8,4],[11,13,5],[0,11,6],[2,11,3],[13,1,11],[2,7,10],[7,10,12],[7,12,10],[12,11,5],[13,12,10],[2,3,9],[4,3,9],[13,2,5],[7,12,6],[12,10,13],[9,8,1],[13,1,5],[9,5,12],[5,11,7],[6,2,9],[8,11,6],[12,5,8],[6,13,1],[7,6,11],[2,3,6],[8,5,9],[1,13,6],[9,3,2],[7,11,1],[3,10,1],[0,11,7],[3,0,5],[1,3,6],[6,0,9],[3,11,4],[8,10,2],[13,1,9],[12,6,9],[3,12,9],[12,8,9],[7,5,0],[8,12,5],[0,11,3],[12,11,13],[0,7,11],[0,3,10],[1,3,11],[2,7,11],[13,2,6],[9,12,13],[8,2,4],[0,7,4],[5,13,0],[13,12,9],[1,9,8],[0,10,3],[3,5,10],[7,12,9],[2,13,4],[12,7,5],[9,2,7],[3,2,9],[6,2,7],[3,1,9],[4,3,2],[5,3,11],[0,7,6],[1,6,13],[7,10,2],[12,4,8],[13,12,6],[7,5,11],[6,2,3],[2,7,6],[2,3,10],[2,7,10],[11,12,6],[0,13,5],[10,2,4],[13,0,11],[7,0,6],[8,9,4],[8,4,11],[7,11,2],[3,4,2],[6,1,3],[7,2,11],[8,9,4],[11,4,8],[10,3,1],[2,10,13],[1,7,11],[13,11,12],[2,6,9],[10,0,13],[7,10,4],[0,11,13],[13,10,1],[7,5,0],[7,12,10],[3,1,4],[7,1,5],[3,11,5],[7,5,0],[1,3,5],[10,5,12],[0,3,9],[7,1,11],[11,8,12],[3,6,2],[7,12,9],[7,11,12],[4,11,3],[0,11,13],[13,2,5],[1,5,8],[0,11,8],[3,5,1],[11,0,6],[3,11,2],[11,8,12],[4,1,3],[10,13,4],[13,9,6],[2,3,10],[12,7,9],[10,0,4],[10,13,2],[3,11,1],[7,2,9],[1,7,4],[13,1,4],[7,0,6],[5,3,9],[10,0,7],[0,7,10],[3,6,10],[13,0,5],[8,4,1],[3,1,10],[2,10,13],[13,0,5],[13,10,2],[12,7,9],[6,8,10],[6,1,8],[10,8,1],[13,5,0],[5,11,3],[7,6,1],[8,5,9],[2,13,11],[10,12,4],[13,4,1],[2,13,4],[11,7,0],[2,9,7],[1,7,6],[8,0,4],[8,1,9],[7,10,12],[13,9,6],[7,6,11],[13,0,4],[1,8,4],[3,12,5],[10,3,1],[10,2,13],[2,4,8],[6,2,3],[3,0,10],[6,7,12],[8,12,5],[3,0,6],[13,12,10],[11,3,6],[9,0,13],[10,0,6],[7,5,2],[1,3,11],[7,10,2],[2,9,8],[11,13,12],[0,8,4],[8,12,11],[6,0,3],[1,13,4],[11,8,2],[12,3,6],[4,7,1],[7,6,12],[3,10,6],[0,10,7],[8,9,1],[0,10,6],[8,10,1]]
+    .map(pg => pg.map(n => 'osd'+n));
+
+const by_osd = {};
+
+for (let i = 0; i < prev_pgs.length; i++)
+{
+    for (let j = 0; j < prev_pgs[i].length; j++)
+    {
+        by_osd[prev_pgs[i][j]] = by_osd[prev_pgs[i][j]] || [];
+        by_osd[prev_pgs[i][j]][j] = (by_osd[prev_pgs[i][j]][j] || 0) + 1;
+    }
+}
+
+/*
+
+This set of PGs was balanced by hand, by heavily tuning OSD weights in Ceph:
+
+{
+  osd0: 4.2,
+  osd1: 3.5,
+  osd2: 3.45409,
+  osd3: 4.5,
+  osd4: 1.4,
+  osd5: 1.4,
+  osd6: 1.75,
+  osd7: 4.5,
+  osd8: 4.4,
+  osd9: 2.2,
+  osd10: 2.7,
+  osd11: 2,
+  osd12: 3.4,
+  osd13: 3.4,
+}
+
+EC+compression is a nightmare in Ceph, yeah :))
+
+To calculate the average ratio between data chunks and parity chunks we
+calculate the number of PG chunks for each chunk role for each OSD:
+
+{
+  osd12: [ 18, 22, 17 ],
+  osd7: [ 35, 22, 8 ],
+  osd5: [ 6, 17, 27 ],
+  osd6: [ 13, 12, 28 ],
+  osd11: [ 13, 26, 20 ],
+  osd3: [ 30, 20, 10 ],
+  osd9: [ 8, 12, 26 ],
+  osd10: [ 15, 23, 20 ],
+  osd0: [ 22, 22, 14 ],
+  osd2: [ 22, 16, 16 ],
+  osd13: [ 29, 19, 13 ],
+  osd8: [ 20, 18, 12 ],
+  osd4: [ 8, 10, 28 ],
+  osd1: [ 17, 17, 17 ]
+}
+
+And now we can pick a pair of OSDs and determine the ratio by solving the following:
+
+osd5 = 23*X + 27*Y = 3249728140
+osd13 = 48*X + 13*Y = 2991675992
+
+=>
+
+osd5 - 27/13*osd13 = 23*X - 27/13*48*X = -76.6923076923077*X = -2963752766.46154
+
+=>
+
+X = 38644720.1243731
+Y = (osd5-23*X)/27 = 87440725.0792377
+Y/X = 2.26268232239284 ~= 2.26
+
+Which means that parity chunks are compressed ~2.26 times worse than data chunks.
+
+Fine, let's try to optimize for it.
+
+*/
+
+async function run()
+{
+    const all_weights = Object.assign({}, ...Object.values(osd_tree));
+    const total_weight = Object.values(all_weights).reduce((a, c) => Number(a) + Number(c), 0);
+    const eff = LPOptimizer.pg_list_space_efficiency(prev_pgs, all_weights, 2, 2.26);
+    const orig = eff*4.26 / total_weight;
+    console.log('Original efficiency was: '+Math.round(orig*10000)/100+' %');
+
+    let prev = await LPOptimizer.optimize_initial({ osd_tree, pg_size: 3, pg_count: 256, parity_space: 2.26 });
+    LPOptimizer.print_change_stats(prev);
+
+    let next = await LPOptimizer.optimize_change({ prev_pgs, osd_tree, pg_size: 3, max_combinations: 10000, parity_space: 2.26 });
+    LPOptimizer.print_change_stats(next);
+}
+
+run().catch(console.error);

mon/test-optimize-undersized.js

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 const LPOptimizer = require('./lp-optimizer.js');
 
 const crush_tree = [
@@ -40,31 +43,31 @@ async function run()
 {
     const cur_tree = {};
     console.log('Empty tree:');
-    let res = await LPOptimizer.optimize_initial(cur_tree, 256);
+    let res = await LPOptimizer.optimize_initial({ osd_tree: cur_tree, pg_size: 3, pg_count: 256 });
     LPOptimizer.print_change_stats(res, false);
     console.log('\nAdding 1st failure domain:');
     cur_tree['dom1'] = osd_tree['dom1'];
-    res = await LPOptimizer.optimize_change(res.int_pgs, cur_tree);
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
     LPOptimizer.print_change_stats(res, false);
     console.log('\nAdding 2nd failure domain:');
     cur_tree['dom2'] = osd_tree['dom2'];
-    res = await LPOptimizer.optimize_change(res.int_pgs, cur_tree);
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
     LPOptimizer.print_change_stats(res, false);
     console.log('\nAdding 3rd failure domain:');
     cur_tree['dom3'] = osd_tree['dom3'];
-    res = await LPOptimizer.optimize_change(res.int_pgs, cur_tree);
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
     LPOptimizer.print_change_stats(res, false);
     console.log('\nRemoving 3rd failure domain:');
     delete cur_tree['dom3'];
-    res = await LPOptimizer.optimize_change(res.int_pgs, cur_tree);
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
     LPOptimizer.print_change_stats(res, false);
     console.log('\nRemoving 2nd failure domain:');
     delete cur_tree['dom2'];
-    res = await LPOptimizer.optimize_change(res.int_pgs, cur_tree);
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
     LPOptimizer.print_change_stats(res, false);
     console.log('\nRemoving 1st failure domain:');
     delete cur_tree['dom1'];
-    res = await LPOptimizer.optimize_change(res.int_pgs, cur_tree);
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
     LPOptimizer.print_change_stats(res, false);
 }
 

mon/test-optimize.js

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 const LPOptimizer = require('./lp-optimizer.js');
 
 const osd_tree = {
@@ -75,19 +78,37 @@ const crush_tree = [
 
 async function run()
 {
+    let res;
+
     // Test: add 1 OSD of almost the same size. Ideal data movement could be 1/12 = 8.33%. Actual is ~13%
-    // Space efficiency is ~99.5% in both cases.
+    // Space efficiency is ~99% in all cases.
-    let res = await LPOptimizer.optimize_initial(osd_tree, 256);
+
+    console.log('256 PGs, size=2');
+    res = await LPOptimizer.optimize_initial({ osd_tree, pg_size: 2, pg_count: 256 });
     LPOptimizer.print_change_stats(res, false);
-    console.log('adding osd.8');
+    console.log('\nAdding osd.8');
     osd_tree[500][8] = 3.58589;
-    res = await LPOptimizer.optimize_change(res.int_pgs, osd_tree);
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2 });
     LPOptimizer.print_change_stats(res, false);
-    console.log('removing osd.8');
+    console.log('\nRemoving osd.8');
     delete osd_tree[500][8];
-    res = await LPOptimizer.optimize_change(res.int_pgs, osd_tree);
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2 });
     LPOptimizer.print_change_stats(res, false);
-    res = await LPOptimizer.optimize_initial(LPOptimizer.flatten_tree(crush_tree, {}, 1, 3), 256);
+
+    console.log('\n256 PGs, size=3');
+    res = await LPOptimizer.optimize_initial({ osd_tree, pg_size: 3, pg_count: 256 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nAdding osd.8');
+    osd_tree[500][8] = 3.58589;
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree, pg_size: 3 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nRemoving osd.8');
+    delete osd_tree[500][8];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree, pg_size: 3 });
+    LPOptimizer.print_change_stats(res, false);
+
+    console.log('\n256 PGs, size=3, failure domain=rack');
+    res = await LPOptimizer.optimize_initial({ osd_tree: LPOptimizer.flatten_tree(crush_tree, {}, 1, 3), pg_size: 3, pg_count: 256 });
     LPOptimizer.print_change_stats(res, false);
 }
 

msgr_receive.cpp

@@ -0,0 +1,307 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+#include "messenger.h"
+
+void osd_messenger_t::read_requests()
+{
+    for (int i = 0; i < read_ready_clients.size(); i++)
+    {
+        int peer_fd = read_ready_clients[i];
+        auto & cl = clients[peer_fd];
+        if (cl.read_remaining < receive_buffer_size)
+        {
+            cl.read_iov.iov_base = cl.in_buf;
+            cl.read_iov.iov_len = receive_buffer_size;
+            cl.read_msg.msg_iov = &cl.read_iov;
+            cl.read_msg.msg_iovlen = 1;
+        }
+        else
+        {
+            cl.read_iov.iov_base = 0;
+            cl.read_iov.iov_len = cl.read_remaining;
+            cl.read_msg.msg_iov = cl.recv_list.get_iovec();
+            cl.read_msg.msg_iovlen = cl.recv_list.get_size();
+        }
+        if (ringloop && !use_sync_send_recv)
+        {
+            io_uring_sqe* sqe = ringloop->get_sqe();
+            if (!sqe)
+            {
+                read_ready_clients.erase(read_ready_clients.begin(), read_ready_clients.begin() + i);
+                return;
+            }
+            ring_data_t* data = ((ring_data_t*)sqe->user_data);
+            data->callback = [this, peer_fd](ring_data_t *data) { handle_read(data->res, peer_fd); };
+            my_uring_prep_recvmsg(sqe, peer_fd, &cl.read_msg, 0);
+        }
+        else
+        {
+            int result = recvmsg(peer_fd, &cl.read_msg, 0);
+            if (result < 0)
+            {
+                result = -errno;
+            }
+            handle_read(result, peer_fd);
+        }
+    }
+    read_ready_clients.clear();
+}
+
+bool osd_messenger_t::handle_read(int result, int peer_fd)
+{
+    bool ret = false;
+    auto cl_it = clients.find(peer_fd);
+    if (cl_it != clients.end())
+    {
+        auto & cl = cl_it->second;
+        if (result <= 0 && result != -EAGAIN)
+        {
+            // this is a client socket, so don't panic on error. just disconnect it
+            if (result != 0)
+            {
+                printf("Client %d socket read error: %d (%s). Disconnecting client\n", peer_fd, -result, strerror(-result));
+            }
+            stop_client(peer_fd);
+            return false;
+        }
+        if (result == -EAGAIN || result < cl.read_iov.iov_len)
+        {
+            cl.read_ready--;
+            if (cl.read_ready > 0)
+                read_ready_clients.push_back(peer_fd);
+        }
+        else
+        {
+            read_ready_clients.push_back(peer_fd);
+        }
+        if (result > 0)
+        {
+            if (cl.read_iov.iov_base == cl.in_buf)
+            {
+                // Compose operation(s) from the buffer
+                int remain = result;
+                void *curbuf = cl.in_buf;
+                while (remain > 0)
+                {
+                    if (!cl.read_op)
+                    {
+                        cl.read_op = new osd_op_t;
+                        cl.read_op->peer_fd = peer_fd;
+                        cl.read_op->op_type = OSD_OP_IN;
+                        cl.recv_list.push_back(cl.read_op->req.buf, OSD_PACKET_SIZE);
+                        cl.read_remaining = OSD_PACKET_SIZE;
+                        cl.read_state = CL_READ_HDR;
+                    }
+                    while (cl.recv_list.done < cl.recv_list.count && remain > 0)
+                    {
+                        iovec* cur = cl.recv_list.get_iovec();
+                        if (cur->iov_len > remain)
+                        {
+                            memcpy(cur->iov_base, curbuf, remain);
+                            cl.read_remaining -= remain;
+                            cur->iov_len -= remain;
+                            cur->iov_base += remain;
+                            remain = 0;
+                        }
+                        else
+                        {
+                            memcpy(cur->iov_base, curbuf, cur->iov_len);
+                            curbuf += cur->iov_len;
+                            cl.read_remaining -= cur->iov_len;
+                            remain -= cur->iov_len;
+                            cur->iov_len = 0;
+                            cl.recv_list.done++;
+                        }
+                    }
+                    if (cl.recv_list.done >= cl.recv_list.count)
+                    {
+                        if (!handle_finished_read(cl))
+                        {
+                            goto fin;
+                        }
+                    }
+                }
+            }
+            else
+            {
+                // Long data
+                cl.read_remaining -= result;
+                cl.recv_list.eat(result);
+                if (cl.recv_list.done >= cl.recv_list.count)
+                {
+                    handle_finished_read(cl);
+                }
+            }
+            if (result >= cl.read_iov.iov_len)
+            {
+                ret = true;
+            }
+        }
+    }
+fin:
+    for (auto cb: set_immediate)
+    {
+        cb();
+    }
+    set_immediate.clear();
+    return ret;
+}
+
+bool osd_messenger_t::handle_finished_read(osd_client_t & cl)
+{
+    cl.recv_list.reset();
+    if (cl.read_state == CL_READ_HDR)
+    {
+        if (cl.read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
+            return handle_reply_hdr(&cl);
+        else
+            handle_op_hdr(&cl);
+    }
+    else if (cl.read_state == CL_READ_DATA)
+    {
+        // Operation is ready
+        cl.received_ops.push_back(cl.read_op);
+        set_immediate.push_back([this, op = cl.read_op]() { exec_op(op); });
+        cl.read_op = NULL;
+        cl.read_state = 0;
+    }
+    else if (cl.read_state == CL_READ_REPLY_DATA)
+    {
+        // Reply is ready
+        handle_reply_ready(cl.read_op);
+        cl.read_op = NULL;
+        cl.read_state = 0;
+    }
+    else
+    {
+        assert(0);
+    }
+    return true;
+}
+
+void osd_messenger_t::handle_op_hdr(osd_client_t *cl)
+{
+    osd_op_t *cur_op = cl->read_op;
+    if (cur_op->req.hdr.opcode == OSD_OP_SEC_READ)
+    {
+        if (cur_op->req.sec_rw.len > 0)
+            cur_op->buf = memalign_or_die(MEM_ALIGNMENT, cur_op->req.sec_rw.len);
+        cl->read_remaining = 0;
+    }
+    else if (cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
+    {
+        if (cur_op->req.sec_rw.len > 0)
+            cur_op->buf = memalign_or_die(MEM_ALIGNMENT, cur_op->req.sec_rw.len);
+        cl->read_remaining = cur_op->req.sec_rw.len;
+    }
+    else if (cur_op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK)
+    {
+        if (cur_op->req.sec_stab.len > 0)
+            cur_op->buf = memalign_or_die(MEM_ALIGNMENT, cur_op->req.sec_stab.len);
+        cl->read_remaining = cur_op->req.sec_stab.len;
+    }
+    else if (cur_op->req.hdr.opcode == OSD_OP_READ)
+    {
+        cl->read_remaining = 0;
+    }
+    else if (cur_op->req.hdr.opcode == OSD_OP_WRITE)
+    {
+        if (cur_op->req.rw.len > 0)
+            cur_op->buf = memalign_or_die(MEM_ALIGNMENT, cur_op->req.rw.len);
+        cl->read_remaining = cur_op->req.rw.len;
+    }
+    if (cl->read_remaining > 0)
+    {
+        // Read data
+        cl->recv_list.push_back(cur_op->buf, cl->read_remaining);
+        cl->read_state = CL_READ_DATA;
+    }
+    else
+    {
+        // Operation is ready
+        cl->received_ops.push_back(cur_op);
+        set_immediate.push_back([this, cur_op]() { exec_op(cur_op); });
+        cl->read_op = NULL;
+        cl->read_state = 0;
+    }
+}
+
+bool osd_messenger_t::handle_reply_hdr(osd_client_t *cl)
+{
+    auto req_it = cl->sent_ops.find(cl->read_op->req.hdr.id);
+    if (req_it == cl->sent_ops.end())
+    {
+        // Command out of sync. Drop connection
+        printf("Client %d command out of sync: id %lu\n", cl->peer_fd, cl->read_op->req.hdr.id);
+        stop_client(cl->peer_fd);
+        return false;
+    }
+    osd_op_t *op = req_it->second;
+    memcpy(op->reply.buf, cl->read_op->req.buf, OSD_PACKET_SIZE);
+    cl->sent_ops.erase(req_it);
+    if ((op->reply.hdr.opcode == OSD_OP_SEC_READ || op->reply.hdr.opcode == OSD_OP_READ) &&
+        op->reply.hdr.retval > 0)
+    {
+        // Read data. In this case we assume that the buffer is preallocated by the caller (!)
+        assert(op->iov.count > 0);
+        cl->recv_list.append(op->iov);
+        delete cl->read_op;
+        cl->read_op = op;
+        cl->read_state = CL_READ_REPLY_DATA;
+        cl->read_remaining = op->reply.hdr.retval;
+    }
+    else if (op->reply.hdr.opcode == OSD_OP_SEC_LIST && op->reply.hdr.retval > 0)
+    {
+        assert(!op->iov.count);
+        delete cl->read_op;
+        cl->read_op = op;
+        cl->read_state = CL_READ_REPLY_DATA;
+        cl->read_remaining = sizeof(obj_ver_id) * op->reply.hdr.retval;
+        op->buf = memalign_or_die(MEM_ALIGNMENT, cl->read_remaining);
+        cl->recv_list.push_back(op->buf, cl->read_remaining);
+    }
+    else if (op->reply.hdr.opcode == OSD_OP_SHOW_CONFIG && op->reply.hdr.retval > 0)
+    {
+        assert(!op->iov.count);
+        delete cl->read_op;
+        cl->read_op = op;
+        cl->read_state = CL_READ_REPLY_DATA;
+        cl->read_remaining = op->reply.hdr.retval;
+        op->buf = malloc_or_die(op->reply.hdr.retval);
+        cl->recv_list.push_back(op->buf, op->reply.hdr.retval);
+    }
+    else
+    {
+        // It's fine to reuse cl->read_op for the next reply
+        handle_reply_ready(op);
+        cl->recv_list.push_back(cl->read_op->req.buf, OSD_PACKET_SIZE);
+        cl->read_remaining = OSD_PACKET_SIZE;
+        cl->read_state = CL_READ_HDR;
+    }
+    return true;
+}
+
+void osd_messenger_t::handle_reply_ready(osd_op_t *op)
+{
+    // Measure subop latency
+    timespec tv_end;
+    clock_gettime(CLOCK_REALTIME, &tv_end);
+    stats.subop_stat_count[op->req.hdr.opcode]++;
+    if (!stats.subop_stat_count[op->req.hdr.opcode])
+    {
+        stats.subop_stat_count[op->req.hdr.opcode]++;
+        stats.subop_stat_sum[op->req.hdr.opcode] = 0;
+    }
+    stats.subop_stat_sum[op->req.hdr.opcode] += (
+        (tv_end.tv_sec - op->tv_begin.tv_sec)*1000000 +
+        (tv_end.tv_nsec - op->tv_begin.tv_nsec)/1000
+    );
+    set_immediate.push_back([this, op]()
+    {
+        // Copy lambda to be unaffected by `delete op`
+        std::function<void(osd_op_t*)>(op->callback)(op);
+    });
+}

msgr_send.cpp

@@ -0,0 +1,186 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+#include "messenger.h"
+
+void osd_messenger_t::outbox_push(osd_op_t *cur_op)
+{
+    assert(cur_op->peer_fd);
+    auto & cl = clients.at(cur_op->peer_fd);
+    if (cur_op->op_type == OSD_OP_OUT)
+    {
+        clock_gettime(CLOCK_REALTIME, &cur_op->tv_begin);
+    }
+    else
+    {
+        // Check that operation actually belongs to this client
+        bool found = false;
+        for (auto it = cl.received_ops.begin(); it != cl.received_ops.end(); it++)
+        {
+            if (*it == cur_op)
+            {
+                found = true;
+                cl.received_ops.erase(it, it+1);
+                break;
+            }
+        }
+        if (!found)
+        {
+            delete cur_op;
+            return;
+        }
+    }
+    cl.outbox.push_back(cur_op);
+    if (!ringloop)
+    {
+        while (cl.write_op || cl.outbox.size())
+        {
+            try_send(cl);
+        }
+    }
+    else if (cl.write_op || cl.outbox.size() > 1 || !try_send(cl))
+    {
+        if (cl.write_state == 0)
+        {
+            cl.write_state = CL_WRITE_READY;
+            write_ready_clients.push_back(cur_op->peer_fd);
+        }
+        ringloop->wakeup();
+    }
+}
+
+bool osd_messenger_t::try_send(osd_client_t & cl)
+{
+    int peer_fd = cl.peer_fd;
+    if (!cl.write_op)
+    {
+        // pick next command
+        cl.write_op = cl.outbox.front();
+        cl.outbox.pop_front();
+        cl.write_state = CL_WRITE_REPLY;
+        if (cl.write_op->op_type == OSD_OP_IN)
+        {
+            // Measure execution latency
+            timespec tv_end;
+            clock_gettime(CLOCK_REALTIME, &tv_end);
+            stats.op_stat_count[cl.write_op->req.hdr.opcode]++;
+            if (!stats.op_stat_count[cl.write_op->req.hdr.opcode])
+            {
+                stats.op_stat_count[cl.write_op->req.hdr.opcode]++;
+                stats.op_stat_sum[cl.write_op->req.hdr.opcode] = 0;
+                stats.op_stat_bytes[cl.write_op->req.hdr.opcode] = 0;
+            }
+            stats.op_stat_sum[cl.write_op->req.hdr.opcode] += (
+                (tv_end.tv_sec - cl.write_op->tv_begin.tv_sec)*1000000 +
+                (tv_end.tv_nsec - cl.write_op->tv_begin.tv_nsec)/1000
+            );
+            if (cl.write_op->req.hdr.opcode == OSD_OP_READ ||
+                cl.write_op->req.hdr.opcode == OSD_OP_WRITE)
+            {
+                stats.op_stat_bytes[cl.write_op->req.hdr.opcode] += cl.write_op->req.rw.len;
+            }
+            else if (cl.write_op->req.hdr.opcode == OSD_OP_SEC_READ ||
+                cl.write_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+                cl.write_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
+            {
+                stats.op_stat_bytes[cl.write_op->req.hdr.opcode] += cl.write_op->req.sec_rw.len;
+            }
+            cl.send_list.push_back(cl.write_op->reply.buf, OSD_PACKET_SIZE);
+            if (cl.write_op->req.hdr.opcode == OSD_OP_READ ||
+                cl.write_op->req.hdr.opcode == OSD_OP_SEC_READ ||
+                cl.write_op->req.hdr.opcode == OSD_OP_SEC_LIST ||
+                cl.write_op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)
+            {
+                cl.send_list.append(cl.write_op->iov);
+            }
+        }
+        else
+        {
+            cl.send_list.push_back(cl.write_op->req.buf, OSD_PACKET_SIZE);
+            if (cl.write_op->req.hdr.opcode == OSD_OP_WRITE ||
+                cl.write_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+                cl.write_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE ||
+                cl.write_op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
+                cl.write_op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK)
+            {
+                cl.send_list.append(cl.write_op->iov);
+            }
+        }
+    }
+    cl.write_msg.msg_iov = cl.send_list.get_iovec();
+    cl.write_msg.msg_iovlen = cl.send_list.get_size();
+    if (ringloop && !use_sync_send_recv)
+    {
+        io_uring_sqe* sqe = ringloop->get_sqe();
+        if (!sqe)
+        {
+            return false;
+        }
+        ring_data_t* data = ((ring_data_t*)sqe->user_data);
+        data->callback = [this, peer_fd](ring_data_t *data) { handle_send(data->res, peer_fd); };
+        my_uring_prep_sendmsg(sqe, peer_fd, &cl.write_msg, 0);
+    }
+    else
+    {
+        int result = sendmsg(peer_fd, &cl.write_msg, MSG_NOSIGNAL);
+        if (result < 0)
+        {
+            result = -errno;
+        }
+        handle_send(result, peer_fd);
+    }
+    return true;
+}
+
+void osd_messenger_t::send_replies()
+{
+    for (int i = 0; i < write_ready_clients.size(); i++)
+    {
+        int peer_fd = write_ready_clients[i];
+        if (!try_send(clients[peer_fd]))
+        {
+            write_ready_clients.erase(write_ready_clients.begin(), write_ready_clients.begin() + i);
+            return;
+        }
+    }
+    write_ready_clients.clear();
+}
+
+void osd_messenger_t::handle_send(int result, int peer_fd)
+{
+    auto cl_it = clients.find(peer_fd);
+    if (cl_it != clients.end())
+    {
+        auto & cl = cl_it->second;
+        if (result < 0 && result != -EAGAIN)
+        {
+            // this is a client socket, so don't panic. just disconnect it
+            printf("Client %d socket write error: %d (%s). Disconnecting client\n", peer_fd, -result, strerror(-result));
+            stop_client(peer_fd);
+            return;
+        }
+        if (result >= 0)
+        {
+            cl.send_list.eat(result);
+            if (cl.send_list.done >= cl.send_list.count)
+            {
+                // Done
+                cl.send_list.reset();
+                if (cl.write_op->op_type == OSD_OP_IN)
+                {
+                    delete cl.write_op;
+                }
+                else
+                {
+                    cl.sent_ops[cl.write_op->req.hdr.id] = cl.write_op;
+                }
+                cl.write_op = NULL;
+                cl.write_state = cl.outbox.size() > 0 ? CL_WRITE_READY : 0;
+            }
+        }
+        if (cl.write_state != 0)
+        {
+            write_ready_clients.push_back(peer_fd);
+        }
+    }
+}

nbd_proxy.cpp

@@ -0,0 +1,673 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+// Similar to qemu-nbd, but sets timeout and uses io_uring
+
+#include <linux/nbd.h>
+#include <sys/ioctl.h>
+
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <netinet/tcp.h>
+#include <arpa/inet.h>
+#include <sys/un.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <signal.h>
+
+#include "epoll_manager.h"
+#include "cluster_client.h"
+
+const char *exe_name = NULL;
+
+class nbd_proxy
+{
+protected:
+    uint64_t inode = 0;
+
+    ring_loop_t *ringloop = NULL;
+    epoll_manager_t *epmgr = NULL;
+    cluster_client_t *cli = NULL;
+    ring_consumer_t consumer;
+
+    std::vector<iovec> send_list, next_send_list;
+    std::vector<void*> to_free;
+    int nbd_fd = -1;
+    void *recv_buf = NULL;
+    int receive_buffer_size = 9000;
+    nbd_request cur_req;
+    cluster_op_t *cur_op = NULL;
+    void *cur_buf = NULL;
+    int cur_left = 0;
+    int read_state = 0;
+    int read_ready = 0;
+    msghdr read_msg = { 0 }, send_msg = { 0 };
+    iovec read_iov = { 0 };
+
+public:
+    static json11::Json::object parse_args(int narg, const char *args[])
+    {
+        json11::Json::object cfg;
+        int pos = 0;
+        for (int i = 1; i < narg; i++)
+        {
+            if (!strcmp(args[i], "-h") || !strcmp(args[i], "--help"))
+            {
+                help();
+            }
+            else if (args[i][0] == '-' && args[i][1] == '-')
+            {
+                const char *opt = args[i]+2;
+                cfg[opt] = !strcmp(opt, "json") || i == narg-1 ? "1" : args[++i];
+            }
+            else if (pos == 0)
+            {
+                cfg["command"] = args[i];
+                pos++;
+            }
+            else if (pos == 1 && (cfg["command"] == "map" || cfg["command"] == "unmap"))
+            {
+                int n = 0;
+                if (sscanf(args[i], "/dev/nbd%d", &n) > 0)
+                    cfg["dev_num"] = n;
+                else
+                    cfg["dev_num"] = args[i];
+                pos++;
+            }
+        }
+        return cfg;
+    }
+
+    void exec(json11::Json cfg)
+    {
+        if (cfg["command"] == "map")
+        {
+            start(cfg);
+        }
+        else if (cfg["command"] == "unmap")
+        {
+            if (cfg["dev_num"].is_null())
+            {
+                fprintf(stderr, "device name or number is missing\n");
+                exit(1);
+            }
+            unmap(cfg["dev_num"].uint64_value());
+        }
+        else if (cfg["command"] == "list" || cfg["command"] == "list-mapped")
+        {
+            auto mapped = list_mapped();
+            print_mapped(mapped, !cfg["json"].is_null());
+        }
+        else
+        {
+            help();
+        }
+    }
+
+    static void help()
+    {
+        printf(
+            "Vitastor NBD proxy\n"
+            "(c) Vitaliy Filippov, 2020 (VNPL-1.0 or GNU GPL 2.0+)\n\n"
+            "USAGE:\n"
+            "  %s map --etcd_address <etcd_address> --pool <pool> --inode <inode> --size <size in bytes>\n"
+            "  %s unmap /dev/nbd0\n"
+            "  %s list [--json]\n",
+            exe_name, exe_name, exe_name
+        );
+        exit(0);
+    }
+
+    void unmap(int dev_num)
+    {
+        char path[64] = { 0 };
+        sprintf(path, "/dev/nbd%d", dev_num);
+        int r, nbd = open(path, O_RDWR);
+        if (nbd < 0)
+        {
+            perror("open");
+            exit(1);
+        }
+        r = ioctl(nbd, NBD_DISCONNECT);
+        if (r < 0)
+        {
+            perror("NBD_DISCONNECT");
+            exit(1);
+        }
+        close(nbd);
+    }
+
+    void start(json11::Json cfg)
+    {
+        // Check options
+        if (cfg["etcd_address"].string_value() == "")
+        {
+            fprintf(stderr, "etcd_address is missing\n");
+            exit(1);
+        }
+        if (!cfg["size"].uint64_value())
+        {
+            fprintf(stderr, "device size is missing\n");
+            exit(1);
+        }
+        inode = cfg["inode"].uint64_value();
+        uint64_t pool = cfg["pool"].uint64_value();
+        if (pool)
+        {
+            inode = (inode & ((1l << (64-POOL_ID_BITS)) - 1)) | (pool << (64-POOL_ID_BITS));
+        }
+        if (!(inode >> (64-POOL_ID_BITS)))
+        {
+            fprintf(stderr, "pool is missing\n");
+            exit(1);
+        }
+        // Initialize NBD
+        int sockfd[2];
+        if (socketpair(AF_UNIX, SOCK_STREAM, 0, sockfd) < 0)
+        {
+            perror("socketpair");
+            exit(1);
+        }
+        fcntl(sockfd[0], F_SETFL, fcntl(sockfd[0], F_GETFL, 0) | O_NONBLOCK);
+        nbd_fd = sockfd[0];
+        load_module();
+        if (!cfg["dev_num"].is_null())
+        {
+            if (run_nbd(sockfd, cfg["dev_num"].int64_value(), cfg["size"].uint64_value(), NBD_FLAG_SEND_FLUSH, 30) < 0)
+            {
+                perror("run_nbd");
+                exit(1);
+            }
+        }
+        else
+        {
+            // Find an unused device
+            int i = 0;
+            while (true)
+            {
+                int r = run_nbd(sockfd, i, cfg["size"].uint64_value(), NBD_FLAG_SEND_FLUSH, 30);
+                if (r == 0)
+                {
+                    printf("/dev/nbd%d\n", i);
+                    break;
+                }
+                else if (r == -1 && errno == ENOENT)
+                {
+                    fprintf(stderr, "No free NBD devices found\n");
+                    exit(1);
+                }
+                else if (r == -2 && errno == EBUSY)
+                {
+                    i++;
+                }
+                else
+                {
+                    printf("%d %d\n", r, errno);
+                    perror("run_nbd");
+                    exit(1);
+                }
+            }
+        }
+        if (cfg["foreground"].is_null())
+        {
+            daemonize();
+        }
+        // Create client
+        ringloop = new ring_loop_t(512);
+        epmgr = new epoll_manager_t(ringloop);
+        cli = new cluster_client_t(ringloop, epmgr->tfd, cfg);
+        // Initialize read state
+        read_state = CL_READ_HDR;
+        recv_buf = malloc_or_die(receive_buffer_size);
+        cur_buf = &cur_req;
+        cur_left = sizeof(nbd_request);
+        consumer.loop = [this]()
+        {
+            submit_read();
+            submit_send();
+            ringloop->submit();
+        };
+        ringloop->register_consumer(&consumer);
+        // Add FD to epoll
+        epmgr->tfd->set_fd_handler(sockfd[0], false, [this](int peer_fd, int epoll_events)
+        {
+            read_ready++;
+            submit_read();
+        });
+        while (1)
+        {
+            ringloop->loop();
+            ringloop->wait();
+        }
+    }
+
+    void load_module()
+    {
+        if (access("/sys/module/nbd", F_OK))
+        {
+            return;
+        }
+        int r;
+        if ((r = system("modprobe nbd")) != 0)
+        {
+            if (r < 0)
+                perror("Failed to load NBD kernel module");
+            else
+                fprintf(stderr, "Failed to load NBD kernel module\n");
+            exit(1);
+        }
+    }
+
+    void daemonize()
+    {
+        if (fork())
+            exit(0);
+        setsid();
+        if (fork())
+            exit(0);
+        chdir("/");
+        close(0);
+        close(1);
+        close(2);
+        open("/dev/null", O_RDONLY);
+        open("/dev/null", O_WRONLY);
+        open("/dev/null", O_WRONLY);
+    }
+
+    json11::Json::object list_mapped()
+    {
+        const char *self_filename = exe_name;
+        for (int i = 0; exe_name[i] != 0; i++)
+        {
+            if (exe_name[i] == '/')
+                self_filename = exe_name+i+1;
+        }
+        char path[64] = { 0 };
+        json11::Json::object mapped;
+        int dev_num = -1;
+        int pid;
+        while (true)
+        {
+            dev_num++;
+            sprintf(path, "/sys/block/nbd%d", dev_num);
+            if (access(path, F_OK) != 0)
+                break;
+            sprintf(path, "/sys/block/nbd%d/pid", dev_num);
+            std::string pid_str = read_file(path);
+            if (pid_str == "")
+                continue;
+            if (sscanf(pid_str.c_str(), "%d", &pid) < 1)
+            {
+                printf("Failed to read pid from /sys/block/nbd%d/pid\n", dev_num);
+                continue;
+            }
+            sprintf(path, "/proc/%d/cmdline", pid);
+            std::string cmdline = read_file(path);
+            std::vector<const char*> argv;
+            int last = 0;
+            for (int i = 0; i < cmdline.size(); i++)
+            {
+                if (cmdline[i] == 0)
+                {
+                    argv.push_back(cmdline.c_str()+last);
+                    last = i+1;
+                }
+            }
+            if (argv.size() > 0)
+            {
+                const char *pid_filename = argv[0];
+                for (int i = 0; argv[0][i] != 0; i++)
+                {
+                    if (argv[0][i] == '/')
+                        pid_filename = argv[0]+i+1;
+                }
+                if (!strcmp(pid_filename, self_filename))
+                {
+                    json11::Json::object cfg = nbd_proxy::parse_args(argv.size(), argv.data());
+                    if (cfg["command"] == "map")
+                    {
+                        cfg.erase("command");
+                        cfg["pid"] = pid;
+                        mapped["/dev/nbd"+std::to_string(dev_num)] = cfg;
+                    }
+                }
+            }
+        }
+        return mapped;
+    }
+
+    void print_mapped(json11::Json mapped, bool json)
+    {
+        if (json)
+        {
+            printf("%s\n", mapped.dump().c_str());
+        }
+        else
+        {
+            for (auto & dev: mapped.object_items())
+            {
+                printf("%s\n", dev.first.c_str());
+                for (auto & k: dev.second.object_items())
+                {
+                    printf("%s: %s\n", k.first.c_str(), k.second.string_value().c_str());
+                }
+                printf("\n");
+            }
+        }
+    }
+
+    std::string read_file(char *path)
+    {
+        int fd = open(path, O_RDONLY);
+        if (fd < 0)
+        {
+            if (errno == ENOENT)
+                return "";
+            auto err = "open "+std::string(path);
+            perror(err.c_str());
+            exit(1);
+        }
+        std::string r;
+        while (true)
+        {
+            int l = r.size();
+            r.resize(l + 1024);
+            int rd = read(fd, (void*)(r.c_str() + l), 1024);
+            if (rd <= 0)
+            {
+                r.resize(l);
+                break;
+            }
+            r.resize(l + rd);
+        }
+        close(fd);
+        return r;
+    }
+
+protected:
+    int run_nbd(int sockfd[2], int dev_num, uint64_t size, uint64_t flags, unsigned timeout)
+    {
+        // Check handle size
+        assert(sizeof(cur_req.handle) == 8);
+        char path[64] = { 0 };
+        sprintf(path, "/dev/nbd%d", dev_num);
+        int r, nbd = open(path, O_RDWR), qd_fd;
+        if (nbd < 0)
+        {
+            return -1;
+        }
+        r = ioctl(nbd, NBD_SET_SOCK, sockfd[1]);
+        if (r < 0)
+        {
+            goto end_close;
+        }
+        r = ioctl(nbd, NBD_SET_BLKSIZE, 4096);
+        if (r < 0)
+        {
+            goto end_unmap;
+        }
+        r = ioctl(nbd, NBD_SET_SIZE, size);
+        if (r < 0)
+        {
+            goto end_unmap;
+        }
+        ioctl(nbd, NBD_SET_FLAGS, flags);
+        if (timeout >= 0)
+        {
+            r = ioctl(nbd, NBD_SET_TIMEOUT, (unsigned long)timeout);
+            if (r < 0)
+            {
+                goto end_unmap;
+            }
+        }
+        // Configure request size
+        sprintf(path, "/sys/block/nbd%d/queue/max_sectors_kb", dev_num);
+        qd_fd = open(path, O_WRONLY);
+        if (qd_fd < 0)
+        {
+            goto end_unmap;
+        }
+        write(qd_fd, "32768", 5);
+        close(qd_fd);
+        if (!fork())
+        {
+            // Run in child
+            close(sockfd[0]);
+            r = ioctl(nbd, NBD_DO_IT);
+            if (r < 0)
+            {
+                fprintf(stderr, "NBD device terminated with error: %s\n", strerror(errno));
+                kill(getppid(), SIGTERM);
+            }
+            close(sockfd[1]);
+            ioctl(nbd, NBD_CLEAR_QUE);
+            ioctl(nbd, NBD_CLEAR_SOCK);
+            exit(0);
+        }
+        close(sockfd[1]);
+        close(nbd);
+        return 0;
+    end_close:
+        r = errno;
+        close(nbd);
+        errno = r;
+        return -2;
+    end_unmap:
+        r = errno;
+        ioctl(nbd, NBD_CLEAR_SOCK);
+        close(nbd);
+        errno = r;
+        return -3;
+    }
+
+    void submit_send()
+    {
+        if (!send_list.size() || send_msg.msg_iovlen > 0)
+        {
+            return;
+        }
+        io_uring_sqe* sqe = ringloop->get_sqe();
+        if (!sqe)
+        {
+            return;
+        }
+        ring_data_t* data = ((ring_data_t*)sqe->user_data);
+        data->callback = [this](ring_data_t *data) { handle_send(data->res); };
+        send_msg.msg_iov = send_list.data();
+        send_msg.msg_iovlen = send_list.size();
+        my_uring_prep_sendmsg(sqe, nbd_fd, &send_msg, MSG_ZEROCOPY);
+    }
+
+    void handle_send(int result)
+    {
+        send_msg.msg_iovlen = 0;
+        if (result < 0 && result != -EAGAIN)
+        {
+            fprintf(stderr, "Socket disconnected: %s\n", strerror(-result));
+            exit(1);
+        }
+        int to_eat = 0;
+        while (result > 0 && to_eat < send_list.size())
+        {
+            if (result >= send_list[to_eat].iov_len)
+            {
+                free(to_free[to_eat]);
+                result -= send_list[to_eat].iov_len;
+                to_eat++;
+            }
+            else
+            {
+                send_list[to_eat].iov_base += result;
+                send_list[to_eat].iov_len -= result;
+                break;
+            }
+        }
+        if (to_eat > 0)
+        {
+            send_list.erase(send_list.begin(), send_list.begin() + to_eat);
+            to_free.erase(to_free.begin(), to_free.begin() + to_eat);
+        }
+        for (int i = 0; i < next_send_list.size(); i++)
+        {
+            send_list.push_back(next_send_list[i]);
+        }
+        next_send_list.clear();
+        if (send_list.size() > 0)
+        {
+            ringloop->wakeup();
+        }
+    }
+
+    void submit_read()
+    {
+        if (!read_ready || read_msg.msg_iovlen > 0)
+        {
+            return;
+        }
+        io_uring_sqe* sqe = ringloop->get_sqe();
+        if (!sqe)
+        {
+            return;
+        }
+        ring_data_t* data = ((ring_data_t*)sqe->user_data);
+        data->callback = [this](ring_data_t *data) { handle_read(data->res); };
+        if (cur_left < receive_buffer_size)
+        {
+            read_iov.iov_base = recv_buf;
+            read_iov.iov_len = receive_buffer_size;
+        }
+        else
+        {
+            read_iov.iov_base = cur_buf;
+            read_iov.iov_len = cur_left;
+        }
+        read_msg.msg_iov = &read_iov;
+        read_msg.msg_iovlen = 1;
+        my_uring_prep_recvmsg(sqe, nbd_fd, &read_msg, 0);
+    }
+
+    void handle_read(int result)
+    {
+        read_msg.msg_iovlen = 0;
+        if (result < 0 && result != -EAGAIN)
+        {
+            fprintf(stderr, "Socket disconnected: %s\n", strerror(-result));
+            exit(1);
+        }
+        if (result == -EAGAIN || result < read_iov.iov_len)
+        {
+            read_ready--;
+        }
+        if (read_ready > 0)
+        {
+            ringloop->wakeup();
+        }
+        void *b = recv_buf;
+        while (result > 0)
+        {
+            if (read_iov.iov_base == recv_buf)
+            {
+                int inc = result >= cur_left ? cur_left : result;
+                memcpy(cur_buf, b, inc);
+                cur_left -= inc;
+                result -= inc;
+                cur_buf += inc;
+                b += inc;
+            }
+            else
+            {
+                assert(result <= cur_left);
+                cur_left -= result;
+                result = 0;
+            }
+            if (cur_left <= 0)
+            {
+                handle_finished_read();
+            }
+        }
+    }
+
+    void handle_finished_read()
+    {
+        if (read_state == CL_READ_HDR)
+        {
+            int req_type = be32toh(cur_req.type);
+            if (be32toh(cur_req.magic) != NBD_REQUEST_MAGIC ||
+                req_type != NBD_CMD_READ && req_type != NBD_CMD_WRITE && req_type != NBD_CMD_FLUSH)
+            {
+                printf("Unexpected request: magic=%x type=%x, terminating\n", cur_req.magic, req_type);
+                exit(1);
+            }
+            uint64_t handle = *((uint64_t*)cur_req.handle);
+#ifdef DEBUG
+            printf("request %lx +%x %lx\n", be64toh(cur_req.from), be32toh(cur_req.len), handle);
+#endif
+            void *buf = NULL;
+            cluster_op_t *op = new cluster_op_t;
+            if (req_type == NBD_CMD_READ || req_type == NBD_CMD_WRITE)
+            {
+                op->opcode = req_type == NBD_CMD_READ ? OSD_OP_READ : OSD_OP_WRITE;
+                op->inode = inode;
+                op->offset = be64toh(cur_req.from);
+                op->len = be32toh(cur_req.len);
+                buf = malloc_or_die(sizeof(nbd_reply) + op->len);
+                op->iov.push_back(buf + sizeof(nbd_reply), op->len);
+            }
+            else if (req_type == NBD_CMD_FLUSH)
+            {
+                op->opcode = OSD_OP_SYNC;
+                buf = malloc_or_die(sizeof(nbd_reply));
+            }
+            op->callback = [this, buf, handle](cluster_op_t *op)
+            {
+#ifdef DEBUG
+                printf("reply %lx e=%d\n", handle, op->retval);
+#endif
+                nbd_reply *reply = (nbd_reply*)buf;
+                reply->magic = htobe32(NBD_REPLY_MAGIC);
+                memcpy(reply->handle, &handle, 8);
+                reply->error = htobe32(op->retval < 0 ? -op->retval : 0);
+                auto & to_list = send_msg.msg_iovlen > 0 ? next_send_list : send_list;
+                if (op->retval < 0 || op->opcode != OSD_OP_READ)
+                    to_list.push_back({ .iov_base = buf, .iov_len = sizeof(nbd_reply) });
+                else
+                    to_list.push_back({ .iov_base = buf, .iov_len = sizeof(nbd_reply) + op->len });
+                to_free.push_back(buf);
+                delete op;
+                ringloop->wakeup();
+            };
+            if (req_type == NBD_CMD_WRITE)
+            {
+                cur_op = op;
+                cur_buf = buf + sizeof(nbd_reply);
+                cur_left = op->len;
+                read_state = CL_READ_DATA;
+            }
+            else
+            {
+                cur_op = NULL;
+                cur_buf = &cur_req;
+                cur_left = sizeof(nbd_request);
+                read_state = CL_READ_HDR;
+                cli->execute(op);
+            }
+        }
+        else
+        {
+            cli->execute(cur_op);
+            cur_op = NULL;
+            cur_buf = &cur_req;
+            cur_left = sizeof(nbd_request);
+            read_state = CL_READ_HDR;
+        }
+    }
+};
+
+int main(int narg, const char *args[])
+{
+    setvbuf(stdout, NULL, _IONBF, 0);
+    setvbuf(stderr, NULL, _IONBF, 0);
+    exe_name = args[0];
+    nbd_proxy *p = new nbd_proxy();
+    p->exec(nbd_proxy::parse_args(narg, args));
+    return 0;
+}

object_id.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #pragma once
 
 #include <stdint.h>

osd.cpp

@@ -1,5 +1,7 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <sys/socket.h>
-#include <sys/epoll.h>
 #include <sys/poll.h>
 #include <netinet/in.h>
 #include <netinet/tcp.h>
@@ -7,23 +9,6 @@
 
 #include "osd.h"
 
-const char* osd_op_names[] = {
-    "",
-    "read",
-    "write",
-    "sync",
-    "stabilize",
-    "rollback",
-    "delete",
-    "sync_stab_all",
-    "list",
-    "show_config",
-    "primary_read",
-    "primary_write",
-    "primary_sync",
-    "primary_delete",
-};
-
 osd_t::osd_t(blockstore_config_t & config, blockstore_t *bs, ring_loop_t *ringloop)
 {
     this->config = config;
@@ -36,13 +21,9 @@ osd_t::osd_t(blockstore_config_t & config, blockstore_t *bs, ring_loop_t *ringlo
 
     parse_config(config);
 
-    epoll_fd = epoll_create(1);
+    epmgr = new epoll_manager_t(ringloop);
-    if (epoll_fd < 0)
+    this->tfd = epmgr->tfd;
-    {
-        throw std::runtime_error(std::string("epoll_create: ") + strerror(errno));
-    }
 
-    this->tfd = new timerfd_manager_t([this](int fd, std::function<void(int, int)> handler) { set_fd_handler(fd, handler); });
     this->tfd->set_timer(print_stats_interval*1000, true, [this](int timer_id)
     {
         print_stats();
@@ -61,41 +42,16 @@ osd_t::osd_t(blockstore_config_t & config, blockstore_t *bs, ring_loop_t *ringlo
 
 osd_t::~osd_t()
 {
-    if (tfd)
-    {
-        delete tfd;
-        tfd = NULL;
-    }
     ringloop->unregister_consumer(&consumer);
-    close(epoll_fd);
+    delete epmgr;
     close(listen_fd);
 }
 
 void osd_t::parse_config(blockstore_config_t & config)
 {
-    int pos;
     // Initial startup configuration
-    {
+    json11::Json json_config = json11::Json(config);
-        std::string ea = config["etcd_address"];
+    st_cli.parse_config(json_config);
-        while (1)
-        {
-            pos = ea.find(',');
-            std::string addr = pos >= 0 ? ea.substr(0, pos) : ea;
-            if (addr.length() > 0)
-            {
-                if (addr.find('/') < 0)
-                    addr += "/v3";
-                st_cli.etcd_addresses.push_back(addr);
-            }
-            if (pos >= 0)
-                ea = ea.substr(pos+1);
-            else
-                break;
-        }
-    }
-    st_cli.etcd_prefix = config["etcd_prefix"];
-    if (st_cli.etcd_prefix == "")
-        st_cli.etcd_prefix = "/microceph";
     etcd_report_interval = strtoull(config["etcd_report_interval"].c_str(), NULL, 10);
     if (etcd_report_interval <= 0)
         etcd_report_interval = 30;
@@ -127,12 +83,6 @@ void osd_t::parse_config(blockstore_config_t & config)
         if (client_queue_depth < 128)
             client_queue_depth = 128;
     }
-    if (config.find("pg_stripe_size") != config.end())
-    {
-        pg_stripe_size = strtoull(config["pg_stripe_size"].c_str(), NULL, 10);
-        if (!pg_stripe_size || !bs_block_size || pg_stripe_size < bs_block_size || (pg_stripe_size % bs_block_size) != 0)
-            pg_stripe_size = DEFAULT_PG_STRIPE_SIZE;
-    }
     recovery_queue_depth = strtoull(config["recovery_queue_depth"].c_str(), NULL, 10);
     if (recovery_queue_depth < 1 || recovery_queue_depth > MAX_RECOVERY_QUEUE)
         recovery_queue_depth = DEFAULT_RECOVERY_QUEUE;
@@ -143,12 +93,11 @@ void osd_t::parse_config(blockstore_config_t & config)
         print_stats_interval = 3;
     c_cli.peer_connect_interval = strtoull(config["peer_connect_interval"].c_str(), NULL, 10);
     if (!c_cli.peer_connect_interval)
-        c_cli.peer_connect_interval = 5;
+        c_cli.peer_connect_interval = DEFAULT_PEER_CONNECT_INTERVAL;
     c_cli.peer_connect_timeout = strtoull(config["peer_connect_timeout"].c_str(), NULL, 10);
     if (!c_cli.peer_connect_timeout)
-        c_cli.peer_connect_timeout = 5;
+        c_cli.peer_connect_timeout = DEFAULT_PEER_CONNECT_TIMEOUT;
     log_level = strtoull(config["log_level"].c_str(), NULL, 10);
-    st_cli.log_level = log_level;
     c_cli.log_level = log_level;
 }
 
@@ -200,15 +149,10 @@ void osd_t::bind_socket()
 
     fcntl(listen_fd, F_SETFL, fcntl(listen_fd, F_GETFL, 0) | O_NONBLOCK);
 
-    epoll_event ev;
+    epmgr->set_fd_handler(listen_fd, false, [this](int fd, int events)
-    ev.data.fd = listen_fd;
-    ev.events = EPOLLIN | EPOLLET;
-    if (epoll_ctl(epoll_fd, EPOLL_CTL_ADD, listen_fd, &ev) < 0)
     {
-        close(listen_fd);
+        c_cli.accept_connections(listen_fd);
-        close(epoll_fd);
+    });
-        throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
-    }
 }
 
 bool osd_t::shutdown()
@@ -223,112 +167,12 @@ bool osd_t::shutdown()
 
 void osd_t::loop()
 {
-    if (!wait_state)
-    {
-        handle_epoll_events();
-        wait_state = 1;
-    }
     handle_peers();
     c_cli.read_requests();
     c_cli.send_replies();
     ringloop->submit();
 }
 
-void osd_t::set_fd_handler(int fd, std::function<void(int, int)> handler)
-{
-    if (handler != NULL)
-    {
-        bool exists = epoll_handlers.find(fd) != epoll_handlers.end();
-        epoll_event ev;
-        ev.data.fd = fd;
-        ev.events = EPOLLOUT | EPOLLIN | EPOLLRDHUP | EPOLLET;
-        if (epoll_ctl(epoll_fd, exists ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev) < 0)
-        {
-            throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
-        }
-        epoll_handlers[fd] = handler;
-    }
-    else
-    {
-        if (epoll_ctl(epoll_fd, EPOLL_CTL_DEL, fd, NULL) < 0 && errno != ENOENT)
-        {
-            throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
-        }
-        epoll_handlers.erase(fd);
-    }
-}
-
-void osd_t::handle_epoll_events()
-{
-    io_uring_sqe *sqe = ringloop->get_sqe();
-    if (!sqe)
-    {
-        throw std::runtime_error("can't get SQE, will fall out of sync with EPOLLET");
-    }
-    ring_data_t *data = ((ring_data_t*)sqe->user_data);
-    my_uring_prep_poll_add(sqe, epoll_fd, POLLIN);
-    data->callback = [this](ring_data_t *data)
-    {
-        if (data->res < 0)
-        {
-            throw std::runtime_error(std::string("epoll failed: ") + strerror(-data->res));
-        }
-        handle_epoll_events();
-    };
-    ringloop->submit();
-    int nfds;
-    epoll_event events[MAX_EPOLL_EVENTS];
-restart:
-    nfds = epoll_wait(epoll_fd, events, MAX_EPOLL_EVENTS, 0);
-    for (int i = 0; i < nfds; i++)
-    {
-        if (events[i].data.fd == listen_fd)
-        {
-            // Accept new connections
-            sockaddr_in addr;
-            socklen_t peer_addr_size = sizeof(addr);
-            int peer_fd;
-            while ((peer_fd = accept(listen_fd, (sockaddr*)&addr, &peer_addr_size)) >= 0)
-            {
-                assert(peer_fd != 0);
-                char peer_str[256];
-                printf("[OSD %lu] new client %d: connection from %s port %d\n", this->osd_num, peer_fd,
-                    inet_ntop(AF_INET, &addr.sin_addr, peer_str, 256), ntohs(addr.sin_port));
-                fcntl(peer_fd, F_SETFL, fcntl(listen_fd, F_GETFL, 0) | O_NONBLOCK);
-                int one = 1;
-                setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
-                c_cli.clients[peer_fd] = {
-                    .peer_addr = addr,
-                    .peer_port = ntohs(addr.sin_port),
-                    .peer_fd = peer_fd,
-                    .peer_state = PEER_CONNECTED,
-                    .in_buf = malloc(c_cli.receive_buffer_size),
-                };
-                // Add FD to epoll
-                set_fd_handler(peer_fd, [this](int peer_fd, int epoll_events)
-                {
-                    c_cli.handle_peer_epoll(peer_fd, epoll_events);
-                });
-                // Try to accept next connection
-                peer_addr_size = sizeof(addr);
-            }
-            if (peer_fd == -1 && errno != EAGAIN)
-            {
-                throw std::runtime_error(std::string("accept: ") + strerror(errno));
-            }
-        }
-        else
-        {
-            auto & cb = epoll_handlers[events[i].data.fd];
-            cb(events[i].data.fd, events[i].events);
-        }
-    }
-    if (nfds == MAX_EPOLL_EVENTS)
-    {
-        goto restart;
-    }
-}
-
 void osd_t::exec_op(osd_op_t *cur_op)
 {
     clock_gettime(CLOCK_REALTIME, &cur_op->tv_begin);
@@ -339,21 +183,28 @@ void osd_t::exec_op(osd_op_t *cur_op)
         return;
     }
     inflight_ops++;
-    cur_op->send_list.push_back(cur_op->reply.buf, OSD_PACKET_SIZE);
     if (cur_op->req.hdr.magic != SECONDARY_OSD_OP_MAGIC ||
         cur_op->req.hdr.opcode < OSD_OP_MIN || cur_op->req.hdr.opcode > OSD_OP_MAX ||
-        (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_READ || cur_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE) &&
+        ((cur_op->req.hdr.opcode == OSD_OP_SEC_READ ||
-        (cur_op->req.sec_rw.len > OSD_RW_MAX || cur_op->req.sec_rw.len % bs_disk_alignment || cur_op->req.sec_rw.offset % bs_disk_alignment) ||
+            cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
-        (cur_op->req.hdr.opcode == OSD_OP_READ || cur_op->req.hdr.opcode == OSD_OP_WRITE || cur_op->req.hdr.opcode == OSD_OP_DELETE) &&
+            cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE) &&
-        (cur_op->req.rw.len > OSD_RW_MAX || cur_op->req.rw.len % bs_disk_alignment || cur_op->req.rw.offset % bs_disk_alignment))
+            (cur_op->req.sec_rw.len > OSD_RW_MAX ||
+            cur_op->req.sec_rw.len % bs_disk_alignment ||
+            cur_op->req.sec_rw.offset % bs_disk_alignment)) ||
+        ((cur_op->req.hdr.opcode == OSD_OP_READ ||
+            cur_op->req.hdr.opcode == OSD_OP_WRITE ||
+            cur_op->req.hdr.opcode == OSD_OP_DELETE) &&
+            (cur_op->req.rw.len > OSD_RW_MAX ||
+            cur_op->req.rw.len % bs_disk_alignment ||
+            cur_op->req.rw.offset % bs_disk_alignment)))
     {
         // Bad command
         finish_op(cur_op, -EINVAL);
         return;
     }
     if (readonly &&
-        cur_op->req.hdr.opcode != OSD_OP_SECONDARY_READ &&
+        cur_op->req.hdr.opcode != OSD_OP_SEC_READ &&
-        cur_op->req.hdr.opcode != OSD_OP_SECONDARY_LIST &&
+        cur_op->req.hdr.opcode != OSD_OP_SEC_LIST &&
         cur_op->req.hdr.opcode != OSD_OP_READ &&
         cur_op->req.hdr.opcode != OSD_OP_SHOW_CONFIG)
     {

osd.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #pragma once
 
 #include <sys/types.h>
@@ -16,8 +19,9 @@
 #include "blockstore.h"
 #include "ringloop.h"
 #include "timerfd_manager.h"
+#include "epoll_manager.h"
 #include "osd_peering_pg.h"
-#include "cluster_client.h"
+#include "messenger.h"
 #include "etcd_state_client.h"
 
 #define OSD_LOADING_PGS 0x01
@@ -33,12 +37,9 @@
 #define DEFAULT_AUTOSYNC_INTERVAL 5
 #define MAX_RECOVERY_QUEUE 2048
 #define DEFAULT_RECOVERY_QUEUE 4
-#define DEFAULT_PG_STRIPE_SIZE 4*1024*1024 // 4 MB by default
 
 //#define OSD_STUB
 
-extern const char* osd_op_names[];
-
 struct osd_object_id_t
 {
     osd_num_t osd_num;
@@ -49,7 +50,6 @@ struct osd_recovery_op_t
 {
     int st = 0;
     bool degraded = false;
-    pg_num_t pg_num = 0;
     object_id oid = { 0 };
     osd_op_t *osd_op = NULL;
 };
@@ -78,23 +78,24 @@ class osd_t
     // cluster state
 
     etcd_state_client_t st_cli;
-    cluster_client_t c_cli;
+    osd_messenger_t c_cli;
     int etcd_failed_attempts = 0;
     std::string etcd_lease_id;
     json11::Json self_state;
     bool loading_peer_config = false;
-    std::set<pg_num_t> pg_state_dirty;
+    std::set<pool_pg_num_t> pg_state_dirty;
     bool pg_config_applied = false;
     bool etcd_reporting_pg_state = false;
     bool etcd_reporting_stats = false;
 
     // peers and PGs
 
-    std::map<pg_num_t, pg_t> pgs;
+    std::map<pool_id_t, pg_num_t> pg_counts;
-    std::set<pg_num_t> dirty_pgs;
+    std::map<pool_pg_num_t, pg_t> pgs;
+    std::set<pool_pg_num_t> dirty_pgs;
+    std::set<osd_num_t> dirty_osds;
     uint64_t misplaced_objects = 0, degraded_objects = 0, incomplete_objects = 0;
     int peering_state = 0;
-    unsigned pg_count = 0;
     std::map<object_id, osd_recovery_op_t> recovery_ops;
     osd_op_t *autosync_op = NULL;
 
@@ -108,16 +109,13 @@ class osd_t
     int inflight_ops = 0;
     blockstore_t *bs;
     uint32_t bs_block_size, bs_disk_alignment;
-    uint64_t pg_stripe_size = DEFAULT_PG_STRIPE_SIZE;
     ring_loop_t *ringloop;
     timerfd_manager_t *tfd = NULL;
+    epoll_manager_t *epmgr = NULL;
 
-    int wait_state = 0;
-    int epoll_fd = 0;
     int listening_port = 0;
     int listen_fd = 0;
     ring_consumer_t consumer;
-    std::map<int, std::function<void(int, int)>> epoll_handlers;
 
     // op statistics
     osd_op_stats_t prev_stats;
@@ -128,7 +126,8 @@ class osd_t
     // cluster connection
     void parse_config(blockstore_config_t & config);
     void init_cluster();
-    void on_change_osd_state_hook(uint64_t osd_num);
+    void on_change_osd_state_hook(osd_num_t peer_osd);
+    void on_change_pg_history_hook(pool_id_t pool_id, pg_num_t pg_num);
     void on_change_etcd_state_hook(json11::Json::object & changes);
     void on_load_config_hook(json11::Json::object & changes);
     json11::Json on_load_pgs_checks_hook();
@@ -149,24 +148,22 @@ class osd_t
 
     // event loop, socket read/write
     void loop();
-    void set_fd_handler(int fd, std::function<void(int, int)> handler);
-    void handle_epoll_events();
 
     // peer handling (primary OSD logic)
     void parse_test_peer(std::string peer);
     void handle_peers();
     void repeer_pgs(osd_num_t osd_num);
-    void start_pg_peering(pg_num_t pg_num);
+    void start_pg_peering(pg_t & pg);
     void submit_sync_and_list_subop(osd_num_t role_osd, pg_peering_state_t *ps);
     void submit_list_subop(osd_num_t role_osd, pg_peering_state_t *ps);
     void discard_list_subop(osd_op_t *list_op);
-    bool stop_pg(pg_num_t pg_num);
+    bool stop_pg(pg_t & pg);
     void finish_stop_pg(pg_t & pg);
 
     // flushing, recovery and backfill
-    void submit_pg_flush_ops(pg_num_t pg_num);
+    void submit_pg_flush_ops(pg_t & pg);
-    void handle_flush_op(bool rollback, pg_num_t pg_num, pg_flush_batch_t *fb, osd_num_t peer_osd, int retval);
+    void handle_flush_op(bool rollback, pool_id_t pool_id, pg_num_t pg_num, pg_flush_batch_t *fb, osd_num_t peer_osd, int retval);
-    void submit_flush_op(pg_num_t pg_num, pg_flush_batch_t *fb, bool rollback, osd_num_t peer_osd, int count, obj_ver_id *data);
+    void submit_flush_op(pool_id_t pool_id, pg_num_t pg_num, pg_flush_batch_t *fb, bool rollback, osd_num_t peer_osd, int count, obj_ver_id *data);
     bool pick_next_recovery(osd_recovery_op_t &op);
     void submit_recovery_op(osd_recovery_op_t *op);
     bool continue_recovery();
@@ -187,22 +184,26 @@ class osd_t
     bool prepare_primary_rw(osd_op_t *cur_op);
     void continue_primary_read(osd_op_t *cur_op);
     void continue_primary_write(osd_op_t *cur_op);
+    void cancel_primary_write(osd_op_t *cur_op);
     void continue_primary_sync(osd_op_t *cur_op);
     void continue_primary_del(osd_op_t *cur_op);
     bool check_write_queue(osd_op_t *cur_op, pg_t & pg);
     void remove_object_from_state(object_id & oid, pg_osd_set_state_t *object_state, pg_t &pg);
-    bool finalize_primary_write(osd_op_t *cur_op, pg_t & pg, pg_osd_set_t & loc_set, int base_state);
+    bool remember_unstable_write(osd_op_t *cur_op, pg_t & pg, pg_osd_set_t & loc_set, int base_state);
-    void handle_primary_subop(uint64_t opcode, osd_op_t *cur_op, int retval, int expected, uint64_t version);
+    void handle_primary_subop(osd_op_t *subop, osd_op_t *cur_op);
     void handle_primary_bs_subop(osd_op_t *subop);
     void add_bs_subop_stats(osd_op_t *subop);
-    void pg_cancel_write_queue(pg_t & pg, object_id oid, int retval);
+    void pg_cancel_write_queue(pg_t & pg, osd_op_t *first_op, object_id oid, int retval);
-    void submit_primary_subops(int submit_type, int read_pg_size, const uint64_t* osd_set, osd_op_t *cur_op);
+    void submit_primary_subops(int submit_type, uint64_t op_version, int pg_size, const uint64_t* osd_set, osd_op_t *cur_op);
-    void submit_primary_del_subops(osd_op_t *cur_op, uint64_t *cur_set, pg_osd_set_t & loc_set);
+    void submit_primary_del_subops(osd_op_t *cur_op, uint64_t *cur_set, uint64_t set_size, pg_osd_set_t & loc_set);
     void submit_primary_sync_subops(osd_op_t *cur_op);
     void submit_primary_stab_subops(osd_op_t *cur_op);
 
-    inline pg_num_t map_to_pg(object_id oid)
+    inline pg_num_t map_to_pg(object_id oid, uint64_t pg_stripe_size)
     {
+        uint64_t pg_count = pg_counts[INODE_POOL(oid.inode)];
+        if (!pg_count)
+            pg_count = 1;
         return (oid.inode + oid.stripe / pg_stripe_size) % pg_count + 1;
     }
 

osd_client.cpp

@@ -1,40 +0,0 @@
-void slice()
-{
-    // Slice the request into blockstore requests to individual objects
-    // Primary OSD still operates individual stripes, except they're twice the size of the blockstore's stripe.
-    std::vector read_parts;
-    int block = bs->get_block_size();
-    uint64_t stripe1 = cur_op->req.rw.offset / block / 2;
-    uint64_t stripe2 = (cur_op->req.rw.offset + cur_op->req.rw.len + block*2 - 1) / block / 2 - 1;
-    for (uint64_t s = stripe1; s <= stripe2; s++)
-    {
-        uint64_t start = s == stripe1 ? cur_op->req.rw.offset - stripe1*block*2 : 0;
-        uint64_t end = s == stripe2 ? cur_op->req.rw.offset + cur_op->req.rw.len - stripe2*block*2 : block*2;
-        if (start < block)
-        {
-            read_parts.push_back({
-                .role = 1,
-                .oid = {
-                    .inode = cur_op->req.rw.inode,
-                    .stripe = (s << STRIPE_ROLE_BITS) | 1,
-                },
-                .version = UINT64_MAX,
-                .offset = start,
-                .len = (block < end ? block : end) - start,
-            });
-        }
-        if (end > block)
-        {
-            read_parts.push_back({
-                .role = 2,
-                .oid = {
-                    .inode = cur_op->req.rw.inode,
-                    .stripe = (s << STRIPE_ROLE_BITS) | 2,
-                },
-                .version = UINT64_MAX,
-                .offset = (start > block ? start-block : 0),
-                .len = end - (start > block ? start-block : 0),
-            });
-        }
-    }
-}

osd_cluster.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "osd.h"
 #include "base64.h"
 #include "etcd_state_client.h"
@@ -14,7 +17,7 @@ void osd_t::init_cluster()
     {
         if (run_primary)
         {
-            // Test version of clustering code with 1 PG and 2 peers
+            // Test version of clustering code with 1 pool, 1 PG and 2 peers
             // Example: peers = 2:127.0.0.1:11204,3:127.0.0.1:11205
             std::string peerstr = config["peers"];
             while (peerstr.size())
@@ -27,15 +30,16 @@ void osd_t::init_cluster()
             {
                 throw std::runtime_error("run_primary requires at least 2 peers");
             }
-            pgs[1] = (pg_t){
+            pgs[{ 1, 1 }] = (pg_t){
                 .state = PG_PEERING,
                 .pg_cursize = 0,
+                .pool_id = 1,
                 .pg_num = 1,
                 .target_set = { 1, 2, 3 },
                 .cur_set = { 0, 0, 0 },
             };
-            report_pg_state(pgs[1]);
+            report_pg_state(pgs[{ 1, 1 }]);
-            pg_count = 1;
+            pg_counts[1] = 1;
         }
         bind_socket();
     }
@@ -43,7 +47,8 @@ void osd_t::init_cluster()
     {
         st_cli.tfd = tfd;
         st_cli.log_level = log_level;
-        st_cli.on_change_osd_state_hook = [this](uint64_t peer_osd) { on_change_osd_state_hook(peer_osd); };
+        st_cli.on_change_osd_state_hook = [this](osd_num_t peer_osd) { on_change_osd_state_hook(peer_osd); };
+        st_cli.on_change_pg_history_hook = [this](pool_id_t pool_id, pg_num_t pg_num) { on_change_pg_history_hook(pool_id, pg_num); };
         st_cli.on_change_hook = [this](json11::Json::object & changes) { on_change_etcd_state_hook(changes); };
         st_cli.on_load_config_hook = [this](json11::Json::object & cfg) { on_load_config_hook(cfg); };
         st_cli.load_pgs_checks_hook = [this]() { return on_load_pgs_checks_hook(); };
@@ -83,7 +88,7 @@ void osd_t::parse_test_peer(std::string peer)
         { "addresses", json11::Json::array { addr } },
         { "port", port },
     };
-    c_cli.connect_peer(peer_osd, json11::Json::array { addr }, port);
+    c_cli.connect_peer(peer_osd, st_cli.peer_states[peer_osd]);
 }
 
 json11::Json osd_t::get_osd_state()
@@ -182,7 +187,7 @@ void osd_t::report_statistics()
         pg_stats["write_osd_set"] = pg.cur_set;
         txn.push_back(json11::Json::object {
             { "request_put", json11::Json::object {
-                { "key", base64_encode(st_cli.etcd_prefix+"/pg/stats/"+std::to_string(pg.pg_num)) },
+                { "key", base64_encode(st_cli.etcd_prefix+"/pg/stats/"+std::to_string(pg.pool_id)+"/"+std::to_string(pg.pg_num)) },
                 { "value", base64_encode(json11::Json(pg_stats).dump()) },
             } }
         });
@@ -207,11 +212,11 @@ void osd_t::report_statistics()
     });
 }
 
-void osd_t::on_change_osd_state_hook(uint64_t peer_osd)
+void osd_t::on_change_osd_state_hook(osd_num_t peer_osd)
 {
     if (c_cli.wanted_peers.find(peer_osd) != c_cli.wanted_peers.end())
     {
-        c_cli.connect_peer(peer_osd, st_cli.peer_states[peer_osd]["addresses"], st_cli.peer_states[peer_osd]["port"].int64_value());
+        c_cli.connect_peer(peer_osd, st_cli.peer_states[peer_osd]);
     }
 }
 
@@ -222,6 +227,30 @@ void osd_t::on_change_etcd_state_hook(json11::Json::object & changes)
     apply_pg_config();
 }
 
+void osd_t::on_change_pg_history_hook(pool_id_t pool_id, pg_num_t pg_num)
+{
+    auto pg_it = pgs.find({
+        .pool_id = pool_id,
+        .pg_num = pg_num,
+    });
+    if (pg_it != pgs.end() && pg_it->second.epoch > pg_it->second.reported_epoch &&
+        st_cli.pool_config[pool_id].pg_config[pg_num].epoch >= pg_it->second.epoch)
+    {
+        pg_it->second.reported_epoch = st_cli.pool_config[pool_id].pg_config[pg_num].epoch;
+        object_id oid = { 0 };
+        bool first = true;
+        for (auto op: pg_it->second.write_queue)
+        {
+            if (first || oid != op.first)
+            {
+                oid = op.first;
+                first = false;
+                continue_primary_write(op.second);
+            }
+        }
+    }
+}
+
 void osd_t::on_load_config_hook(json11::Json::object & global_config)
 {
     blockstore_config_t osd_config = this->config;
@@ -229,8 +258,14 @@ void osd_t::on_load_config_hook(json11::Json::object & global_config)
     {
         if (this->config.find(cfg_var.first) == this->config.end())
         {
-            // FIXME Convert int to str
+            if (cfg_var.second.is_string())
-            osd_config[cfg_var.first] = cfg_var.second.string_value();
+            {
+                osd_config[cfg_var.first] = cfg_var.second.string_value();
+            }
+            else
+            {
+                osd_config[cfg_var.first] = cfg_var.second.dump();
+            }
         }
     }
     parse_config(osd_config);
@@ -423,149 +458,156 @@ void osd_t::on_load_pgs_hook(bool success)
 
 void osd_t::apply_pg_count()
 {
-    pg_num_t pg_count = st_cli.pg_config.size();
+    for (auto & pool_item: st_cli.pool_config)
-    if (pg_count > 0 && (st_cli.pg_config.begin()->first != 1 || std::prev(st_cli.pg_config.end())->first != pg_count))
     {
-        printf("Invalid PG configuration: PG numbers don't cover the whole 1..%d range\n", pg_count);
+        if (pool_item.second.real_pg_count != 0 &&
-        force_stop(1);
+            pool_item.second.real_pg_count != pg_counts[pool_item.first])
-        return;
-    }
-    if (this->pg_count != 0 && this->pg_count != pg_count)
-    {
-        // Check that all PGs are offline. It is not allowed to change PG count when any PGs are online
-        // The external tool must wait for all PGs to come down before changing PG count
-        // If it doesn't wait, a restarted OSD may apply the new count immediately which will lead to bugs
-        // So an OSD just dies if it detects PG count change while there are active PGs
-        int still_active = 0;
-        for (auto & kv: pgs)
         {
-            if (kv.second.state & PG_ACTIVE)
+            // Check that all pool PGs are offline. It is not allowed to change PG count when any PGs are online
+            // The external tool must wait for all PGs to come down before changing PG count
+            // If it doesn't wait, a restarted OSD may apply the new count immediately which will lead to bugs
+            // So an OSD just dies if it detects PG count change while there are active PGs
+            int still_active = 0;
+            for (auto & kv: pgs)
             {
-                still_active++;
+                if (kv.first.pool_id == pool_item.first && (kv.second.state & PG_ACTIVE))
+                {
+                    still_active++;
+                }
+            }
+            if (still_active > 0)
+            {
+                printf("[OSD %lu] PG count change detected, but %d PG(s) are still active. This is not allowed. Exiting\n", this->osd_num, still_active);
+                force_stop(1);
+                return;
             }
         }
-        if (still_active > 0)
+        this->pg_counts[pool_item.first] = pool_item.second.real_pg_count;
-        {
-            printf("[OSD %lu] PG count change detected, but %d PG(s) are still active. This is not allowed. Exiting\n", this->osd_num, still_active);
-            force_stop(1);
-            return;
-        }
     }
-    this->pg_count = pg_count;
 }
 
 void osd_t::apply_pg_config()
 {
     bool all_applied = true;
-    for (auto & kv: st_cli.pg_config)
+    for (auto & pool_item: st_cli.pool_config)
     {
-        pg_num_t pg_num = kv.first;
+        auto pool_id = pool_item.first;
-        auto & pg_cfg = kv.second;
+        for (auto & kv: pool_item.second.pg_config)
-        bool take = pg_cfg.exists && pg_cfg.primary == this->osd_num &&
-            !pg_cfg.pause && (!pg_cfg.cur_primary || pg_cfg.cur_primary == this->osd_num);
-        bool currently_taken = this->pgs.find(pg_num) != this->pgs.end() &&
-            this->pgs[pg_num].state != PG_OFFLINE;
-        if (currently_taken && !take)
         {
-            // Stop this PG
+            pg_num_t pg_num = kv.first;
-            stop_pg(pg_num);
+            auto & pg_cfg = kv.second;
-        }
+            bool take = pg_cfg.exists && pg_cfg.primary == this->osd_num &&
-        else if (take)
+                !pg_cfg.pause && (!pg_cfg.cur_primary || pg_cfg.cur_primary == this->osd_num);
-        {
+            auto pg_it = this->pgs.find({ .pool_id = pool_id, .pg_num = pg_num });
-            // Take this PG
+            bool currently_taken = pg_it != this->pgs.end() && pg_it->second.state != PG_OFFLINE;
-            std::set<osd_num_t> all_peers;
+            if (currently_taken && !take)
-            for (osd_num_t pg_osd: pg_cfg.target_set)
             {
-                if (pg_osd != 0)
+                // Stop this PG
-                {
+                stop_pg(pg_it->second);
-                    all_peers.insert(pg_osd);
-                }
             }
-            for (osd_num_t pg_osd: pg_cfg.all_peers)
+            else if (take)
             {
-                if (pg_osd != 0)
+                // Take this PG
-                {
+                std::set<osd_num_t> all_peers;
-                    all_peers.insert(pg_osd);
+                for (osd_num_t pg_osd: pg_cfg.target_set)
-                }
-            }
-            for (auto & hist_item: pg_cfg.target_history)
-            {
-                for (auto pg_osd: hist_item)
                 {
                     if (pg_osd != 0)
                     {
                         all_peers.insert(pg_osd);
                     }
                 }
-            }
+                for (osd_num_t pg_osd: pg_cfg.all_peers)
-            if (currently_taken)
-            {
-                if (this->pgs[pg_num].state & (PG_ACTIVE | PG_INCOMPLETE | PG_PEERING))
                 {
-                    if (this->pgs[pg_num].target_set == pg_cfg.target_set)
+                    if (pg_osd != 0)
                     {
-                        // No change in osd_set; history changes are ignored
+                        all_peers.insert(pg_osd);
-                        continue;
                     }
-                    else
+                }
+                for (auto & hist_item: pg_cfg.target_history)
+                {
+                    for (auto pg_osd: hist_item)
                     {
-                        // Stop PG, reapply change after stopping
+                        if (pg_osd != 0)
-                        stop_pg(pg_num);
+                        {
+                            all_peers.insert(pg_osd);
+                        }
+                    }
+                }
+                if (currently_taken)
+                {
+                    if (pg_it->second.state & (PG_ACTIVE | PG_INCOMPLETE | PG_PEERING))
+                    {
+                        if (pg_it->second.target_set == pg_cfg.target_set)
+                        {
+                            // No change in osd_set; history changes are ignored
+                            continue;
+                        }
+                        else
+                        {
+                            // Stop PG, reapply change after stopping
+                            stop_pg(pg_it->second);
+                            all_applied = false;
+                            continue;
+                        }
+                    }
+                    else if (pg_it->second.state & PG_STOPPING)
+                    {
+                        // Reapply change after stopping
                         all_applied = false;
                         continue;
                     }
-                }
+                    else if (pg_it->second.state & PG_STARTING)
-                else if (this->pgs[pg_num].state & PG_STOPPING)
-                {
-                    // Reapply change after stopping
-                    all_applied = false;
-                    continue;
-                }
-                else if (this->pgs[pg_num].state & PG_STARTING)
-                {
-                    if (pg_cfg.cur_primary == this->osd_num)
                     {
-                        // PG locked, continue
+                        if (pg_cfg.cur_primary == this->osd_num)
+                        {
+                            // PG locked, continue
+                        }
+                        else
+                        {
+                            // Reapply change after locking the PG
+                            all_applied = false;
+                            continue;
+                        }
                     }
                     else
                     {
-                        // Reapply change after locking the PG
+                        throw std::runtime_error("Unexpected PG "+std::to_string(pg_num)+" state: "+std::to_string(pg_it->second.state));
-                        all_applied = false;
-                        continue;
                     }
                 }
+                auto & pg = this->pgs[{ .pool_id = pool_id, .pg_num = pg_num }];
+                pg = (pg_t){
+                    .state = pg_cfg.cur_primary == this->osd_num ? PG_PEERING : PG_STARTING,
+                    .scheme = pool_item.second.scheme,
+                    .pg_cursize = 0,
+                    .pg_size = pool_item.second.pg_size,
+                    .pg_minsize = pool_item.second.pg_minsize,
+                    .pool_id = pool_id,
+                    .pg_num = pg_num,
+                    .reported_epoch = pg_cfg.epoch,
+                    .target_history = pg_cfg.target_history,
+                    .all_peers = std::vector<osd_num_t>(all_peers.begin(), all_peers.end()),
+                    .target_set = pg_cfg.target_set,
+                };
+                this->pg_state_dirty.insert({ .pool_id = pool_id, .pg_num = pg_num });
+                pg.print_state();
+                if (pg_cfg.cur_primary == this->osd_num)
+                {
+                    // Add peers
+                    for (auto pg_osd: all_peers)
+                    {
+                        if (pg_osd != this->osd_num && c_cli.osd_peer_fds.find(pg_osd) == c_cli.osd_peer_fds.end())
+                        {
+                            c_cli.connect_peer(pg_osd, st_cli.peer_states[pg_osd]);
+                        }
+                    }
+                    start_pg_peering(pg);
+                }
                 else
                 {
-                    throw std::runtime_error("Unexpected PG "+std::to_string(pg_num)+" state: "+std::to_string(this->pgs[pg_num].state));
+                    // Reapply change after locking the PG
+                    all_applied = false;
                 }
             }
-            this->pgs[pg_num] = (pg_t){
-                .state = pg_cfg.cur_primary == this->osd_num ? PG_PEERING : PG_STARTING,
-                .pg_cursize = 0,
-                .pg_num = pg_num,
-                .target_history = pg_cfg.target_history,
-                .all_peers = std::vector<osd_num_t>(all_peers.begin(), all_peers.end()),
-                .target_set = pg_cfg.target_set,
-            };
-            this->pg_state_dirty.insert(pg_num);
-            this->pgs[pg_num].print_state();
-            if (pg_cfg.cur_primary == this->osd_num)
-            {
-                // Add peers
-                for (auto pg_osd: all_peers)
-                {
-                    if (pg_osd != this->osd_num && c_cli.osd_peer_fds.find(pg_osd) == c_cli.osd_peer_fds.end())
-                    {
-                        c_cli.connect_peer(pg_osd, st_cli.peer_states[pg_osd]["addresses"], st_cli.peer_states[pg_osd]["port"].int64_value());
-                    }
-                }
-                start_pg_peering(pg_num);
-            }
-            else
-            {
-                // Reapply change after locking the PG
-                all_applied = false;
-            }
         }
     }
     report_pg_states();
@@ -578,8 +620,7 @@ void osd_t::report_pg_states()
     {
         return;
     }
-    etcd_reporting_pg_state = true;
+    std::vector<std::pair<pool_pg_num_t,bool>> reporting_pgs;
-    std::vector<std::pair<pg_num_t,bool>> reporting_pgs;
     json11::Json::array checks;
     json11::Json::array success;
     json11::Json::array failure;
@@ -591,8 +632,8 @@ void osd_t::report_pg_states()
             continue;
         }
         auto & pg = pg_it->second;
-        reporting_pgs.push_back({ pg.pg_num, pg.history_changed });
+        reporting_pgs.push_back({ *it, pg.history_changed });
-        std::string state_key_base64 = base64_encode(st_cli.etcd_prefix+"/pg/state/"+std::to_string(pg.pg_num));
+        std::string state_key_base64 = base64_encode(st_cli.etcd_prefix+"/pg/state/"+std::to_string(pg.pool_id)+"/"+std::to_string(pg.pg_num));
         if (pg.state == PG_STARTING)
         {
             // Check that the PG key does not exist
@@ -634,7 +675,7 @@ void osd_t::report_pg_states()
             }
             success.push_back(json11::Json::object {
                 { "request_put", json11::Json::object {
-                    { "key", base64_encode(st_cli.etcd_prefix+"/pg/state/"+std::to_string(pg.pg_num)) },
+                    { "key", state_key_base64 },
                     { "value", base64_encode(json11::Json(json11::Json::object {
                         { "primary", this->osd_num },
                         { "state", pg_state_keywords },
@@ -645,26 +686,26 @@ void osd_t::report_pg_states()
             });
             if (pg.history_changed)
             {
+                // Prevent race conditions (for the case when the monitor is updating this key at the same time)
                 pg.history_changed = false;
-                if (pg.state == PG_ACTIVE)
+                std::string history_key = base64_encode(st_cli.etcd_prefix+"/pg/history/"+std::to_string(pg.pool_id)+"/"+std::to_string(pg.pg_num));
-                {
+                json11::Json::object history_value = {
-                    success.push_back(json11::Json::object {
+                    { "epoch", pg.epoch },
-                        { "request_delete_range", json11::Json::object {
+                    { "all_peers", pg.all_peers },
-                            { "key", base64_encode(st_cli.etcd_prefix+"/pg/history/"+std::to_string(pg.pg_num)) },
+                    { "osd_sets", pg.target_history },
-                        } }
+                };
-                    });
+                checks.push_back(json11::Json::object {
-                }
+                    { "target", "MOD" },
-                else if (pg.state == (PG_ACTIVE|PG_LEFT_ON_DEAD))
+                    { "key", history_key },
-                {
+                    { "result", "LESS" },
-                    success.push_back(json11::Json::object {
+                    { "mod_revision", st_cli.etcd_watch_revision+1 },
-                        { "request_put", json11::Json::object {
+                });
-                            { "key", base64_encode(st_cli.etcd_prefix+"/pg/history/"+std::to_string(pg.pg_num)) },
+                success.push_back(json11::Json::object {
-                            { "value", base64_encode(json11::Json(json11::Json::object {
+                    { "request_put", json11::Json::object {
-                                { "all_peers", pg.all_peers },
+                        { "key", history_key },
-                            }).dump()) },
+                        { "value", base64_encode(json11::Json(history_value).dump()) },
-                        } }
+                    } }
-                    });
+                });
-                }
             }
         }
         failure.push_back(json11::Json::object {
@@ -674,6 +715,7 @@ void osd_t::report_pg_states()
         });
     }
     pg_state_dirty.clear();
+    etcd_reporting_pg_state = true;
     st_cli.etcd_txn(json11::Json::object {
         { "compare", checks }, { "success", success }, { "failure", failure }
     }, ETCD_QUICK_TIMEOUT, [this, reporting_pgs](std::string err, json11::Json data)
@@ -699,14 +741,23 @@ void osd_t::report_pg_states()
                 if (res["kvs"].array_items().size())
                 {
                     auto kv = st_cli.parse_etcd_kv(res["kvs"][0]);
-                    pg_num_t pg_num = stoull_full(kv.key.substr(st_cli.etcd_prefix.length()+10));
+                    if (kv.key.substr(st_cli.etcd_prefix.length()+10) == st_cli.etcd_prefix+"/pg/state/")
-                    auto pg_it = pgs.find(pg_num);
-                    if (pg_it != pgs.end() && pg_it->second.state != PG_OFFLINE && pg_it->second.state != PG_STARTING)
                     {
-                        // Live PG state update failed
+                        pool_id_t pool_id = 0;
-                        printf("Failed to report state of PG %u which is live. Race condition detected, exiting\n", pg_num);
+                        pg_num_t pg_num = 0;
-                        force_stop(1);
+                        char null_byte = 0;
-                        return;
+                        sscanf(kv.key.c_str() + st_cli.etcd_prefix.length()+10, "%u/%u%c", &pool_id, &pg_num, &null_byte);
+                        if (null_byte == 0)
+                        {
+                            auto pg_it = pgs.find({ .pool_id = pool_id, .pg_num = pg_num });
+                            if (pg_it != pgs.end() && pg_it->second.state != PG_OFFLINE && pg_it->second.state != PG_STARTING)
+                            {
+                                // Live PG state update failed
+                                printf("Failed to report state of pool %u PG %u which is live. Race condition detected, exiting\n", pool_id, pg_num);
+                                force_stop(1);
+                                return;
+                            }
+                        }
                     }
                 }
             }

osd_flush.cpp

@@ -1,10 +1,12 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "osd.h"
 
 #define FLUSH_BATCH 512
 
-void osd_t::submit_pg_flush_ops(pg_num_t pg_num)
+void osd_t::submit_pg_flush_ops(pg_t & pg)
 {
-    pg_t & pg = pgs[pg_num];
     pg_flush_batch_t *fb = new pg_flush_batch_t();
     pg.flush_batch = fb;
     auto it = pg.flush_actions.begin(), prev_it = pg.flush_actions.begin();
@@ -45,7 +47,7 @@ void osd_t::submit_pg_flush_ops(pg_num_t pg_num)
         if (l.second.size() > 0)
         {
             fb->flush_ops++;
-            submit_flush_op(pg.pg_num, fb, true, l.first, l.second.size(), l.second.data());
+            submit_flush_op(pg.pool_id, pg.pg_num, fb, true, l.first, l.second.size(), l.second.data());
         }
     }
     for (auto & l: fb->stable_lists)
@@ -53,14 +55,15 @@ void osd_t::submit_pg_flush_ops(pg_num_t pg_num)
         if (l.second.size() > 0)
         {
             fb->flush_ops++;
-            submit_flush_op(pg.pg_num, fb, false, l.first, l.second.size(), l.second.data());
+            submit_flush_op(pg.pool_id, pg.pg_num, fb, false, l.first, l.second.size(), l.second.data());
         }
     }
 }
 
-void osd_t::handle_flush_op(bool rollback, pg_num_t pg_num, pg_flush_batch_t *fb, osd_num_t peer_osd, int retval)
+void osd_t::handle_flush_op(bool rollback, pool_id_t pool_id, pg_num_t pg_num, pg_flush_batch_t *fb, osd_num_t peer_osd, int retval)
 {
-    if (pgs.find(pg_num) == pgs.end() || pgs[pg_num].flush_batch != fb)
+    pool_pg_num_t pg_id = { .pool_id = pool_id, .pg_num = pg_num };
+    if (pgs.find(pg_id) == pgs.end() || pgs[pg_id].flush_batch != fb)
     {
         // Throw the result away
         return;
@@ -78,9 +81,12 @@ void osd_t::handle_flush_op(bool rollback, pg_num_t pg_num, pg_flush_batch_t *fb
         }
         else
         {
-            printf("Error while doing flush on OSD %lu: %s\n", osd_num, strerror(-retval));
+            printf("Error while doing flush on OSD %lu: %d (%s)\n", osd_num, retval, strerror(-retval));
-            assert(c_cli.osd_peer_fds.find(peer_osd) != c_cli.osd_peer_fds.end());
+            auto fd_it = c_cli.osd_peer_fds.find(peer_osd);
-            c_cli.stop_client(c_cli.osd_peer_fds[peer_osd]);
+            if (fd_it != c_cli.osd_peer_fds.end())
+            {
+                c_cli.stop_client(fd_it->second);
+            }
             return;
         }
     }
@@ -89,7 +95,7 @@ void osd_t::handle_flush_op(bool rollback, pg_num_t pg_num, pg_flush_batch_t *fb
     {
         // This flush batch is done
         std::vector<osd_op_t*> continue_ops;
-        auto & pg = pgs[pg_num];
+        auto & pg = pgs[pg_id];
         auto it = pg.flush_actions.begin(), prev_it = it;
         auto erase_start = it;
         while (1)
@@ -150,11 +156,11 @@ void osd_t::handle_flush_op(bool rollback, pg_num_t pg_num, pg_flush_batch_t *fb
     }
 }
 
-void osd_t::submit_flush_op(pg_num_t pg_num, pg_flush_batch_t *fb, bool rollback, osd_num_t peer_osd, int count, obj_ver_id *data)
+void osd_t::submit_flush_op(pool_id_t pool_id, pg_num_t pg_num, pg_flush_batch_t *fb, bool rollback, osd_num_t peer_osd, int count, obj_ver_id *data)
 {
     osd_op_t *op = new osd_op_t();
     // Copy buffer so it gets freed along with the operation
-    op->buf = malloc(sizeof(obj_ver_id) * count);
+    op->buf = malloc_or_die(sizeof(obj_ver_id) * count);
     memcpy(op->buf, data, sizeof(obj_ver_id) * count);
     if (peer_osd == this->osd_num)
     {
@@ -162,10 +168,10 @@ void osd_t::submit_flush_op(pg_num_t pg_num, pg_flush_batch_t *fb, bool rollback
         clock_gettime(CLOCK_REALTIME, &op->tv_begin);
         op->bs_op = new blockstore_op_t({
             .opcode = (uint64_t)(rollback ? BS_OP_ROLLBACK : BS_OP_STABLE),
-            .callback = [this, op, pg_num, fb](blockstore_op_t *bs_op)
+            .callback = [this, op, pool_id, pg_num, fb](blockstore_op_t *bs_op)
             {
                 add_bs_subop_stats(op);
-                handle_flush_op(bs_op->opcode == BS_OP_ROLLBACK, pg_num, fb, this->osd_num, bs_op->retval);
+                handle_flush_op(bs_op->opcode == BS_OP_ROLLBACK, pool_id, pg_num, fb, this->osd_num, bs_op->retval);
                 delete op->bs_op;
                 op->bs_op = NULL;
                 delete op;
@@ -180,22 +186,21 @@ void osd_t::submit_flush_op(pg_num_t pg_num, pg_flush_batch_t *fb, bool rollback
         // Peer
         int peer_fd = c_cli.osd_peer_fds[peer_osd];
         op->op_type = OSD_OP_OUT;
-        op->send_list.push_back(op->req.buf, OSD_PACKET_SIZE);
+        op->iov.push_back(op->buf, count * sizeof(obj_ver_id));
-        op->send_list.push_back(op->buf, count * sizeof(obj_ver_id));
         op->peer_fd = peer_fd;
         op->req = {
             .sec_stab = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,
                     .id = c_cli.next_subop_id++,
-                    .opcode = (uint64_t)(rollback ? OSD_OP_SECONDARY_ROLLBACK : OSD_OP_SECONDARY_STABILIZE),
+                    .opcode = (uint64_t)(rollback ? OSD_OP_SEC_ROLLBACK : OSD_OP_SEC_STABILIZE),
                 },
                 .len = count * sizeof(obj_ver_id),
             },
         };
-        op->callback = [this, pg_num, fb, peer_osd](osd_op_t *op)
+        op->callback = [this, pool_id, pg_num, fb, peer_osd](osd_op_t *op)
         {
-            handle_flush_op(op->req.hdr.opcode == OSD_OP_SECONDARY_ROLLBACK, pg_num, fb, peer_osd, op->reply.hdr.retval);
+            handle_flush_op(op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK, pool_id, pg_num, fb, peer_osd, op->reply.hdr.retval);
             delete op;
         };
         c_cli.outbox_push(op);
@@ -213,7 +218,6 @@ bool osd_t::pick_next_recovery(osd_recovery_op_t &op)
                 if (recovery_ops.find(obj_it->first) == recovery_ops.end())
                 {
                     op.degraded = true;
-                    op.pg_num = pg_it->first;
                     op.oid = obj_it->first;
                     return true;
                 }
@@ -229,7 +233,6 @@ bool osd_t::pick_next_recovery(osd_recovery_op_t &op)
                 if (recovery_ops.find(obj_it->first) == recovery_ops.end())
                 {
                     op.degraded = false;
-                    op.pg_num = pg_it->first;
                     op.oid = obj_it->first;
                     return true;
                 }
@@ -270,9 +273,10 @@ void osd_t::submit_recovery_op(osd_recovery_op_t *op)
                 throw std::runtime_error("Failed to recover an object");
             }
         }
+        // CAREFUL! op = &recovery_ops[op->oid]. Don't access op->* after recovery_ops.erase()
+        op->osd_op = NULL;
         recovery_ops.erase(op->oid);
         delete osd_op;
-        op->osd_op = NULL;
         continue_recovery();
     };
     exec_op(op->osd_op);

osd_id.h

@@ -1,4 +1,27 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #pragma once
 
+#define POOL_SCHEME_REPLICATED 1
+#define POOL_SCHEME_XOR 2
+#define POOL_ID_MAX 0x10000
+#define POOL_ID_BITS 16
+#define INODE_POOL(inode) (pool_id_t)((inode) >> (64 - POOL_ID_BITS))
+
+// Pool ID is 16 bits long
+typedef uint32_t pool_id_t;
+
 typedef uint64_t osd_num_t;
 typedef uint32_t pg_num_t;
+
+struct pool_pg_num_t
+{
+    pool_id_t pool_id;
+    pg_num_t pg_num;
+};
+
+inline bool operator < (const pool_pg_num_t & a, const pool_pg_num_t & b)
+{
+    return a.pool_id < b.pool_id || a.pool_id == b.pool_id && a.pg_num < b.pg_num;
+}

osd_main.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "osd.h"
 
 #include <signal.h>
@@ -18,6 +21,8 @@ static void handle_sigint(int sig)
 
 int main(int narg, char *args[])
 {
+    setvbuf(stdout, NULL, _IONBF, 0);
+    setvbuf(stderr, NULL, _IONBF, 0);
     if (sizeof(osd_any_op_t) > OSD_PACKET_SIZE ||
         sizeof(osd_any_reply_t) > OSD_PACKET_SIZE)
     {

osd_ops.cpp

@@ -0,0 +1,22 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+#include "osd_ops.h"
+
+const char* osd_op_names[] = {
+    "",
+    "read",
+    "write",
+    "write_stable",
+    "sync",
+    "stabilize",
+    "rollback",
+    "delete",
+    "sync_stab_all",
+    "list",
+    "show_config",
+    "primary_read",
+    "primary_write",
+    "primary_sync",
+    "primary_delete",
+};

osd_ops.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #pragma once
 
 #include "object_id.h"
@@ -10,20 +13,21 @@
 #define OSD_PACKET_SIZE             0x80
 // Opcodes
 #define OSD_OP_MIN                  1
-#define OSD_OP_SECONDARY_READ       1
+#define OSD_OP_SEC_READ             1
-#define OSD_OP_SECONDARY_WRITE      2
+#define OSD_OP_SEC_WRITE            2
-#define OSD_OP_SECONDARY_SYNC       3
+#define OSD_OP_SEC_WRITE_STABLE     3
-#define OSD_OP_SECONDARY_STABILIZE  4
+#define OSD_OP_SEC_SYNC             4
-#define OSD_OP_SECONDARY_ROLLBACK   5
+#define OSD_OP_SEC_STABILIZE        5
-#define OSD_OP_SECONDARY_DELETE     6
+#define OSD_OP_SEC_ROLLBACK         6
-#define OSD_OP_TEST_SYNC_STAB_ALL   7
+#define OSD_OP_SEC_DELETE           7
-#define OSD_OP_SECONDARY_LIST       8
+#define OSD_OP_TEST_SYNC_STAB_ALL   8
-#define OSD_OP_SHOW_CONFIG          9
+#define OSD_OP_SEC_LIST             9
-#define OSD_OP_READ                 10
+#define OSD_OP_SHOW_CONFIG          10
-#define OSD_OP_WRITE                11
+#define OSD_OP_READ                 11
-#define OSD_OP_SYNC                 12
+#define OSD_OP_WRITE                12
-#define OSD_OP_DELETE               13
+#define OSD_OP_SYNC                 13
-#define OSD_OP_MAX                  13
+#define OSD_OP_DELETE               14
+#define OSD_OP_MAX                  14
 // Alignment & limit for read/write operations
 #ifndef MEM_ALIGNMENT
 #define MEM_ALIGNMENT               512
@@ -134,7 +138,10 @@ struct __attribute__((__packed__)) osd_op_secondary_list_t
     osd_op_header_t header;
     // placement group total number and total count
     pg_num_t list_pg, pg_count;
+    // size of an area that maps to one PG continuously
     uint64_t pg_stripe_size;
+    // inode range (used to select pools)
+    uint64_t min_inode, max_inode;
 };
 
 struct __attribute__((__packed__)) osd_reply_secondary_list_t
@@ -202,3 +209,5 @@ union osd_any_reply_t
     osd_reply_sync_t sync;
     uint8_t buf[OSD_PACKET_SIZE];
 };
+
+extern const char* osd_op_names[];

osd_peering.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <netinet/tcp.h>
 #include <sys/epoll.h>
 
@@ -26,7 +29,7 @@ void osd_t::handle_peers()
                     degraded_objects += p.second.degraded_objects.size();
                     if ((p.second.state & (PG_ACTIVE | PG_HAS_UNCLEAN)) == (PG_ACTIVE | PG_HAS_UNCLEAN))
                         peering_state = peering_state | OSD_FLUSHING_PGS;
-                    else
+                    else if (p.second.state & PG_ACTIVE)
                         peering_state = peering_state | OSD_RECOVERING;
                 }
                 else
@@ -50,7 +53,7 @@ void osd_t::handle_peers()
             {
                 if (!p.second.flush_batch)
                 {
-                    submit_pg_flush_ops(p.first);
+                    submit_pg_flush_ops(p.second);
                 }
                 still = true;
             }
@@ -89,16 +92,15 @@ void osd_t::repeer_pgs(osd_num_t peer_osd)
             {
                 // Repeer this pg
                 printf("[PG %u] Repeer because of OSD %lu\n", p.second.pg_num, peer_osd);
-                start_pg_peering(p.second.pg_num);
+                start_pg_peering(p.second);
             }
         }
     }
 }
 
 // Repeer on each connect/disconnect peer event
-void osd_t::start_pg_peering(pg_num_t pg_num)
+void osd_t::start_pg_peering(pg_t & pg)
 {
-    auto & pg = pgs[pg_num];
     pg.state = PG_PEERING;
     this->peering_state |= OSD_PEERING_PGS;
     report_pg_state(pg);
@@ -120,20 +122,35 @@ void osd_t::start_pg_peering(pg_num_t pg_num)
     pg.flush_batch = NULL;
     for (auto p: pg.write_queue)
     {
-        finish_op(p.second, -EPIPE);
+        cancel_primary_write(p.second);
     }
     pg.write_queue.clear();
+    uint64_t pg_stripe_size = st_cli.pool_config[pg.pool_id].pg_stripe_size;
     for (auto it = unstable_writes.begin(); it != unstable_writes.end(); )
     {
         // Forget this PG's unstable writes
-        pg_num_t n = (it->first.oid.inode + it->first.oid.stripe / pg_stripe_size) % pg_count + 1;
+        if (INODE_POOL(it->first.oid.inode) == pg.pool_id && map_to_pg(it->first.oid, pg_stripe_size) == pg.pg_num)
-        if (n == pg.pg_num)
             unstable_writes.erase(it++);
         else
             it++;
     }
-    pg.inflight = 0;
+    dirty_pgs.erase({ .pool_id = pg.pool_id, .pg_num = pg.pg_num });
-    dirty_pgs.erase(pg.pg_num);
+    // Drop connections of clients who have this PG in dirty_pgs
+    if (immediate_commit != IMMEDIATE_ALL)
+    {
+        std::vector<int> to_stop;
+        for (auto & cp: c_cli.clients)
+        {
+            if (cp.second.dirty_pgs.find({ .pool_id = pg.pool_id, .pg_num = pg.pg_num }) != cp.second.dirty_pgs.end())
+            {
+                to_stop.push_back(cp.first);
+            }
+        }
+        for (auto peer_fd: to_stop)
+        {
+            c_cli.stop_client(peer_fd);
+        }
+    }
     // Calculate current write OSD set
     pg.pg_cursize = 0;
     pg.cur_set.resize(pg.target_set.size());
@@ -171,6 +188,7 @@ void osd_t::start_pg_peering(pg_num_t pg_num)
             {
                 pg.state = PG_INCOMPLETE;
                 report_pg_state(pg);
+                return;
             }
         }
     }
@@ -178,6 +196,7 @@ void osd_t::start_pg_peering(pg_num_t pg_num)
     {
         pg.state = PG_INCOMPLETE;
         report_pg_state(pg);
+        return;
     }
     std::set<osd_num_t> cur_peers;
     for (auto pg_osd: pg.all_peers)
@@ -188,7 +207,7 @@ void osd_t::start_pg_peering(pg_num_t pg_num)
         }
         else if (c_cli.wanted_peers.find(pg_osd) == c_cli.wanted_peers.end())
         {
-            c_cli.connect_peer(pg_osd, st_cli.peer_states[pg_osd]["addresses"], st_cli.peer_states[pg_osd]["port"].int64_value());
+            c_cli.connect_peer(pg_osd, st_cli.peer_states[pg_osd]);
         }
     }
     pg.cur_peers.insert(pg.cur_peers.begin(), cur_peers.begin(), cur_peers.end());
@@ -234,6 +253,7 @@ void osd_t::start_pg_peering(pg_num_t pg_num)
     if (!pg.peering_state)
     {
         pg.peering_state = new pg_peering_state_t();
+        pg.peering_state->pool_id = pg.pool_id;
         pg.peering_state->pg_num = pg.pg_num;
     }
     for (osd_num_t peer_osd: cur_peers)
@@ -288,14 +308,13 @@ void osd_t::submit_sync_and_list_subop(osd_num_t role_osd, pg_peering_state_t *p
         auto & cl = c_cli.clients.at(c_cli.osd_peer_fds[role_osd]);
         osd_op_t *op = new osd_op_t();
         op->op_type = OSD_OP_OUT;
-        op->send_list.push_back(op->req.buf, OSD_PACKET_SIZE);
         op->peer_fd = cl.peer_fd;
         op->req = {
             .sec_sync = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,
                     .id = c_cli.next_subop_id++,
-                    .opcode = OSD_OP_SECONDARY_SYNC,
+                    .opcode = OSD_OP_SEC_SYNC,
                 },
             },
         };
@@ -330,8 +349,10 @@ void osd_t::submit_list_subop(osd_num_t role_osd, pg_peering_state_t *ps)
         clock_gettime(CLOCK_REALTIME, &op->tv_begin);
         op->bs_op = new blockstore_op_t();
         op->bs_op->opcode = BS_OP_LIST;
-        op->bs_op->oid.stripe = pg_stripe_size;
+        op->bs_op->oid.stripe = st_cli.pool_config[ps->pool_id].pg_stripe_size;
-        op->bs_op->len = pg_count;
+        op->bs_op->oid.inode = ((uint64_t)ps->pool_id << (64 - POOL_ID_BITS));
+        op->bs_op->version = ((uint64_t)(ps->pool_id+1) << (64 - POOL_ID_BITS)) - 1;
+        op->bs_op->len = pg_counts[ps->pool_id];
         op->bs_op->offset = ps->pg_num-1;
         op->bs_op->callback = [this, ps, op, role_osd](blockstore_op_t *bs_op)
         {
@@ -362,18 +383,19 @@ void osd_t::submit_list_subop(osd_num_t role_osd, pg_peering_state_t *ps)
         // Peer
         osd_op_t *op = new osd_op_t();
         op->op_type = OSD_OP_OUT;
-        op->send_list.push_back(op->req.buf, OSD_PACKET_SIZE);
         op->peer_fd = c_cli.osd_peer_fds[role_osd];
         op->req = {
             .sec_list = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,
                     .id = c_cli.next_subop_id++,
-                    .opcode = OSD_OP_SECONDARY_LIST,
+                    .opcode = OSD_OP_SEC_LIST,
                 },
                 .list_pg = ps->pg_num,
-                .pg_count = pg_count,
+                .pg_count = pg_counts[ps->pool_id],
-                .pg_stripe_size = pg_stripe_size,
+                .pg_stripe_size = st_cli.pool_config[ps->pool_id].pg_stripe_size,
+                .min_inode = ((uint64_t)(ps->pool_id) << (64 - POOL_ID_BITS)),
+                .max_inode = ((uint64_t)(ps->pool_id+1) << (64 - POOL_ID_BITS)) - 1,
             },
         };
         op->callback = [this, ps, role_osd](osd_op_t *op)
@@ -429,14 +451,8 @@ void osd_t::discard_list_subop(osd_op_t *list_op)
     }
 }
 
-bool osd_t::stop_pg(pg_num_t pg_num)
+bool osd_t::stop_pg(pg_t & pg)
 {
-    auto pg_it = pgs.find(pg_num);
-    if (pg_it == pgs.end())
-    {
-        return false;
-    }
-    auto & pg = pg_it->second;
     if (pg.peering_state)
     {
         // Stop peering
@@ -479,7 +495,7 @@ void osd_t::finish_stop_pg(pg_t & pg)
 void osd_t::report_pg_state(pg_t & pg)
 {
     pg.print_state();
-    this->pg_state_dirty.insert(pg.pg_num);
+    this->pg_state_dirty.insert({ .pool_id = pg.pool_id, .pg_num = pg.pg_num });
     if (pg.state == PG_ACTIVE && (pg.target_history.size() > 0 || pg.all_peers.size() > pg.target_set.size()))
     {
         // Clear history of active+clean PGs

osd_peering_pg.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "osd_peering_pg.h"
 
 struct obj_ver_role
@@ -33,6 +36,7 @@ struct obj_piece_ver_t
 struct pg_obj_state_check_t
 {
     pg_t *pg;
+    bool replicated = false;
     std::vector<obj_ver_role> list;
     int list_pos;
     int obj_start = 0, obj_end = 0, ver_start = 0, ver_end = 0;
@@ -41,7 +45,7 @@ struct pg_obj_state_check_t
     uint64_t last_ver = 0;
     uint64_t target_ver = 0;
     uint64_t n_copies = 0, has_roles = 0, n_roles = 0, n_stable = 0, n_mismatched = 0;
-    uint64_t n_unstable = 0, n_buggy = 0;
+    uint64_t n_unstable = 0, n_invalid = 0;
     pg_osd_set_t osd_set;
     int log_level;
 
@@ -73,6 +77,12 @@ void pg_obj_state_check_t::walk()
     {
         finish_object();
     }
+    if (pg->state & PG_HAS_INVALID)
+    {
+        // Stop PGs with "invalid" objects
+        pg->state = PG_INCOMPLETE | PG_HAS_INVALID;
+        return;
+    }
     if (pg->pg_cursize < pg->pg_size)
     {
         pg->state |= PG_DEGRADED;
@@ -92,7 +102,7 @@ void pg_obj_state_check_t::start_object()
     target_ver = 0;
     ver_start = list_pos;
     has_roles = n_copies = n_roles = n_stable = n_mismatched = 0;
-    n_unstable = n_buggy = 0;
+    n_unstable = n_invalid = 0;
 }
 
 void pg_obj_state_check_t::handle_version()
@@ -111,11 +121,11 @@ void pg_obj_state_check_t::handle_version()
             has_roles = n_copies = n_roles = n_stable = n_mismatched = 0;
             last_ver = list[list_pos].version;
         }
-        int replica = (list[list_pos].oid.stripe & STRIPE_MASK);
+        unsigned replica = (list[list_pos].oid.stripe & STRIPE_MASK);
         n_copies++;
-        if (replica >= pg->pg_size)
+        if (replicated && replica > 0 || replica >= pg->pg_size)
         {
-            n_buggy++;
+            n_invalid++;
         }
         else
         {
@@ -123,14 +133,32 @@ void pg_obj_state_check_t::handle_version()
             {
                 n_stable++;
             }
-            if (pg->cur_set[replica] != list[list_pos].osd_num)
+            if (replicated)
             {
-                n_mismatched++;
+                int i;
+                for (i = 0; i < pg->cur_set.size(); i++)
+                {
+                    if (pg->cur_set[i] == list[list_pos].osd_num)
+                    {
+                        break;
+                    }
+                }
+                if (i == pg->cur_set.size())
+                {
+                    n_mismatched++;
+                }
             }
-            if (!(has_roles & (1 << replica)))
+            else
             {
-                has_roles = has_roles | (1 << replica);
+                if (pg->cur_set[replica] != list[list_pos].osd_num)
-                n_roles++;
+                {
+                    n_mismatched++;
+                }
+                if (!(has_roles & (1 << replica)))
+                {
+                    has_roles = has_roles | (1 << replica);
+                    n_roles++;
+                }
             }
         }
     }
@@ -151,11 +179,14 @@ void pg_obj_state_check_t::finish_object()
     obj_end = list_pos;
     // Remember the decision
     uint64_t state = 0;
-    if (n_buggy > 0)
+    if (n_invalid > 0)
     {
-        state = OBJ_BUGGY;
+        // It's not allowed to change the replication scheme for a pool other than by recreating it
-        // FIXME: bring pg offline
+        // So we must bring the PG offline
-        throw std::runtime_error("buggy object state");
+        state = OBJ_INCOMPLETE;
+        pg->state |= PG_HAS_INVALID;
+        pg->total_count++;
+        return;
     }
     if (n_unstable > 0)
     {
@@ -201,51 +232,52 @@ void pg_obj_state_check_t::finish_object()
     {
         return;
     }
-    if (n_roles < pg->pg_minsize)
+    if (!replicated && n_roles < pg->pg_minsize)
     {
         if (log_level > 1)
         {
             printf("Object is incomplete: inode=%lu stripe=%lu version=%lu/%lu\n", oid.inode, oid.stripe, target_ver, max_ver);
-            for (int i = ver_start; i < ver_end; i++)
-            {
-                printf("Present on: osd %lu, role %ld%s\n", list[i].osd_num, (list[i].oid.stripe & STRIPE_MASK), list[i].is_stable ? " (stable)" : "");
-            }
-        }
-        if (log_level > 2)
-        {
-            for (int i = obj_start; i < obj_end; i++)
-            {
-                printf("v%lu present on: osd %lu, role %ld%s\n", list[i].version, list[i].osd_num, (list[i].oid.stripe & STRIPE_MASK), list[i].is_stable ? " (stable)" : "");
-            }
         }
         state = OBJ_INCOMPLETE;
         pg->state = pg->state | PG_HAS_INCOMPLETE;
     }
-    else if (n_roles < pg->pg_cursize)
+    else if ((replicated ? n_copies : n_roles) < pg->pg_cursize)
     {
         if (log_level > 1)
         {
             printf("Object is degraded: inode=%lu stripe=%lu version=%lu/%lu\n", oid.inode, oid.stripe, target_ver, max_ver);
-            for (int i = ver_start; i < ver_end; i++)
-            {
-                printf("Present on: osd %lu, role %ld%s\n", list[i].osd_num, (list[i].oid.stripe & STRIPE_MASK), list[i].is_stable ? " (stable)" : "");
-            }
-        }
-        if (log_level > 2)
-        {
-            for (int i = obj_start; i < obj_end; i++)
-            {
-                printf("v%lu present on: osd %lu, role %ld%s\n", list[i].version, list[i].osd_num, (list[i].oid.stripe & STRIPE_MASK), list[i].is_stable ? " (stable)" : "");
-            }
         }
         state = OBJ_DEGRADED;
         pg->state = pg->state | PG_HAS_DEGRADED;
     }
-    if (n_mismatched > 0)
+    else if (n_mismatched > 0)
     {
+        if (log_level > 1 && (replicated || n_roles >= pg->pg_cursize))
+        {
+            printf("Object is misplaced: inode=%lu stripe=%lu version=%lu/%lu\n", oid.inode, oid.stripe, target_ver, max_ver);
+        }
         state |= OBJ_MISPLACED;
         pg->state = pg->state | PG_HAS_MISPLACED;
     }
+    if (log_level > 1 && ((replicated ? n_copies : n_roles) < pg->pg_cursize || n_mismatched > 0))
+    {
+        if (log_level > 2)
+        {
+            for (int i = obj_start; i < obj_end; i++)
+            {
+                printf("v%lu present on: osd %lu, role %ld%s\n", list[i].version, list[i].osd_num,
+                    (list[i].oid.stripe & STRIPE_MASK), list[i].is_stable ? " (stable)" : "");
+            }
+        }
+        else
+        {
+            for (int i = ver_start; i < ver_end; i++)
+            {
+                printf("Target version present on: osd %lu, role %ld%s\n", list[i].osd_num,
+                    (list[i].oid.stripe & STRIPE_MASK), list[i].is_stable ? " (stable)" : "");
+            }
+        }
+    }
     pg->total_count++;
     if (state != 0 || ver_end < obj_end)
     {
@@ -279,8 +311,11 @@ void pg_obj_state_check_t::finish_object()
                     .osd_num = list[i].osd_num,
                     .outdated = true,
                 });
-                state |= OBJ_MISPLACED;
+                if (!(state & (OBJ_INCOMPLETE | OBJ_DEGRADED)))
-                pg->state = pg->state | PG_HAS_MISPLACED;
+                {
+                    state |= OBJ_MISPLACED;
+                    pg->state = pg->state | PG_HAS_MISPLACED;
+                }
             }
         }
     }
@@ -298,16 +333,34 @@ void pg_obj_state_check_t::finish_object()
         if (it == pg->state_dict.end())
         {
             std::vector<uint64_t> read_target;
-            read_target.resize(pg->pg_size);
+            if (replicated)
-            for (int i = 0; i < pg->pg_size; i++)
             {
-                read_target[i] = 0;
+                for (auto & o: osd_set)
-            }
-            for (auto & o: osd_set)
-            {
-                if (!o.outdated)
                 {
-                    read_target[o.role] = o.osd_num;
+                    if (!o.outdated)
+                    {
+                        read_target.push_back(o.osd_num);
+                    }
+                }
+                while (read_target.size() < pg->pg_size)
+                {
+                    // FIXME: This is because we then use .data() and assume it's at least <pg_size> long
+                    read_target.push_back(0);
+                }
+            }
+            else
+            {
+                read_target.resize(pg->pg_size);
+                for (int i = 0; i < pg->pg_size; i++)
+                {
+                    read_target[i] = 0;
+                }
+                for (auto & o: osd_set)
+                {
+                    if (!o.outdated)
+                    {
+                        read_target[o.role] = o.osd_num;
+                    }
                 }
             }
             pg->state_dict[osd_set] = {
@@ -344,7 +397,9 @@ void pg_t::calc_object_states(int log_level)
     pg_obj_state_check_t st;
     st.log_level = log_level;
     st.pg = this;
+    st.replicated = (this->scheme == POOL_SCHEME_REPLICATED);
     auto ps = peering_state;
+    epoch = 0;
     for (auto it: ps->list_results)
     {
         auto nstab = it.second.stable_count;
@@ -355,6 +410,10 @@ void pg_t::calc_object_states(int log_level)
         obj_ver_id *ov = it.second.buf;
         for (uint64_t i = 0; i < n; i++, ov++)
         {
+            if ((ov->version >> (64-PG_EPOCH_BITS)) > epoch)
+            {
+                epoch = (ov->version >> (64-PG_EPOCH_BITS));
+            }
             st.list[start+i] = {
                 .oid = ov->oid,
                 .version = ov->version,
@@ -370,12 +429,17 @@ void pg_t::calc_object_states(int log_level)
     std::sort(st.list.begin(), st.list.end());
     // Walk over it and check object states
     st.walk();
+    if (this->state & (PG_DEGRADED|PG_LEFT_ON_DEAD))
+    {
+        assert(epoch != ((1ul << PG_EPOCH_BITS)-1));
+        epoch++;
+    }
 }
 
 void pg_t::print_state()
 {
     printf(
-        "[PG %u] is %s%s%s%s%s%s%s%s%s%s%s (%lu objects)\n", pg_num,
+        "[PG %u] is %s%s%s%s%s%s%s%s%s%s%s%s%s (%lu objects)\n", pg_num,
         (state & PG_STARTING) ? "starting" : "",
         (state & PG_OFFLINE) ? "offline" : "",
         (state & PG_PEERING) ? "peering" : "",
@@ -387,6 +451,8 @@ void pg_t::print_state()
         (state & PG_HAS_DEGRADED) ? " + has_degraded" : "",
         (state & PG_HAS_MISPLACED) ? " + has_misplaced" : "",
         (state & PG_HAS_UNCLEAN) ? " + has_unclean" : "",
+        (state & PG_HAS_INVALID) ? " + has_invalid" : "",
+        (state & PG_LEFT_ON_DEAD) ? " + left_on_dead" : "",
         total_count
     );
 }

osd_peering_pg.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <map>
 #include <unordered_map>
 #include <vector>
@@ -9,6 +12,8 @@
 #include "osd_ops.h"
 #include "pg_states.h"
 
+#define PG_EPOCH_BITS 48
+
 struct pg_obj_loc_t
 {
     uint64_t role;
@@ -42,6 +47,7 @@ struct pg_peering_state_t
     // osd_num -> list result
     std::unordered_map<osd_num_t, osd_op_t*> list_ops;
     std::unordered_map<osd_num_t, pg_list_result_t> list_results;
+    pool_id_t pool_id = 0;
     pg_num_t pg_num = 0;
 };
 
@@ -69,9 +75,13 @@ struct pg_flush_batch_t
 struct pg_t
 {
     int state = 0;
-    uint64_t pg_cursize = 3, pg_size = 3, pg_minsize = 2;
+    uint64_t scheme = 0;
-    pg_num_t pg_num;
+    uint64_t pg_cursize = 0, pg_size = 0, pg_minsize = 0;
+    pool_id_t pool_id = 0;
+    pg_num_t pg_num = 0;
     uint64_t clean_count = 0, total_count = 0;
+    // epoch number - should increase with each non-clean activation of the PG
+    uint64_t epoch = 0, reported_epoch = 0;
     // target history and all potential peers
     std::vector<std::vector<osd_num_t>> target_history;
     std::vector<osd_num_t> all_peers;

osd_peering_pg_test.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #define _LARGEFILE64_SOURCE
 
 #include "osd_peering_pg.h"
@@ -28,7 +31,7 @@ int main(int argc, char *argv[])
     for (uint64_t osd_num = 1; osd_num <= 3; osd_num++)
     {
         pg_list_result_t r = {
-            .buf = (obj_ver_id*)malloc(sizeof(obj_ver_id) * 1024*1024*8),
+            .buf = (obj_ver_id*)malloc_or_die(sizeof(obj_ver_id) * 1024*1024*8),
             .total_count = 1024*1024*8,
             .stable_count = (uint64_t)(1024*1024*8 - (osd_num == 1 ? 10 : 0)),
         };

osd_primary.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "osd_primary.h"
 
 // read: read directly or read paired stripe(s), reconstruct, return
@@ -13,23 +16,23 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
 {
     // PG number is calculated from the offset
     // Our EC scheme stores data in fixed chunks equal to (K*block size)
-    // But we must not use K in the process of calculating the PG number
+    // K = pg_minsize in case of EC/XOR, or 1 for replicated pools
-    // So we calculate the PG number using a separate setting which should be per-inode (FIXME)
+    pool_id_t pool_id = INODE_POOL(cur_op->req.rw.inode);
-    pg_num_t pg_num = (cur_op->req.rw.inode + cur_op->req.rw.offset / pg_stripe_size) % pg_count + 1;
+    auto & pool_cfg = st_cli.pool_config[pool_id];
-    auto pg_it = pgs.find(pg_num);
+    uint64_t pg_block_size = bs_block_size * (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_minsize);
+    object_id oid = {
+        .inode = cur_op->req.rw.inode,
+        // oid.stripe = starting offset of the parity stripe
+        .stripe = (cur_op->req.rw.offset/pg_block_size)*pg_block_size,
+    };
+    pg_num_t pg_num = (cur_op->req.rw.inode + oid.stripe/pool_cfg.pg_stripe_size) % pg_counts[pool_id] + 1;
+    auto pg_it = pgs.find({ .pool_id = pool_id, .pg_num = pg_num });
     if (pg_it == pgs.end() || !(pg_it->second.state & PG_ACTIVE))
     {
         // This OSD is not primary for this PG or the PG is inactive
         finish_op(cur_op, -EPIPE);
         return false;
     }
-    uint64_t pg_block_size = bs_block_size * pg_it->second.pg_minsize;
-    object_id oid = {
-        .inode = cur_op->req.rw.inode,
-        // oid.stripe = starting offset of the parity stripe, so it can be mapped back to the PG
-        .stripe = (cur_op->req.rw.offset / pg_stripe_size) * pg_stripe_size +
-            ((cur_op->req.rw.offset % pg_stripe_size) / pg_block_size) * pg_block_size
-    };
     if ((cur_op->req.rw.offset + cur_op->req.rw.len) > (oid.stripe + pg_block_size) ||
         (cur_op->req.rw.offset % bs_disk_alignment) != 0 ||
         (cur_op->req.rw.len % bs_disk_alignment) != 0)
@@ -37,14 +40,16 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
         finish_op(cur_op, -EINVAL);
         return false;
     }
-    osd_primary_op_data_t *op_data = (osd_primary_op_data_t*)calloc(
+    osd_primary_op_data_t *op_data = (osd_primary_op_data_t*)calloc_or_die(
-        sizeof(osd_primary_op_data_t) + sizeof(osd_rmw_stripe_t) * pg_it->second.pg_size, 1
+        1, sizeof(osd_primary_op_data_t) + sizeof(osd_rmw_stripe_t) * (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pg_it->second.pg_size)
     );
     op_data->pg_num = pg_num;
     op_data->oid = oid;
     op_data->stripes = ((osd_rmw_stripe_t*)(op_data+1));
+    op_data->scheme = pool_cfg.scheme;
     cur_op->op_data = op_data;
-    split_stripes(pg_it->second.pg_minsize, bs_block_size, (uint32_t)(cur_op->req.rw.offset - oid.stripe), cur_op->req.rw.len, op_data->stripes);
+    split_stripes((pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pg_it->second.pg_minsize),
+        bs_block_size, (uint32_t)(cur_op->req.rw.offset - oid.stripe), cur_op->req.rw.len, op_data->stripes);
     pg_it->second.inflight++;
     return true;
 }
@@ -88,8 +93,8 @@ void osd_t::continue_primary_read(osd_op_t *cur_op)
     if (op_data->st == 1)      goto resume_1;
     else if (op_data->st == 2) goto resume_2;
     {
-        auto & pg = pgs[op_data->pg_num];
+        auto & pg = pgs[{ .pool_id = INODE_POOL(op_data->oid.inode), .pg_num = op_data->pg_num }];
-        for (int role = 0; role < pg.pg_minsize; role++)
+        for (int role = 0; role < (op_data->scheme == POOL_SCHEME_REPLICATED ? 1 : pg.pg_minsize); role++)
         {
             op_data->stripes[role].read_start = op_data->stripes[role].req_start;
             op_data->stripes[role].read_end = op_data->stripes[role].req_end;
@@ -97,12 +102,13 @@ void osd_t::continue_primary_read(osd_op_t *cur_op)
         // Determine version
         auto vo_it = pg.ver_override.find(op_data->oid);
         op_data->target_ver = vo_it != pg.ver_override.end() ? vo_it->second : UINT64_MAX;
-        if (pg.state == PG_ACTIVE)
+        if (pg.state == PG_ACTIVE || op_data->scheme == POOL_SCHEME_REPLICATED)
         {
             // Fast happy-path
-            cur_op->buf = alloc_read_buffer(op_data->stripes, pg.pg_minsize, 0);
+            cur_op->buf = alloc_read_buffer(op_data->stripes,
-            submit_primary_subops(SUBMIT_READ, pg.pg_minsize, pg.cur_set.data(), cur_op);
+                (op_data->scheme == POOL_SCHEME_REPLICATED ? 1 : pg.pg_minsize), 0);
-            cur_op->send_list.push_back(cur_op->buf, cur_op->req.rw.len);
+            submit_primary_subops(SUBMIT_READ, op_data->target_ver,
+                (op_data->scheme == POOL_SCHEME_REPLICATED ? pg.pg_size : pg.pg_minsize), pg.cur_set.data(), cur_op);
             op_data->st = 1;
         }
         else
@@ -119,7 +125,7 @@ void osd_t::continue_primary_read(osd_op_t *cur_op)
             op_data->pg_size = pg.pg_size;
             op_data->degraded = 1;
             cur_op->buf = alloc_read_buffer(op_data->stripes, pg.pg_size, 0);
-            submit_primary_subops(SUBMIT_READ, pg.pg_size, cur_set, cur_op);
+            submit_primary_subops(SUBMIT_READ, op_data->target_ver, pg.pg_size, cur_set, cur_op);
             op_data->st = 1;
         }
     }
@@ -140,18 +146,22 @@ resume_2:
         {
             if (stripes[role].read_end != 0 && stripes[role].missing)
             {
-                reconstruct_stripe(stripes, op_data->pg_size, role);
+                reconstruct_stripe_xor(stripes, op_data->pg_size, role);
             }
             if (stripes[role].req_end != 0)
             {
                 // Send buffer in parts to avoid copying
-                cur_op->send_list.push_back(
+                cur_op->iov.push_back(
                     stripes[role].read_buf + (stripes[role].req_start - stripes[role].read_start),
                     stripes[role].req_end - stripes[role].req_start
                 );
             }
         }
     }
+    else
+    {
+        cur_op->iov.push_back(cur_op->buf, cur_op->req.rw.len);
+    }
     finish_op(cur_op, cur_op->req.rw.len);
 }
 
@@ -189,7 +199,7 @@ void osd_t::continue_primary_write(osd_op_t *cur_op)
         return;
     }
     osd_primary_op_data_t *op_data = cur_op->op_data;
-    auto & pg = pgs[op_data->pg_num];
+    auto & pg = pgs[{ .pool_id = INODE_POOL(op_data->oid.inode), .pg_num = op_data->pg_num }];
     if (op_data->st == 1)      goto resume_1;
     else if (op_data->st == 2) goto resume_2;
     else if (op_data->st == 3) goto resume_3;
@@ -198,6 +208,8 @@ void osd_t::continue_primary_write(osd_op_t *cur_op)
     else if (op_data->st == 6) goto resume_6;
     else if (op_data->st == 7) goto resume_7;
     else if (op_data->st == 8) goto resume_8;
+    else if (op_data->st == 9) goto resume_9;
+    else if (op_data->st == 10) goto resume_10;
     assert(op_data->st == 0);
     if (!check_write_queue(cur_op, pg))
     {
@@ -206,34 +218,98 @@ void osd_t::continue_primary_write(osd_op_t *cur_op)
 resume_1:
     // Determine blocks to read and write
     // Missing chunks are allowed to be overwritten even in incomplete objects
-    // FIXME: Allow to do small writes to the old (degraded/misplaced) OSD set for the lower performance impact
+    // FIXME: Allow to do small writes to the old (degraded/misplaced) OSD set for lower performance impact
     op_data->prev_set = get_object_osd_set(pg, op_data->oid, pg.cur_set.data(), &op_data->object_state);
-    cur_op->rmw_buf = calc_rmw(cur_op->buf, op_data->stripes, op_data->prev_set,
+    if (op_data->scheme == POOL_SCHEME_REPLICATED)
-        pg.pg_size, pg.pg_minsize, pg.pg_cursize, pg.cur_set.data(), bs_block_size);
+    {
+        // Simplified algorithm
+        op_data->stripes[0].write_start = op_data->stripes[0].req_start;
+        op_data->stripes[0].write_end = op_data->stripes[0].req_end;
+        op_data->stripes[0].write_buf = cur_op->buf;
+        if (pg.cur_set.data() != op_data->prev_set && (op_data->stripes[0].write_start != 0 ||
+            op_data->stripes[0].write_end != bs_block_size))
+        {
+            // Object is degraded/misplaced and will be moved to <write_osd_set>
+            op_data->stripes[0].read_start = 0;
+            op_data->stripes[0].read_end = bs_block_size;
+            cur_op->rmw_buf = op_data->stripes[0].read_buf = memalign_or_die(MEM_ALIGNMENT, bs_block_size);
+        }
+    }
+    else
+    {
+        cur_op->rmw_buf = calc_rmw(cur_op->buf, op_data->stripes, op_data->prev_set,
+            pg.pg_size, pg.pg_minsize, pg.pg_cursize, pg.cur_set.data(), bs_block_size);
+    }
     // Read required blocks
-    submit_primary_subops(SUBMIT_RMW_READ, pg.pg_size, op_data->prev_set, cur_op);
+    submit_primary_subops(SUBMIT_RMW_READ, UINT64_MAX, pg.pg_size, op_data->prev_set, cur_op);
 resume_2:
     op_data->st = 2;
     return;
 resume_3:
     if (op_data->errors > 0)
     {
-        pg_cancel_write_queue(pg, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
+        pg_cancel_write_queue(pg, cur_op, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
         return;
     }
     // Save version override for parallel reads
     pg.ver_override[op_data->oid] = op_data->fact_ver;
-    // Recover missing stripes, calculate parity
+    if (op_data->scheme == POOL_SCHEME_REPLICATED)
-    calc_rmw_parity(op_data->stripes, pg.pg_size, op_data->prev_set, pg.cur_set.data(), bs_block_size);
+    {
+        // Only (possibly) copy new data from the request into the recovery buffer
+        if (pg.cur_set.data() != op_data->prev_set && (op_data->stripes[0].write_start != 0 ||
+            op_data->stripes[0].write_end != bs_block_size))
+        {
+            memcpy(
+                op_data->stripes[0].read_buf + op_data->stripes[0].req_start,
+                op_data->stripes[0].write_buf,
+                op_data->stripes[0].req_end - op_data->stripes[0].req_start
+            );
+            op_data->stripes[0].write_buf = op_data->stripes[0].read_buf;
+            op_data->stripes[0].write_start = 0;
+            op_data->stripes[0].write_end = bs_block_size;
+        }
+    }
+    else
+    {
+        // Recover missing stripes, calculate parity
+        calc_rmw_parity_xor(op_data->stripes, pg.pg_size, op_data->prev_set, pg.cur_set.data(), bs_block_size);
+    }
     // Send writes
-    submit_primary_subops(SUBMIT_WRITE, pg.pg_size, pg.cur_set.data(), cur_op);
+    if ((op_data->fact_ver >> (64-PG_EPOCH_BITS)) < pg.epoch)
+    {
+        op_data->target_ver = ((uint64_t)pg.epoch << (64-PG_EPOCH_BITS)) | 1;
+    }
+    else
+    {
+        if ((op_data->fact_ver & (1ul<<(64-PG_EPOCH_BITS) - 1)) == (1ul<<(64-PG_EPOCH_BITS) - 1))
+        {
+            assert(pg.epoch != ((1ul << PG_EPOCH_BITS)-1));
+            pg.epoch++;
+        }
+        op_data->target_ver = op_data->fact_ver + 1;
+    }
+    if (pg.epoch > pg.reported_epoch)
+    {
+        // Report newer epoch before writing
+        // FIXME: We may report only one PG state here...
+        this->pg_state_dirty.insert({ .pool_id = pg.pool_id, .pg_num = pg.pg_num });
+        pg.history_changed = true;
+        report_pg_states();
+resume_10:
+        if (pg.epoch > pg.reported_epoch)
+        {
+            op_data->st = 10;
+            return;
+        }
+    }
+    submit_primary_subops(SUBMIT_WRITE, op_data->target_ver, pg.pg_size, pg.cur_set.data(), cur_op);
 resume_4:
     op_data->st = 4;
     return;
 resume_5:
     if (op_data->errors > 0)
     {
-        pg_cancel_write_queue(pg, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
+        pg_cancel_write_queue(pg, cur_op, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
         return;
     }
     if (op_data->fact_ver == 1)
@@ -252,7 +328,7 @@ resume_5:
                 recovery_stat_count[0][recovery_type]++;
                 recovery_stat_bytes[0][recovery_type] = 0;
             }
-            for (int role = 0; role < pg.pg_size; role++)
+            for (int role = 0; role < (op_data->scheme == POOL_SCHEME_REPLICATED ? 1 : pg.pg_size); role++)
             {
                 recovery_stat_bytes[0][recovery_type] += op_data->stripes[role].write_end - op_data->stripes[role].write_start;
             }
@@ -260,15 +336,16 @@ resume_5:
         if (op_data->object_state->state & OBJ_MISPLACED)
         {
             // Remove extra chunks
-            submit_primary_del_subops(cur_op, pg.cur_set.data(), op_data->object_state->osd_set);
+            submit_primary_del_subops(cur_op, pg.cur_set.data(), pg.pg_size, op_data->object_state->osd_set);
             if (op_data->n_subops > 0)
             {
+resume_8:
                 op_data->st = 8;
                 return;
-resume_8:
+resume_9:
                 if (op_data->errors > 0)
                 {
-                    pg_cancel_write_queue(pg, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
+                    pg_cancel_write_queue(pg, cur_op, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
                     return;
                 }
             }
@@ -282,7 +359,7 @@ resume_8:
     // FIXME: Check for immediate_commit == IMMEDIATE_SMALL
 resume_6:
 resume_7:
-    if (!finalize_primary_write(cur_op, pg, pg.cur_loc_set, 6))
+    if (!remember_unstable_write(cur_op, pg, pg.cur_loc_set, 6))
     {
         return;
     }
@@ -291,17 +368,19 @@ resume_7:
     // Continue other write operations to the same object
     auto next_it = pg.write_queue.find(oid);
     auto this_it = next_it;
-    next_it++;
+    if (this_it != pg.write_queue.end() && this_it->second == cur_op)
-    pg.write_queue.erase(this_it);
-    if (next_it != pg.write_queue.end() &&
-        next_it->first == oid)
     {
-        osd_op_t *next_op = next_it->second;
+        next_it++;
-        continue_primary_write(next_op);
+        pg.write_queue.erase(this_it);
+        if (next_it != pg.write_queue.end() && next_it->first == oid)
+        {
+            osd_op_t *next_op = next_it->second;
+            continue_primary_write(next_op);
+        }
     }
 }
 
-bool osd_t::finalize_primary_write(osd_op_t *cur_op, pg_t & pg, pg_osd_set_t & loc_set, int base_state)
+bool osd_t::remember_unstable_write(osd_op_t *cur_op, pg_t & pg, pg_osd_set_t & loc_set, int base_state)
 {
     osd_primary_op_data_t *op_data = cur_op->op_data;
     if (op_data->st == base_state)
@@ -314,60 +393,76 @@ bool osd_t::finalize_primary_write(osd_op_t *cur_op, pg_t & pg, pg_osd_set_t & l
     }
     if (immediate_commit == IMMEDIATE_ALL)
     {
-        op_data->unstable_write_osds = new std::vector<unstable_osd_num_t>();
+        if (op_data->scheme != POOL_SCHEME_REPLICATED)
-        op_data->unstable_writes = new obj_ver_id[loc_set.size()];
         {
-            int last_start = 0;
+            // Send STABILIZE ops immediately
-            for (auto & chunk: loc_set)
+            op_data->unstable_write_osds = new std::vector<unstable_osd_num_t>();
+            op_data->unstable_writes = new obj_ver_id[loc_set.size()];
             {
-                op_data->unstable_writes[last_start] = (obj_ver_id){
+                int last_start = 0;
-                    .oid = {
+                for (auto & chunk: loc_set)
-                        .inode = op_data->oid.inode,
+                {
-                        .stripe = op_data->oid.stripe | chunk.role,
+                    op_data->unstable_writes[last_start] = (obj_ver_id){
-                    },
+                        .oid = {
-                    .version = op_data->fact_ver,
+                            .inode = op_data->oid.inode,
-                };
+                            .stripe = op_data->oid.stripe | chunk.role,
-                op_data->unstable_write_osds->push_back((unstable_osd_num_t){
+                        },
-                    .osd_num = chunk.osd_num,
+                        .version = op_data->fact_ver,
-                    .start = last_start,
+                    };
-                    .len = 1,
+                    op_data->unstable_write_osds->push_back((unstable_osd_num_t){
-                });
+                        .osd_num = chunk.osd_num,
-                last_start++;
+                        .start = last_start,
+                        .len = 1,
+                    });
+                    last_start++;
+                }
             }
-        }
+            submit_primary_stab_subops(cur_op);
-        submit_primary_stab_subops(cur_op);
 resume_6:
-        op_data->st = 6;
+            op_data->st = 6;
-        return false;
-resume_7:
-        // FIXME: Free those in the destructor?
-        delete op_data->unstable_write_osds;
-        delete[] op_data->unstable_writes;
-        op_data->unstable_writes = NULL;
-        op_data->unstable_write_osds = NULL;
-        if (op_data->errors > 0)
-        {
-            pg_cancel_write_queue(pg, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
             return false;
+resume_7:
+            // FIXME: Free those in the destructor?
+            delete op_data->unstable_write_osds;
+            delete[] op_data->unstable_writes;
+            op_data->unstable_writes = NULL;
+            op_data->unstable_write_osds = NULL;
+            if (op_data->errors > 0)
+            {
+                pg_cancel_write_queue(pg, cur_op, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
+                return false;
+            }
         }
     }
     else
     {
-        // Remember version as unstable
+        if (op_data->scheme != POOL_SCHEME_REPLICATED)
-        for (auto & chunk: loc_set)
         {
-            this->unstable_writes[(osd_object_id_t){
+            // Remember version as unstable for EC/XOR
-                .osd_num = chunk.osd_num,
+            for (auto & chunk: loc_set)
-                .oid = {
+            {
-                    .inode = op_data->oid.inode,
+                this->dirty_osds.insert(chunk.osd_num);
-                    .stripe = op_data->oid.stripe | chunk.role,
+                this->unstable_writes[(osd_object_id_t){
-                },
+                    .osd_num = chunk.osd_num,
-            }] = op_data->fact_ver;
+                    .oid = {
+                        .inode = op_data->oid.inode,
+                        .stripe = op_data->oid.stripe | chunk.role,
+                    },
+                }] = op_data->fact_ver;
+            }
+        }
+        else
+        {
+            // Only remember to sync OSDs for replicated pools
+            for (auto & chunk: loc_set)
+            {
+                this->dirty_osds.insert(chunk.osd_num);
+            }
         }
         // Remember PG as dirty to drop the connection when PG goes offline
         // (this is required because of the "lazy sync")
-        c_cli.clients[cur_op->peer_fd].dirty_pgs.insert(op_data->pg_num);
+        c_cli.clients[cur_op->peer_fd].dirty_pgs.insert({ .pool_id = pg.pool_id, .pg_num = pg.pg_num });
-        dirty_pgs.insert(op_data->pg_num);
+        dirty_pgs.insert({ .pool_id = pg.pool_id, .pg_num = pg.pg_num });
     }
     return true;
 }
@@ -377,7 +472,7 @@ void osd_t::continue_primary_sync(osd_op_t *cur_op)
 {
     if (!cur_op->op_data)
     {
-        cur_op->op_data = (osd_primary_op_data_t*)calloc(sizeof(osd_primary_op_data_t), 1);
+        cur_op->op_data = (osd_primary_op_data_t*)calloc_or_die(1, sizeof(osd_primary_op_data_t));
     }
     osd_primary_op_data_t *op_data = cur_op->op_data;
     if (op_data->st == 1)      goto resume_1;
@@ -401,7 +496,7 @@ resume_1:
         syncs_in_progress.push_back(cur_op);
     }
 resume_2:
-    if (unstable_writes.size() == 0)
+    if (dirty_osds.size() == 0)
     {
         // Nothing to sync
         goto finish;
@@ -409,11 +504,10 @@ resume_2:
     // Save and clear unstable_writes
     // In theory it is possible to do in on a per-client basis, but this seems to be an unnecessary complication
     // It would be cool not to copy these here at all, but someone has to deduplicate them by object IDs anyway
+    if (unstable_writes.size() > 0)
     {
         op_data->unstable_write_osds = new std::vector<unstable_osd_num_t>();
         op_data->unstable_writes = new obj_ver_id[this->unstable_writes.size()];
-        op_data->dirty_pgs = new pg_num_t[dirty_pgs.size()];
-        op_data->dirty_pg_count = dirty_pgs.size();
         osd_num_t last_osd = 0;
         int last_start = 0, last_end = 0;
         for (auto it = this->unstable_writes.begin(); it != this->unstable_writes.end(); it++)
@@ -445,6 +539,14 @@ resume_2:
                 .len = last_end - last_start,
             });
         }
+        this->unstable_writes.clear();
+    }
+    {
+        void *dirty_buf = malloc_or_die(sizeof(pool_pg_num_t)*dirty_pgs.size() + sizeof(osd_num_t)*dirty_osds.size());
+        op_data->dirty_pgs = (pool_pg_num_t*)dirty_buf;
+        op_data->dirty_osds = (osd_num_t*)(dirty_buf + sizeof(pool_pg_num_t)*dirty_pgs.size());
+        op_data->dirty_pg_count = dirty_pgs.size();
+        op_data->dirty_osd_count = dirty_osds.size();
         int dpg = 0;
         for (auto dirty_pg_num: dirty_pgs)
         {
@@ -452,7 +554,12 @@ resume_2:
             op_data->dirty_pgs[dpg++] = dirty_pg_num;
         }
         dirty_pgs.clear();
-        this->unstable_writes.clear();
+        dpg = 0;
+        for (auto osd_num: dirty_osds)
+        {
+            op_data->dirty_osds[dpg++] = osd_num;
+        }
+        dirty_osds.clear();
     }
     if (immediate_commit != IMMEDIATE_ALL)
     {
@@ -467,30 +574,48 @@ resume_4:
             goto resume_6;
         }
     }
-    // Stabilize version sets
+    if (op_data->unstable_writes)
-    submit_primary_stab_subops(cur_op);
+    {
+        // Stabilize version sets, if any
+        submit_primary_stab_subops(cur_op);
 resume_5:
-    op_data->st = 5;
+        op_data->st = 5;
-    return;
+        return;
+    }
 resume_6:
     if (op_data->errors > 0)
     {
-        // Return objects back into the unstable write set
+        // Return PGs and OSDs back into their dirty sets
-        for (auto unstable_osd: *(op_data->unstable_write_osds))
+        for (int i = 0; i < op_data->dirty_pg_count; i++)
         {
-            for (int i = 0; i < unstable_osd.len; i++)
+            dirty_pgs.insert(op_data->dirty_pgs[i]);
+        }
+        for (int i = 0; i < op_data->dirty_osd_count; i++)
+        {
+            dirty_osds.insert(op_data->dirty_osds[i]);
+        }
+        if (op_data->unstable_writes)
+        {
+            // Return objects back into the unstable write set
+            for (auto unstable_osd: *(op_data->unstable_write_osds))
             {
-                // Except those from peered PGs
+                for (int i = 0; i < unstable_osd.len; i++)
-                auto & w = op_data->unstable_writes[i];
-                pg_num_t wpg = map_to_pg(w.oid);
-                if (pgs[wpg].state & PG_ACTIVE)
                 {
-                    uint64_t & dest = this->unstable_writes[(osd_object_id_t){
+                    // Except those from peered PGs
-                        .osd_num = unstable_osd.osd_num,
+                    auto & w = op_data->unstable_writes[i];
-                        .oid = w.oid,
+                    pool_pg_num_t wpg = {
-                    }];
+                        .pool_id = INODE_POOL(w.oid.inode),
-                    dest = dest < w.version ? w.version : dest;
+                        .pg_num = map_to_pg(w.oid, st_cli.pool_config.at(INODE_POOL(w.oid.inode)).pg_stripe_size),
-                    dirty_pgs.insert(wpg);
+                    };
+                    if (pgs[wpg].state & PG_ACTIVE)
+                    {
+                        uint64_t & dest = this->unstable_writes[(osd_object_id_t){
+                            .osd_num = unstable_osd.osd_num,
+                            .oid = w.oid,
+                        }];
+                        dest = dest < w.version ? w.version : dest;
+                        dirty_pgs.insert(wpg);
+                    }
                 }
             }
         }
@@ -505,11 +630,16 @@ resume_6:
         }
     }
     // FIXME: Free those in the destructor?
-    delete op_data->dirty_pgs;
+    free(op_data->dirty_pgs);
-    delete op_data->unstable_write_osds;
+    op_data->dirty_pgs = NULL;
-    delete[] op_data->unstable_writes;
+    op_data->dirty_osds = NULL;
-    op_data->unstable_writes = NULL;
+    if (op_data->unstable_writes)
-    op_data->unstable_write_osds = NULL;
+    {
+        delete op_data->unstable_write_osds;
+        delete[] op_data->unstable_writes;
+        op_data->unstable_writes = NULL;
+        op_data->unstable_write_osds = NULL;
+    }
     if (op_data->errors > 0)
     {
         finish_op(cur_op, op_data->epipe > 0 ? -EPIPE : -EIO);
@@ -588,14 +718,12 @@ void osd_t::continue_primary_del(osd_op_t *cur_op)
         return;
     }
     osd_primary_op_data_t *op_data = cur_op->op_data;
-    auto & pg = pgs[op_data->pg_num];
+    auto & pg = pgs[{ .pool_id = INODE_POOL(op_data->oid.inode), .pg_num = op_data->pg_num }];
     if (op_data->st == 1)      goto resume_1;
     else if (op_data->st == 2) goto resume_2;
     else if (op_data->st == 3) goto resume_3;
     else if (op_data->st == 4) goto resume_4;
     else if (op_data->st == 5) goto resume_5;
-    else if (op_data->st == 6) goto resume_6;
-    else if (op_data->st == 7) goto resume_7;
     assert(op_data->st == 0);
     // Delete is forbidden even in active PGs if they're also degraded or have previous dead OSDs
     if (pg.state & (PG_DEGRADED | PG_LEFT_ON_DEAD))
@@ -611,38 +739,32 @@ resume_1:
     // Determine which OSDs contain this object and delete it
     op_data->prev_set = get_object_osd_set(pg, op_data->oid, pg.cur_set.data(), &op_data->object_state);
     // Submit 1 read to determine the actual version number
-    submit_primary_subops(SUBMIT_RMW_READ, pg.pg_size, op_data->prev_set, cur_op);
+    submit_primary_subops(SUBMIT_RMW_READ, UINT64_MAX, pg.pg_size, op_data->prev_set, cur_op);
 resume_2:
     op_data->st = 2;
     return;
 resume_3:
     if (op_data->errors > 0)
     {
-        pg_cancel_write_queue(pg, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
+        pg_cancel_write_queue(pg, cur_op, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
         return;
     }
     // Save version override for parallel reads
     pg.ver_override[op_data->oid] = op_data->fact_ver;
     // Submit deletes
     op_data->fact_ver++;
-    submit_primary_del_subops(cur_op, NULL, op_data->object_state ? op_data->object_state->osd_set : pg.cur_loc_set);
+    submit_primary_del_subops(cur_op, NULL, 0, op_data->object_state ? op_data->object_state->osd_set : pg.cur_loc_set);
 resume_4:
     op_data->st = 4;
     return;
 resume_5:
     if (op_data->errors > 0)
     {
-        pg_cancel_write_queue(pg, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
+        pg_cancel_write_queue(pg, cur_op, op_data->oid, op_data->epipe > 0 ? -EPIPE : -EIO);
         return;
     }
     // Remove version override
     pg.ver_override.erase(op_data->oid);
-resume_6:
-resume_7:
-    if (!finalize_primary_write(cur_op, pg, op_data->object_state ? op_data->object_state->osd_set : pg.cur_loc_set, 6))
-    {
-        return;
-    }
     // Adjust PG stats after "instant stabilize", because we need object_state above
     if (!op_data->object_state)
     {
@@ -658,12 +780,15 @@ resume_7:
     // Continue other write operations to the same object
     auto next_it = pg.write_queue.find(oid);
     auto this_it = next_it;
-    next_it++;
+    if (this_it != pg.write_queue.end() && this_it->second == cur_op)
-    pg.write_queue.erase(this_it);
-    if (next_it != pg.write_queue.end() &&
-        next_it->first == oid)
     {
-        osd_op_t *next_op = next_it->second;
+        next_it++;
-        continue_primary_write(next_op);
+        pg.write_queue.erase(this_it);
+        if (next_it != pg.write_queue.end() &&
+            next_it->first == oid)
+        {
+            osd_op_t *next_op = next_it->second;
+            continue_primary_write(next_op);
+        }
     }
 }

osd_primary.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #pragma once
 
 #include "osd.h"
@@ -20,6 +23,7 @@ struct osd_primary_op_data_t
     object_id oid;
     uint64_t target_ver;
     uint64_t fact_ver = 0;
+    uint64_t scheme = 0;
     int n_subops = 0, done = 0, errors = 0, epipe = 0;
     int degraded = 0, pg_size, pg_minsize;
     osd_rmw_stripe_t *stripes;
@@ -29,7 +33,9 @@ struct osd_primary_op_data_t
 
     // for sync. oops, requires freeing
     std::vector<unstable_osd_num_t> *unstable_write_osds = NULL;
-    pg_num_t *dirty_pgs = NULL;
+    pool_pg_num_t *dirty_pgs = NULL;
     int dirty_pg_count = 0;
+    osd_num_t *dirty_osds = NULL;
+    int dirty_osd_count = 0;
     obj_ver_id *unstable_writes = NULL;
 };

osd_primary_subops.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "osd_primary.h"
 
 void osd_t::autosync()
@@ -33,15 +36,24 @@ void osd_t::autosync()
 void osd_t::finish_op(osd_op_t *cur_op, int retval)
 {
     inflight_ops--;
-    if (cur_op->op_data && cur_op->op_data->pg_num > 0)
+    if (cur_op->op_data)
     {
-        auto & pg = pgs[cur_op->op_data->pg_num];
+        if (cur_op->op_data->pg_num > 0)
-        pg.inflight--;
-        assert(pg.inflight >= 0);
-        if ((pg.state & PG_STOPPING) && pg.inflight == 0 && !pg.flush_batch)
         {
-            finish_stop_pg(pg);
+            auto & pg = pgs[{ .pool_id = INODE_POOL(cur_op->op_data->oid.inode), .pg_num = cur_op->op_data->pg_num }];
+            pg.inflight--;
+            assert(pg.inflight >= 0);
+            if ((pg.state & PG_STOPPING) && pg.inflight == 0 && !pg.flush_batch)
+            {
+                finish_stop_pg(pg);
+            }
         }
+        assert(!cur_op->op_data->subops);
+        assert(!cur_op->op_data->unstable_write_osds);
+        assert(!cur_op->op_data->unstable_writes);
+        assert(!cur_op->op_data->dirty_pgs);
+        free(cur_op->op_data);
+        cur_op->op_data = NULL;
     }
     if (!cur_op->peer_fd)
     {
@@ -67,11 +79,12 @@ void osd_t::finish_op(osd_op_t *cur_op, int retval)
     }
 }
 
-void osd_t::submit_primary_subops(int submit_type, int pg_size, const uint64_t* osd_set, osd_op_t *cur_op)
+void osd_t::submit_primary_subops(int submit_type, uint64_t op_version, int pg_size, const uint64_t* osd_set, osd_op_t *cur_op)
 {
-    bool w = submit_type == SUBMIT_WRITE;
+    bool wr = submit_type == SUBMIT_WRITE;
     osd_primary_op_data_t *op_data = cur_op->op_data;
     osd_rmw_stripe_t *stripes = op_data->stripes;
+    bool rep = op_data->scheme == POOL_SCHEME_REPLICATED;
     // Allocate subops
     int n_subops = 0, zero_read = -1;
     for (int role = 0; role < pg_size; role++)
@@ -80,12 +93,12 @@ void osd_t::submit_primary_subops(int submit_type, int pg_size, const uint64_t*
         {
             zero_read = role;
         }
-        if (osd_set[role] != 0 && (w || stripes[role].read_end != 0))
+        if (osd_set[role] != 0 && (wr || !rep && stripes[role].read_end != 0))
         {
             n_subops++;
         }
     }
-    if (!n_subops && submit_type == SUBMIT_RMW_READ)
+    if (!n_subops && (submit_type == SUBMIT_RMW_READ || rep))
     {
         n_subops = 1;
     }
@@ -93,7 +106,6 @@ void osd_t::submit_primary_subops(int submit_type, int pg_size, const uint64_t*
     {
         zero_read = -1;
     }
-    uint64_t op_version = w ? op_data->fact_ver+1 : (submit_type == SUBMIT_RMW_READ ? UINT64_MAX : op_data->target_ver);
     osd_op_t *subops = new osd_op_t[n_subops];
     op_data->fact_ver = 0;
     op_data->done = op_data->errors = 0;
@@ -103,68 +115,86 @@ void osd_t::submit_primary_subops(int submit_type, int pg_size, const uint64_t*
     for (int role = 0; role < pg_size; role++)
     {
         // We always submit zero-length writes to all replicas, even if the stripe is not modified
-        if (!(w || stripes[role].read_end != 0 || zero_read == role))
+        if (!(wr || !rep && stripes[role].read_end != 0 || zero_read == role))
         {
             continue;
         }
         osd_num_t role_osd_num = osd_set[role];
         if (role_osd_num != 0)
         {
+            int stripe_num = rep ? 0 : role;
             if (role_osd_num == this->osd_num)
             {
                 clock_gettime(CLOCK_REALTIME, &subops[i].tv_begin);
                 subops[i].op_type = (uint64_t)cur_op;
                 subops[i].bs_op = new blockstore_op_t({
-                    .opcode = (uint64_t)(w ? BS_OP_WRITE : BS_OP_READ),
+                    .opcode = (uint64_t)(wr ? (rep ? BS_OP_WRITE_STABLE : BS_OP_WRITE) : BS_OP_READ),
                     .callback = [subop = &subops[i], this](blockstore_op_t *bs_subop)
                     {
                         handle_primary_bs_subop(subop);
                     },
                     .oid = {
                         .inode = op_data->oid.inode,
-                        .stripe = op_data->oid.stripe | role,
+                        .stripe = op_data->oid.stripe | stripe_num,
                     },
                     .version = op_version,
-                    .offset = w ? stripes[role].write_start : stripes[role].read_start,
+                    .offset = wr ? stripes[stripe_num].write_start : stripes[stripe_num].read_start,
-                    .len = w ? stripes[role].write_end - stripes[role].write_start : stripes[role].read_end - stripes[role].read_start,
+                    .len = wr ? stripes[stripe_num].write_end - stripes[stripe_num].write_start : stripes[stripe_num].read_end - stripes[stripe_num].read_start,
-                    .buf = w ? stripes[role].write_buf : stripes[role].read_buf,
+                    .buf = wr ? stripes[stripe_num].write_buf : stripes[stripe_num].read_buf,
                 });
+#ifdef OSD_DEBUG
+                printf(
+                    "Submit %s to local: %lx:%lx v%lu %u-%u\n", wr ? "write" : "read",
+                    op_data->oid.inode, op_data->oid.stripe | stripe_num, op_version,
+                    subops[i].bs_op->offset, subops[i].bs_op->len
+                );
+#endif
                 bs->enqueue_op(subops[i].bs_op);
             }
             else
             {
                 subops[i].op_type = OSD_OP_OUT;
-                subops[i].send_list.push_back(subops[i].req.buf, OSD_PACKET_SIZE);
                 subops[i].peer_fd = c_cli.osd_peer_fds.at(role_osd_num);
                 subops[i].req.sec_rw = {
                     .header = {
                         .magic = SECONDARY_OSD_OP_MAGIC,
                         .id = c_cli.next_subop_id++,
-                        .opcode = (uint64_t)(w ? OSD_OP_SECONDARY_WRITE : OSD_OP_SECONDARY_READ),
+                        .opcode = (uint64_t)(wr ? (rep ? OSD_OP_SEC_WRITE_STABLE : OSD_OP_SEC_WRITE) : OSD_OP_SEC_READ),
                     },
                     .oid = {
                         .inode = op_data->oid.inode,
-                        .stripe = op_data->oid.stripe | role,
+                        .stripe = op_data->oid.stripe | stripe_num,
                     },
                     .version = op_version,
-                    .offset = w ? stripes[role].write_start : stripes[role].read_start,
+                    .offset = wr ? stripes[stripe_num].write_start : stripes[stripe_num].read_start,
-                    .len = w ? stripes[role].write_end - stripes[role].write_start : stripes[role].read_end - stripes[role].read_start,
+                    .len = wr ? stripes[stripe_num].write_end - stripes[stripe_num].write_start : stripes[stripe_num].read_end - stripes[stripe_num].read_start,
                 };
-                subops[i].buf = w ? stripes[role].write_buf : stripes[role].read_buf;
+#ifdef OSD_DEBUG
-                if (w && stripes[role].write_end > 0)
+                printf(
+                    "Submit %s to osd %lu: %lx:%lx v%lu %u-%u\n", wr ? "write" : "read", role_osd_num,
+                    op_data->oid.inode, op_data->oid.stripe | stripe_num, op_version,
+                    subops[i].req.sec_rw.offset, subops[i].req.sec_rw.len
+                );
+#endif
+                if (wr)
                 {
-                    subops[i].send_list.push_back(stripes[role].write_buf, stripes[role].write_end - stripes[role].write_start);
+                    if (stripes[stripe_num].write_end > stripes[stripe_num].write_start)
+                    {
+                        subops[i].iov.push_back(stripes[stripe_num].write_buf, stripes[stripe_num].write_end - stripes[stripe_num].write_start);
+                    }
+                }
+                else
+                {
+                    if (stripes[stripe_num].read_end > stripes[stripe_num].read_start)
+                    {
+                        subops[i].iov.push_back(stripes[stripe_num].read_buf, stripes[stripe_num].read_end - stripes[stripe_num].read_start);
+                    }
                 }
                 subops[i].callback = [cur_op, this](osd_op_t *subop)
                 {
-                    int fail_fd = subop->req.hdr.opcode == OSD_OP_SECONDARY_WRITE &&
+                    int fail_fd = subop->req.hdr.opcode == OSD_OP_SEC_WRITE &&
                         subop->reply.hdr.retval != subop->req.sec_rw.len ? subop->peer_fd : -1;
-                    // so it doesn't get freed
+                    handle_primary_subop(subop, cur_op);
-                    subop->buf = NULL;
-                    handle_primary_subop(
-                        subop->req.hdr.opcode, cur_op, subop->reply.hdr.retval,
-                        subop->req.sec_rw.len, subop->reply.sec_rw.version
-                    );
                     if (fail_fd >= 0)
                     {
                         // write operation failed, drop the connection
@@ -180,21 +210,23 @@ void osd_t::submit_primary_subops(int submit_type, int pg_size, const uint64_t*
 
 static uint64_t bs_op_to_osd_op[] = {
     0,
-    OSD_OP_SECONDARY_READ,      // BS_OP_READ
+    OSD_OP_SEC_READ,            // BS_OP_READ = 1
-    OSD_OP_SECONDARY_WRITE,     // BS_OP_WRITE
+    OSD_OP_SEC_WRITE,           // BS_OP_WRITE = 2
-    OSD_OP_SECONDARY_SYNC,      // BS_OP_SYNC
+    OSD_OP_SEC_WRITE_STABLE,    // BS_OP_WRITE_STABLE = 3
-    OSD_OP_SECONDARY_STABILIZE, // BS_OP_STABLE
+    OSD_OP_SEC_SYNC,            // BS_OP_SYNC = 4
-    OSD_OP_SECONDARY_DELETE,    // BS_OP_DELETE
+    OSD_OP_SEC_STABILIZE,       // BS_OP_STABLE = 5
-    OSD_OP_SECONDARY_LIST,      // BS_OP_LIST
+    OSD_OP_SEC_DELETE,          // BS_OP_DELETE = 6
-    OSD_OP_SECONDARY_ROLLBACK,  // BS_OP_ROLLBACK
+    OSD_OP_SEC_LIST,            // BS_OP_LIST = 7
-    OSD_OP_TEST_SYNC_STAB_ALL,  // BS_OP_SYNC_STAB_ALL
+    OSD_OP_SEC_ROLLBACK,        // BS_OP_ROLLBACK = 8
+    OSD_OP_TEST_SYNC_STAB_ALL,  // BS_OP_SYNC_STAB_ALL = 9
 };
 
 void osd_t::handle_primary_bs_subop(osd_op_t *subop)
 {
     osd_op_t *cur_op = (osd_op_t*)subop->op_type;
     blockstore_op_t *bs_op = subop->bs_op;
-    int expected = bs_op->opcode == BS_OP_READ || bs_op->opcode == BS_OP_WRITE ? bs_op->len : 0;
+    int expected = bs_op->opcode == BS_OP_READ || bs_op->opcode == BS_OP_WRITE
+        || bs_op->opcode == BS_OP_WRITE_STABLE ? bs_op->len : 0;
     if (bs_op->retval != expected && bs_op->opcode != BS_OP_READ)
     {
         // die
@@ -204,12 +236,16 @@ void osd_t::handle_primary_bs_subop(osd_op_t *subop)
         );
     }
     add_bs_subop_stats(subop);
-    uint64_t opcode = bs_op_to_osd_op[bs_op->opcode];
+    subop->req.hdr.opcode = bs_op_to_osd_op[bs_op->opcode];
-    int retval = bs_op->retval;
+    subop->reply.hdr.retval = bs_op->retval;
-    uint64_t version = bs_op->version;
+    if (bs_op->opcode == BS_OP_READ || bs_op->opcode == BS_OP_WRITE || bs_op->opcode == BS_OP_WRITE_STABLE)
+    {
+        subop->req.sec_rw.len = bs_op->len;
+        subop->reply.sec_rw.version = bs_op->version;
+    }
     delete bs_op;
     subop->bs_op = NULL;
-    handle_primary_subop(opcode, cur_op, retval, expected, version);
+    handle_primary_subop(subop, cur_op);
 }
 
 void osd_t::add_bs_subop_stats(osd_op_t *subop)
@@ -229,14 +265,18 @@ void osd_t::add_bs_subop_stats(osd_op_t *subop)
         (tv_end.tv_sec - subop->tv_begin.tv_sec)*1000000 +
         (tv_end.tv_nsec - subop->tv_begin.tv_nsec)/1000
     );
-    if (opcode == OSD_OP_SECONDARY_READ || opcode == OSD_OP_SECONDARY_WRITE)
+    if (opcode == OSD_OP_SEC_READ || opcode == OSD_OP_SEC_WRITE)
     {
         c_cli.stats.op_stat_bytes[opcode] += subop->bs_op->len;
     }
 }
 
-void osd_t::handle_primary_subop(uint64_t opcode, osd_op_t *cur_op, int retval, int expected, uint64_t version)
+void osd_t::handle_primary_subop(osd_op_t *subop, osd_op_t *cur_op)
 {
+    uint64_t opcode = subop->req.hdr.opcode;
+    int retval = subop->reply.hdr.retval;
+    int expected = opcode == OSD_OP_SEC_READ || opcode == OSD_OP_SEC_WRITE
+        || opcode == OSD_OP_SEC_WRITE_STABLE ? subop->req.sec_rw.len : 0;
     osd_primary_op_data_t *op_data = cur_op->op_data;
     if (retval != expected)
     {
@@ -250,8 +290,14 @@ void osd_t::handle_primary_subop(uint64_t opcode, osd_op_t *cur_op, int retval,
     else
     {
         op_data->done++;
-        if (opcode == OSD_OP_SECONDARY_READ || opcode == OSD_OP_SECONDARY_WRITE)
+        if (opcode == OSD_OP_SEC_READ || opcode == OSD_OP_SEC_WRITE || opcode == OSD_OP_SEC_WRITE_STABLE)
         {
+            uint64_t version = subop->reply.sec_rw.version;
+#ifdef OSD_DEBUG
+            uint64_t peer_osd = c_cli.clients.find(subop->peer_fd) != c_cli.clients.end()
+                ? c_cli.clients[subop->peer_fd].osd_num : osd_num;
+            printf("subop %lu from osd %lu: version = %lu\n", opcode, peer_osd, version);
+#endif
             if (op_data->fact_ver != 0 && op_data->fact_ver != version)
             {
                 throw std::runtime_error(
@@ -290,13 +336,44 @@ void osd_t::handle_primary_subop(uint64_t opcode, osd_op_t *cur_op, int retval,
     }
 }
 
-void osd_t::submit_primary_del_subops(osd_op_t *cur_op, uint64_t *cur_set, pg_osd_set_t & loc_set)
+void osd_t::cancel_primary_write(osd_op_t *cur_op)
+{
+    if (cur_op->op_data && cur_op->op_data->subops)
+    {
+        // Primary-write operation is waiting for subops, subops
+        // are sent to peer OSDs, so we can't just throw them away.
+        // Mark them with an extra EPIPE.
+        cur_op->op_data->errors++;
+        cur_op->op_data->epipe++;
+        cur_op->op_data->done--; // Caution: `done` must be signed because may become -1 here
+    }
+    else
+    {
+        finish_op(cur_op, -EPIPE);
+    }
+}
+
+static bool contains_osd(osd_num_t *osd_set, uint64_t size, osd_num_t osd_num)
+{
+    for (uint64_t i = 0; i < size; i++)
+    {
+        if (osd_set[i] == osd_num)
+        {
+            return true;
+        }
+    }
+    return false;
+}
+
+void osd_t::submit_primary_del_subops(osd_op_t *cur_op, osd_num_t *cur_set, uint64_t set_size, pg_osd_set_t & loc_set)
 {
     osd_primary_op_data_t *op_data = cur_op->op_data;
+    bool rep = op_data->scheme == POOL_SCHEME_REPLICATED;
     int extra_chunks = 0;
+    // ordered comparison for EC/XOR, unordered for replicated pools
     for (auto & chunk: loc_set)
     {
-        if (!cur_set || chunk.osd_num != cur_set[chunk.role])
+        if (!cur_set || (rep ? !contains_osd(cur_set, set_size, chunk.osd_num) : chunk.osd_num != cur_set[chunk.role]))
         {
             extra_chunks++;
         }
@@ -312,8 +389,9 @@ void osd_t::submit_primary_del_subops(osd_op_t *cur_op, uint64_t *cur_set, pg_os
     int i = 0;
     for (auto & chunk: loc_set)
     {
-        if (!cur_set || chunk.osd_num != cur_set[chunk.role])
+        if (!cur_set || (rep ? !contains_osd(cur_set, set_size, chunk.osd_num) : chunk.osd_num != cur_set[chunk.role]))
         {
+            int stripe_num = op_data->scheme == POOL_SCHEME_REPLICATED ? 0 : chunk.role;
             if (chunk.osd_num == this->osd_num)
             {
                 clock_gettime(CLOCK_REALTIME, &subops[i].tv_begin);
@@ -326,7 +404,7 @@ void osd_t::submit_primary_del_subops(osd_op_t *cur_op, uint64_t *cur_set, pg_os
                     },
                     .oid = {
                         .inode = op_data->oid.inode,
-                        .stripe = op_data->oid.stripe | chunk.role,
+                        .stripe = op_data->oid.stripe | stripe_num,
                     },
                     // Same version as write
                     .version = op_data->fact_ver,
@@ -336,17 +414,16 @@ void osd_t::submit_primary_del_subops(osd_op_t *cur_op, uint64_t *cur_set, pg_os
             else
             {
                 subops[i].op_type = OSD_OP_OUT;
-                subops[i].send_list.push_back(subops[i].req.buf, OSD_PACKET_SIZE);
                 subops[i].peer_fd = c_cli.osd_peer_fds.at(chunk.osd_num);
                 subops[i].req.sec_del = {
                     .header = {
                         .magic = SECONDARY_OSD_OP_MAGIC,
                         .id = c_cli.next_subop_id++,
-                        .opcode = OSD_OP_SECONDARY_DELETE,
+                        .opcode = OSD_OP_SEC_DELETE,
                     },
                     .oid = {
                         .inode = op_data->oid.inode,
-                        .stripe = op_data->oid.stripe | chunk.role,
+                        .stripe = op_data->oid.stripe | stripe_num,
                     },
                     // Same version as write
                     .version = op_data->fact_ver,
@@ -354,7 +431,7 @@ void osd_t::submit_primary_del_subops(osd_op_t *cur_op, uint64_t *cur_set, pg_os
                 subops[i].callback = [cur_op, this](osd_op_t *subop)
                 {
                     int fail_fd = subop->reply.hdr.retval != 0 ? subop->peer_fd : -1;
-                    handle_primary_subop(OSD_OP_SECONDARY_DELETE, cur_op, subop->reply.hdr.retval, 0, 0);
+                    handle_primary_subop(subop, cur_op);
                     if (fail_fd >= 0)
                     {
                         // delete operation failed, drop the connection
@@ -371,14 +448,14 @@ void osd_t::submit_primary_del_subops(osd_op_t *cur_op, uint64_t *cur_set, pg_os
 void osd_t::submit_primary_sync_subops(osd_op_t *cur_op)
 {
     osd_primary_op_data_t *op_data = cur_op->op_data;
-    int n_osds = op_data->unstable_write_osds->size();
+    int n_osds = op_data->dirty_osd_count;
     osd_op_t *subops = new osd_op_t[n_osds];
     op_data->done = op_data->errors = 0;
     op_data->n_subops = n_osds;
     op_data->subops = subops;
     for (int i = 0; i < n_osds; i++)
     {
-        osd_num_t sync_osd = (*(op_data->unstable_write_osds))[i].osd_num;
+        osd_num_t sync_osd = op_data->dirty_osds[i];
         if (sync_osd == this->osd_num)
         {
             clock_gettime(CLOCK_REALTIME, &subops[i].tv_begin);
@@ -395,19 +472,18 @@ void osd_t::submit_primary_sync_subops(osd_op_t *cur_op)
         else
         {
             subops[i].op_type = OSD_OP_OUT;
-            subops[i].send_list.push_back(subops[i].req.buf, OSD_PACKET_SIZE);
             subops[i].peer_fd = c_cli.osd_peer_fds.at(sync_osd);
             subops[i].req.sec_sync = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,
                     .id = c_cli.next_subop_id++,
-                    .opcode = OSD_OP_SECONDARY_SYNC,
+                    .opcode = OSD_OP_SEC_SYNC,
                 },
             };
             subops[i].callback = [cur_op, this](osd_op_t *subop)
             {
                 int fail_fd = subop->reply.hdr.retval != 0 ? subop->peer_fd : -1;
-                handle_primary_subop(OSD_OP_SECONDARY_SYNC, cur_op, subop->reply.hdr.retval, 0, 0);
+                handle_primary_subop(subop, cur_op);
                 if (fail_fd >= 0)
                 {
                     // sync operation failed, drop the connection
@@ -448,21 +524,20 @@ void osd_t::submit_primary_stab_subops(osd_op_t *cur_op)
         else
         {
             subops[i].op_type = OSD_OP_OUT;
-            subops[i].send_list.push_back(subops[i].req.buf, OSD_PACKET_SIZE);
             subops[i].peer_fd = c_cli.osd_peer_fds.at(stab_osd.osd_num);
             subops[i].req.sec_stab = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,
                     .id = c_cli.next_subop_id++,
-                    .opcode = OSD_OP_SECONDARY_STABILIZE,
+                    .opcode = OSD_OP_SEC_STABILIZE,
                 },
                 .len = (uint64_t)(stab_osd.len * sizeof(obj_ver_id)),
             };
-            subops[i].send_list.push_back(op_data->unstable_writes + stab_osd.start, stab_osd.len * sizeof(obj_ver_id));
+            subops[i].iov.push_back(op_data->unstable_writes + stab_osd.start, stab_osd.len * sizeof(obj_ver_id));
             subops[i].callback = [cur_op, this](osd_op_t *subop)
             {
                 int fail_fd = subop->reply.hdr.retval != 0 ? subop->peer_fd : -1;
-                handle_primary_subop(OSD_OP_SECONDARY_STABILIZE, cur_op, subop->reply.hdr.retval, 0, 0);
+                handle_primary_subop(subop, cur_op);
                 if (fail_fd >= 0)
                 {
                     // sync operation failed, drop the connection
@@ -474,9 +549,20 @@ void osd_t::submit_primary_stab_subops(osd_op_t *cur_op)
     }
 }
 
-void osd_t::pg_cancel_write_queue(pg_t & pg, object_id oid, int retval)
+void osd_t::pg_cancel_write_queue(pg_t & pg, osd_op_t *first_op, object_id oid, int retval)
 {
     auto st_it = pg.write_queue.find(oid), it = st_it;
+    finish_op(first_op, retval);
+    if (it != pg.write_queue.end() && it->second == first_op)
+    {
+        it++;
+    }
+    else
+    {
+        // Write queue doesn't match the first operation.
+        // first_op is a leftover operation from the previous peering of the same PG.
+        return;
+    }
     while (it != pg.write_queue.end() && it->first == oid)
     {
         finish_op(it->second, retval);

osd_receive.cpp

@@ -1,272 +0,0 @@
-#include "cluster_client.h"
-
-void cluster_client_t::read_requests()
-{
-    for (int i = 0; i < read_ready_clients.size(); i++)
-    {
-        int peer_fd = read_ready_clients[i];
-        auto & cl = clients[peer_fd];
-        io_uring_sqe* sqe = ringloop->get_sqe();
-        if (!sqe)
-        {
-            read_ready_clients.erase(read_ready_clients.begin(), read_ready_clients.begin() + i);
-            return;
-        }
-        ring_data_t* data = ((ring_data_t*)sqe->user_data);
-        if (!cl.read_op || cl.read_remaining < receive_buffer_size)
-        {
-            cl.read_iov.iov_base = cl.in_buf;
-            cl.read_iov.iov_len = receive_buffer_size;
-        }
-        else
-        {
-            cl.read_iov.iov_base = cl.read_buf;
-            cl.read_iov.iov_len = cl.read_remaining;
-        }
-        cl.read_msg.msg_iov = &cl.read_iov;
-        cl.read_msg.msg_iovlen = 1;
-        data->callback = [this, peer_fd](ring_data_t *data) { handle_read(data, peer_fd); };
-        my_uring_prep_recvmsg(sqe, peer_fd, &cl.read_msg, 0);
-    }
-    read_ready_clients.clear();
-}
-
-void cluster_client_t::handle_read(ring_data_t *data, int peer_fd)
-{
-    auto cl_it = clients.find(peer_fd);
-    if (cl_it != clients.end())
-    {
-        auto & cl = cl_it->second;
-        if (data->res < 0 && data->res != -EAGAIN)
-        {
-            // this is a client socket, so don't panic. just disconnect it
-            printf("Client %d socket read error: %d (%s). Disconnecting client\n", peer_fd, -data->res, strerror(-data->res));
-            stop_client(peer_fd);
-            return;
-        }
-        if (data->res == -EAGAIN || cl.read_iov.iov_base == cl.in_buf && data->res < receive_buffer_size)
-        {
-            cl.read_ready--;
-            if (cl.read_ready > 0)
-                read_ready_clients.push_back(peer_fd);
-        }
-        else
-        {
-            read_ready_clients.push_back(peer_fd);
-        }
-        if (data->res == -EAGAIN)
-        {
-            return;
-        }
-        if (data->res > 0)
-        {
-            if (cl.read_iov.iov_base == cl.in_buf)
-            {
-                // Compose operation(s) from the buffer
-                int remain = data->res;
-                void *curbuf = cl.in_buf;
-                while (remain > 0)
-                {
-                    if (!cl.read_op)
-                    {
-                        cl.read_op = new osd_op_t;
-                        cl.read_op->peer_fd = peer_fd;
-                        cl.read_op->op_type = OSD_OP_IN;
-                        cl.read_buf = cl.read_op->req.buf;
-                        cl.read_remaining = OSD_PACKET_SIZE;
-                        cl.read_state = CL_READ_HDR;
-                    }
-                    if (cl.read_remaining > remain)
-                    {
-                        memcpy(cl.read_buf, curbuf, remain);
-                        cl.read_remaining -= remain;
-                        cl.read_buf += remain;
-                        remain = 0;
-                        if (cl.read_remaining <= 0)
-                            handle_finished_read(cl);
-                    }
-                    else
-                    {
-                        memcpy(cl.read_buf, curbuf, cl.read_remaining);
-                        curbuf += cl.read_remaining;
-                        remain -= cl.read_remaining;
-                        cl.read_remaining = 0;
-                        cl.read_buf = NULL;
-                        handle_finished_read(cl);
-                    }
-                }
-            }
-            else
-            {
-                // Long data
-                cl.read_remaining -= data->res;
-                cl.read_buf += data->res;
-                if (cl.read_remaining <= 0)
-                {
-                    handle_finished_read(cl);
-                }
-            }
-        }
-    }
-}
-
-void cluster_client_t::handle_finished_read(osd_client_t & cl)
-{
-    if (cl.read_state == CL_READ_HDR)
-    {
-        if (cl.read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
-            handle_reply_hdr(&cl);
-        else
-            handle_op_hdr(&cl);
-    }
-    else if (cl.read_state == CL_READ_DATA)
-    {
-        // Operation is ready
-        exec_op(cl.read_op);
-        cl.read_op = NULL;
-        cl.read_state = 0;
-    }
-    else if (cl.read_state == CL_READ_REPLY_DATA)
-    {
-        // Reply is ready
-        auto req_it = cl.sent_ops.find(cl.read_reply_id);
-        osd_op_t *request = req_it->second;
-        cl.sent_ops.erase(req_it);
-        cl.read_reply_id = 0;
-        delete cl.read_op;
-        cl.read_op = NULL;
-        cl.read_state = 0;
-        // Measure subop latency
-        timespec tv_end;
-        clock_gettime(CLOCK_REALTIME, &tv_end);
-        stats.subop_stat_count[request->req.hdr.opcode]++;
-        if (!stats.subop_stat_count[request->req.hdr.opcode])
-        {
-            stats.subop_stat_count[request->req.hdr.opcode]++;
-            stats.subop_stat_sum[request->req.hdr.opcode] = 0;
-        }
-        stats.subop_stat_sum[request->req.hdr.opcode] += (
-            (tv_end.tv_sec - request->tv_begin.tv_sec)*1000000 +
-            (tv_end.tv_nsec - request->tv_begin.tv_nsec)/1000
-        );
-        request->callback(request);
-    }
-    else
-    {
-        assert(0);
-    }
-}
-
-void cluster_client_t::handle_op_hdr(osd_client_t *cl)
-{
-    osd_op_t *cur_op = cl->read_op;
-    if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_READ)
-    {
-        if (cur_op->req.sec_rw.len > 0)
-            cur_op->buf = memalign(MEM_ALIGNMENT, cur_op->req.sec_rw.len);
-        cl->read_remaining = 0;
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE)
-    {
-        if (cur_op->req.sec_rw.len > 0)
-            cur_op->buf = memalign(MEM_ALIGNMENT, cur_op->req.sec_rw.len);
-        cl->read_remaining = cur_op->req.sec_rw.len;
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_STABILIZE ||
-        cur_op->req.hdr.opcode == OSD_OP_SECONDARY_ROLLBACK)
-    {
-        if (cur_op->req.sec_stab.len > 0)
-            cur_op->buf = memalign(MEM_ALIGNMENT, cur_op->req.sec_stab.len);
-        cl->read_remaining = cur_op->req.sec_stab.len;
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_READ)
-    {
-        if (cur_op->req.rw.len > 0)
-            cur_op->buf = memalign(MEM_ALIGNMENT, cur_op->req.rw.len);
-        cl->read_remaining = 0;
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_WRITE)
-    {
-        if (cur_op->req.rw.len > 0)
-            cur_op->buf = memalign(MEM_ALIGNMENT, cur_op->req.rw.len);
-        cl->read_remaining = cur_op->req.rw.len;
-    }
-    if (cl->read_remaining > 0)
-    {
-        // Read data
-        cl->read_buf = cur_op->buf;
-        cl->read_state = CL_READ_DATA;
-    }
-    else
-    {
-        // Operation is ready
-        cl->read_op = NULL;
-        cl->read_state = 0;
-        exec_op(cur_op);
-    }
-}
-
-void cluster_client_t::handle_reply_hdr(osd_client_t *cl)
-{
-    osd_op_t *cur_op = cl->read_op;
-    auto req_it = cl->sent_ops.find(cur_op->req.hdr.id);
-    if (req_it == cl->sent_ops.end())
-    {
-        // Command out of sync. Drop connection
-        printf("Client %d command out of sync: id %lu\n", cl->peer_fd, cur_op->req.hdr.id);
-        stop_client(cl->peer_fd);
-        return;
-    }
-    osd_op_t *op = req_it->second;
-    memcpy(op->reply.buf, cur_op->req.buf, OSD_PACKET_SIZE);
-    if (op->reply.hdr.opcode == OSD_OP_SECONDARY_READ &&
-        op->reply.hdr.retval > 0)
-    {
-        // Read data. In this case we assume that the buffer is preallocated by the caller (!)
-        assert(op->buf);
-        cl->read_state = CL_READ_REPLY_DATA;
-        cl->read_reply_id = op->req.hdr.id;
-        cl->read_buf = op->buf;
-        cl->read_remaining = op->reply.hdr.retval;
-    }
-    else if (op->reply.hdr.opcode == OSD_OP_SECONDARY_LIST &&
-        op->reply.hdr.retval > 0)
-    {
-        op->buf = memalign(MEM_ALIGNMENT, sizeof(obj_ver_id) * op->reply.hdr.retval);
-        cl->read_state = CL_READ_REPLY_DATA;
-        cl->read_reply_id = op->req.hdr.id;
-        cl->read_buf = op->buf;
-        cl->read_remaining = sizeof(obj_ver_id) * op->reply.hdr.retval;
-    }
-    else if (op->reply.hdr.opcode == OSD_OP_SHOW_CONFIG &&
-        op->reply.hdr.retval > 0)
-    {
-        op->buf = malloc(op->reply.hdr.retval);
-        cl->read_state = CL_READ_REPLY_DATA;
-        cl->read_reply_id = op->req.hdr.id;
-        cl->read_buf = op->buf;
-        cl->read_remaining = op->reply.hdr.retval;
-    }
-    else
-    {
-        delete cl->read_op;
-        cl->read_state = 0;
-        cl->read_op = NULL;
-        cl->sent_ops.erase(req_it);
-        // Measure subop latency
-        timespec tv_end;
-        clock_gettime(CLOCK_REALTIME, &tv_end);
-        stats.subop_stat_count[op->req.hdr.opcode]++;
-        if (!stats.subop_stat_count[op->req.hdr.opcode])
-        {
-            stats.subop_stat_count[op->req.hdr.opcode]++;
-            stats.subop_stat_sum[op->req.hdr.opcode] = 0;
-        }
-        stats.subop_stat_sum[op->req.hdr.opcode] += (
-            (tv_end.tv_sec - op->tv_begin.tv_sec)*1000000 +
-            (tv_end.tv_nsec - op->tv_begin.tv_nsec)/1000
-        );
-        // Copy lambda to be unaffected by `delete op`
-        std::function<void(osd_op_t*)>(op->callback)(op);
-    }
-}

osd_rmw.cpp

@@ -1,8 +1,11 @@
-#include <malloc.h>
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <string.h>
 #include <assert.h>
 #include "xor.h"
 #include "osd_rmw.h"
+#include "malloc_or_die.h"
 
 static inline void extend_read(uint32_t start, uint32_t end, osd_rmw_stripe_t & stripe)
 {
@@ -72,7 +75,7 @@ void split_stripes(uint64_t pg_minsize, uint32_t bs_block_size, uint32_t start,
     }
 }
 
-void reconstruct_stripe(osd_rmw_stripe_t *stripes, int pg_size, int role)
+void reconstruct_stripe_xor(osd_rmw_stripe_t *stripes, int pg_size, int role)
 {
     int prev = -2;
     for (int other = 0; other < pg_size; other++)
@@ -152,7 +155,7 @@ void* alloc_read_buffer(osd_rmw_stripe_t *stripes, int read_pg_size, uint64_t ad
         }
     }
     // Allocate buffer
-    void *buf = memalign(MEM_ALIGNMENT, buf_size);
+    void *buf = memalign_or_die(MEM_ALIGNMENT, buf_size);
     uint64_t buf_pos = add_size;
     for (int role = 0; role < read_pg_size; role++)
     {
@@ -207,9 +210,8 @@ void* calc_rmw(void *request_buf, osd_rmw_stripe_t *stripes, uint64_t *read_osd_
         // Object is degraded/misplaced and will be moved to <write_osd_set>
         for (int role = 0; role < pg_size; role++)
         {
-            if (write_osd_set[role] != read_osd_set[role])
+            if (write_osd_set[role] != read_osd_set[role] && write_osd_set[role] != 0)
             {
-                // FIXME: For EC more than 2+1: handle case when write_osd_set == 0 and read_osd_set != 0
                 // We need to get data for any moved / recovered chunk
                 // And we need a continuous write buffer so we'll only optimize
                 // for the case when the whole chunk is ovewritten in the request
@@ -357,21 +359,22 @@ static void xor_multiple_buffers(buf_len_t *xor1, int n1, buf_len_t *xor2, int n
     }
 }
 
-void calc_rmw_parity(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size)
+void calc_rmw_parity_xor(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size)
 {
     int pg_minsize = pg_size-1;
     for (int role = 0; role < pg_size; role++)
     {
         if (stripes[role].read_end != 0 && stripes[role].missing)
         {
-            // Reconstruct missing stripe (EC k+1)
+            // Reconstruct missing stripe (XOR k+1)
-            reconstruct_stripe(stripes, pg_size, role);
+            reconstruct_stripe_xor(stripes, pg_size, role);
             break;
         }
     }
     uint32_t start = 0, end = 0;
-    if (!stripes[pg_minsize].missing || write_osd_set != read_osd_set)
+    if (write_osd_set[pg_minsize] != 0 || write_osd_set != read_osd_set)
     {
+        // Required for the next two if()s
         for (int role = 0; role < pg_minsize; role++)
         {
             if (stripes[role].req_end != 0)
@@ -385,10 +388,9 @@ void calc_rmw_parity(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_osd_
     {
         for (int role = 0; role < pg_minsize; role++)
         {
-            if (write_osd_set[role] != read_osd_set[role] &&
+            if (write_osd_set[role] != read_osd_set[role] && write_osd_set[role] != 0 &&
                 (stripes[role].req_start != 0 || stripes[role].req_end != chunk_size))
             {
-                // FIXME again, handle case when write_osd_set[role] is 0
                 // Copy modified chunk into the read buffer to write it back
                 memcpy(
                     stripes[role].read_buf + stripes[role].req_start,
@@ -401,9 +403,9 @@ void calc_rmw_parity(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_osd_
             }
         }
     }
-    if (!stripes[pg_minsize].missing && end != 0)
+    if (write_osd_set[pg_minsize] != 0 && end != 0)
     {
-        // Calculate new parity (EC k+1)
+        // Calculate new parity (XOR k+1)
         int parity = pg_minsize, prev = -2;
         for (int other = 0; other < pg_minsize; other++)
         {

osd_rmw.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #pragma once
 
 #include <stdint.h>
@@ -25,7 +28,7 @@ struct osd_rmw_stripe_t
 
 void split_stripes(uint64_t pg_minsize, uint32_t bs_block_size, uint32_t start, uint32_t len, osd_rmw_stripe_t *stripes);
 
-void reconstruct_stripe(osd_rmw_stripe_t *stripes, int pg_size, int role);
+void reconstruct_stripe_xor(osd_rmw_stripe_t *stripes, int pg_size, int role);
 
 int extend_missing_stripes(osd_rmw_stripe_t *stripes, osd_num_t *osd_set, int minsize, int size);
 
@@ -34,4 +37,4 @@ void* alloc_read_buffer(osd_rmw_stripe_t *stripes, int read_pg_size, uint64_t ad
 void* calc_rmw(void *request_buf, osd_rmw_stripe_t *stripes, uint64_t *read_osd_set,
     uint64_t pg_size, uint64_t pg_minsize, uint64_t pg_cursize, uint64_t *write_osd_set, uint64_t chunk_size);
 
-void calc_rmw_parity(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size);
+void calc_rmw_parity_xor(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size);

osd_rmw_test.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <string.h>
 #include "osd_rmw.cpp"
 #include "test_pattern.h"
@@ -58,7 +61,7 @@ Cases:
      input buffer: [ write0, write1 ],
      rmw buffer: [ write2, read0, read1, read2 ],
    }
-   then, after calc_rmw_parity(): {
+   then, after calc_rmw_parity_xor(): {
      write: [ [ 128K-4K, 128K ], [ 0, 128K ], [ 0, 128K ] ],
      write1==read1,
    }
@@ -82,7 +85,7 @@ Cases:
      input buffer: NULL,
      rmw buffer: [ read0, read1, read2 ],
    }
-   then, after calc_rmw_parity(): {
+   then, after calc_rmw_parity_xor(): {
      write: [ [ 0, 128K ], [ 0, 0 ], [ 0, 0 ] ],
      write0==read0,
    }
@@ -182,7 +185,7 @@ void test4()
     set_pattern(stripes[0].read_buf, 128*1024, PATTERN1); // old data
     set_pattern(stripes[1].read_buf, 128*1024-4096, UINT64_MAX); // didn't read it, it's missing
     set_pattern(stripes[2].read_buf, 128*1024-4096, 0); // old parity = 0
-    calc_rmw_parity(stripes, 3, osd_set, osd_set, 128*1024);
+    calc_rmw_parity_xor(stripes, 3, osd_set, osd_set, 128*1024);
     check_pattern(stripes[2].write_buf, 4096, PATTERN0^PATTERN1); // new parity
     check_pattern(stripes[2].write_buf+4096, 128*1024-4096*2, 0); // new parity
     check_pattern(stripes[2].write_buf+128*1024-4096, 4096, PATTERN0^PATTERN1); // new parity
@@ -268,7 +271,7 @@ void test7()
     set_pattern(stripes[0].read_buf, 128*1024, PATTERN1); // old data
     set_pattern(stripes[1].read_buf, 128*1024, UINT64_MAX); // didn't read it, it's missing
     set_pattern(stripes[2].read_buf, 128*1024, 0); // old parity = 0
-    calc_rmw_parity(stripes, 3, osd_set, write_osd_set, 128*1024);
+    calc_rmw_parity_xor(stripes, 3, osd_set, write_osd_set, 128*1024);
     assert(stripes[0].write_start == 128*1024-4096 && stripes[0].write_end == 128*1024);
     assert(stripes[1].write_start == 0 && stripes[1].write_end == 128*1024);
     assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024);
@@ -306,7 +309,7 @@ void test8()
     // Test 8.2
     set_pattern(write_buf, 128*1024+4096, PATTERN0);
     set_pattern(stripes[1].read_buf, 128*1024-4096, PATTERN1);
-    calc_rmw_parity(stripes, 3, osd_set, write_osd_set, 128*1024);
+    calc_rmw_parity_xor(stripes, 3, osd_set, write_osd_set, 128*1024);
     assert(stripes[0].write_start == 0 && stripes[0].write_end == 128*1024); // recheck again
     assert(stripes[1].write_start == 0 && stripes[1].write_end == 4096);     // recheck again
     assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024); // recheck again
@@ -344,10 +347,10 @@ void test9()
     assert(stripes[0].write_buf == NULL);
     assert(stripes[1].write_buf == NULL);
     assert(stripes[2].write_buf == NULL);
-    // Test 8.2
+    // Test 9.2
     set_pattern(stripes[1].read_buf, 128*1024, 0);
     set_pattern(stripes[2].read_buf, 128*1024, PATTERN1);
-    calc_rmw_parity(stripes, 3, osd_set, write_osd_set, 128*1024);
+    calc_rmw_parity_xor(stripes, 3, osd_set, write_osd_set, 128*1024);
     assert(stripes[0].write_start == 0 && stripes[0].write_end == 128*1024);
     assert(stripes[1].write_start == 0 && stripes[1].write_end == 0);
     assert(stripes[2].write_start == 0 && stripes[2].write_end == 0);

osd_secondary.cpp

@@ -1,30 +1,34 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include "osd.h"
 
 #include "json11/json11.hpp"
 
 void osd_t::secondary_op_callback(osd_op_t *op)
 {
-    if (op->req.hdr.opcode == OSD_OP_SECONDARY_READ ||
+    if (op->req.hdr.opcode == OSD_OP_SEC_READ ||
-        op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE)
+        op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+        op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
     {
         op->reply.sec_rw.version = op->bs_op->version;
     }
-    else if (op->req.hdr.opcode == OSD_OP_SECONDARY_DELETE)
+    else if (op->req.hdr.opcode == OSD_OP_SEC_DELETE)
     {
         op->reply.sec_del.version = op->bs_op->version;
     }
-    if (op->req.hdr.opcode == OSD_OP_SECONDARY_READ &&
+    if (op->req.hdr.opcode == OSD_OP_SEC_READ &&
         op->bs_op->retval > 0)
     {
-        op->send_list.push_back(op->buf, op->bs_op->retval);
+        op->iov.push_back(op->buf, op->bs_op->retval);
     }
-    else if (op->req.hdr.opcode == OSD_OP_SECONDARY_LIST)
+    else if (op->req.hdr.opcode == OSD_OP_SEC_LIST)
     {
         // allocated by blockstore
         op->buf = op->bs_op->buf;
         if (op->bs_op->retval > 0)
         {
-            op->send_list.push_back(op->buf, op->bs_op->retval * sizeof(obj_ver_id));
+            op->iov.push_back(op->buf, op->bs_op->retval * sizeof(obj_ver_id));
         }
         op->reply.sec_list.stable_count = op->bs_op->version;
     }
@@ -38,16 +42,18 @@ void osd_t::exec_secondary(osd_op_t *cur_op)
 {
     cur_op->bs_op = new blockstore_op_t();
     cur_op->bs_op->callback = [this, cur_op](blockstore_op_t* bs_op) { secondary_op_callback(cur_op); };
-    cur_op->bs_op->opcode = (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_READ ? BS_OP_READ
+    cur_op->bs_op->opcode = (cur_op->req.hdr.opcode == OSD_OP_SEC_READ ? BS_OP_READ
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE ? BS_OP_WRITE
+        : (cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE ? BS_OP_WRITE
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_SYNC ? BS_OP_SYNC
+        : (cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE ? BS_OP_WRITE_STABLE
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_STABILIZE ? BS_OP_STABLE
+        : (cur_op->req.hdr.opcode == OSD_OP_SEC_SYNC ? BS_OP_SYNC
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_ROLLBACK ? BS_OP_ROLLBACK
+        : (cur_op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ? BS_OP_STABLE
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_DELETE ? BS_OP_DELETE
+        : (cur_op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK ? BS_OP_ROLLBACK
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_LIST ? BS_OP_LIST
+        : (cur_op->req.hdr.opcode == OSD_OP_SEC_DELETE ? BS_OP_DELETE
-        : -1)))))));
+        : (cur_op->req.hdr.opcode == OSD_OP_SEC_LIST ? BS_OP_LIST
-    if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_READ ||
+        : -1))))))));
-        cur_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE)
+    if (cur_op->req.hdr.opcode == OSD_OP_SEC_READ ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
     {
         cur_op->bs_op->oid = cur_op->req.sec_rw.oid;
         cur_op->bs_op->version = cur_op->req.sec_rw.version;
@@ -58,7 +64,7 @@ void osd_t::exec_secondary(osd_op_t *cur_op)
         cur_op->bs_op->retval = cur_op->bs_op->len;
 #endif
     }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_DELETE)
+    else if (cur_op->req.hdr.opcode == OSD_OP_SEC_DELETE)
     {
         cur_op->bs_op->oid = cur_op->req.sec_del.oid;
         cur_op->bs_op->version = cur_op->req.sec_del.version;
@@ -66,8 +72,8 @@ void osd_t::exec_secondary(osd_op_t *cur_op)
         cur_op->bs_op->retval = 0;
 #endif
     }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_STABILIZE ||
+    else if (cur_op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
-        cur_op->req.hdr.opcode == OSD_OP_SECONDARY_ROLLBACK)
+        cur_op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK)
     {
         cur_op->bs_op->len = cur_op->req.sec_stab.len/sizeof(obj_ver_id);
         cur_op->bs_op->buf = cur_op->buf;
@@ -75,11 +81,12 @@ void osd_t::exec_secondary(osd_op_t *cur_op)
         cur_op->bs_op->retval = 0;
 #endif
     }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_LIST)
+    else if (cur_op->req.hdr.opcode == OSD_OP_SEC_LIST)
     {
         if (cur_op->req.sec_list.pg_count < cur_op->req.sec_list.list_pg)
         {
             // requested pg number is greater than total pg count
+            printf("Invalid LIST request: pg count %u < pg number %u\n", cur_op->req.sec_list.pg_count, cur_op->req.sec_list.list_pg);
             cur_op->bs_op->retval = -EINVAL;
             secondary_op_callback(cur_op);
             return;
@@ -87,6 +94,8 @@ void osd_t::exec_secondary(osd_op_t *cur_op)
         cur_op->bs_op->oid.stripe = cur_op->req.sec_list.pg_stripe_size;
         cur_op->bs_op->len = cur_op->req.sec_list.pg_count;
         cur_op->bs_op->offset = cur_op->req.sec_list.list_pg - 1;
+        cur_op->bs_op->oid.inode = cur_op->req.sec_list.min_inode;
+        cur_op->bs_op->version = cur_op->req.sec_list.max_inode;
 #ifdef OSD_STUB
         cur_op->bs_op->retval = 0;
         cur_op->bs_op->buf = NULL;
@@ -103,9 +112,9 @@ void osd_t::exec_show_config(osd_op_t *cur_op)
 {
     // FIXME: Send the real config, not its source
     std::string cfg_str = json11::Json(config).dump();
-    cur_op->buf = malloc(cfg_str.size()+1);
+    cur_op->buf = malloc_or_die(cfg_str.size()+1);
     memcpy(cur_op->buf, cfg_str.c_str(), cfg_str.size()+1);
-    cur_op->send_list.push_back(cur_op->buf, cfg_str.size()+1);
+    cur_op->iov.push_back(cur_op->buf, cfg_str.size()+1);
     finish_op(cur_op, cfg_str.size()+1);
 }
 

osd_send.cpp

@@ -1,138 +0,0 @@
-#include "cluster_client.h"
-
-void cluster_client_t::outbox_push(osd_op_t *cur_op)
-{
-    assert(cur_op->peer_fd);
-    auto & cl = clients.at(cur_op->peer_fd);
-    if (cur_op->op_type == OSD_OP_OUT)
-    {
-        clock_gettime(CLOCK_REALTIME, &cur_op->tv_begin);
-    }
-    cl.outbox.push_back(cur_op);
-    if (cl.write_op || cl.outbox.size() > 1 || !try_send(cl))
-    {
-        if (cl.write_state == 0)
-        {
-            cl.write_state = CL_WRITE_READY;
-            write_ready_clients.push_back(cur_op->peer_fd);
-        }
-        ringloop->wakeup();
-    }
-}
-
-bool cluster_client_t::try_send(osd_client_t & cl)
-{
-    int peer_fd = cl.peer_fd;
-    io_uring_sqe* sqe = ringloop->get_sqe();
-    if (!sqe)
-    {
-        return false;
-    }
-    ring_data_t* data = ((ring_data_t*)sqe->user_data);
-    if (!cl.write_op)
-    {
-        // pick next command
-        cl.write_op = cl.outbox.front();
-        cl.outbox.pop_front();
-        cl.write_state = CL_WRITE_REPLY;
-        if (cl.write_op->op_type == OSD_OP_IN)
-        {
-            // Measure execution latency
-            timespec tv_end;
-            clock_gettime(CLOCK_REALTIME, &tv_end);
-            stats.op_stat_count[cl.write_op->req.hdr.opcode]++;
-            if (!stats.op_stat_count[cl.write_op->req.hdr.opcode])
-            {
-                stats.op_stat_count[cl.write_op->req.hdr.opcode]++;
-                stats.op_stat_sum[cl.write_op->req.hdr.opcode] = 0;
-                stats.op_stat_bytes[cl.write_op->req.hdr.opcode] = 0;
-            }
-            stats.op_stat_sum[cl.write_op->req.hdr.opcode] += (
-                (tv_end.tv_sec - cl.write_op->tv_begin.tv_sec)*1000000 +
-                (tv_end.tv_nsec - cl.write_op->tv_begin.tv_nsec)/1000
-            );
-            if (cl.write_op->req.hdr.opcode == OSD_OP_READ ||
-                cl.write_op->req.hdr.opcode == OSD_OP_WRITE)
-            {
-                stats.op_stat_bytes[cl.write_op->req.hdr.opcode] += cl.write_op->req.rw.len;
-            }
-            else if (cl.write_op->req.hdr.opcode == OSD_OP_SECONDARY_READ ||
-                cl.write_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE)
-            {
-                stats.op_stat_bytes[cl.write_op->req.hdr.opcode] += cl.write_op->req.sec_rw.len;
-            }
-        }
-    }
-    cl.write_msg.msg_iov = cl.write_op->send_list.get_iovec();
-    cl.write_msg.msg_iovlen = cl.write_op->send_list.get_size();
-    data->callback = [this, peer_fd](ring_data_t *data) { handle_send(data, peer_fd); };
-    my_uring_prep_sendmsg(sqe, peer_fd, &cl.write_msg, 0);
-    return true;
-}
-
-void cluster_client_t::send_replies()
-{
-    for (int i = 0; i < write_ready_clients.size(); i++)
-    {
-        int peer_fd = write_ready_clients[i];
-        if (!try_send(clients[peer_fd]))
-        {
-            write_ready_clients.erase(write_ready_clients.begin(), write_ready_clients.begin() + i);
-            return;
-        }
-    }
-    write_ready_clients.clear();
-}
-
-void cluster_client_t::handle_send(ring_data_t *data, int peer_fd)
-{
-    auto cl_it = clients.find(peer_fd);
-    if (cl_it != clients.end())
-    {
-        auto & cl = cl_it->second;
-        if (data->res < 0 && data->res != -EAGAIN)
-        {
-            // this is a client socket, so don't panic. just disconnect it
-            printf("Client %d socket write error: %d (%s). Disconnecting client\n", peer_fd, -data->res, strerror(-data->res));
-            stop_client(peer_fd);
-            return;
-        }
-        if (data->res >= 0)
-        {
-            osd_op_t *cur_op = cl.write_op;
-            while (data->res > 0 && cur_op->send_list.sent < cur_op->send_list.count)
-            {
-                iovec & iov = cur_op->send_list.buf[cur_op->send_list.sent];
-                if (iov.iov_len <= data->res)
-                {
-                    data->res -= iov.iov_len;
-                    cur_op->send_list.sent++;
-                }
-                else
-                {
-                    iov.iov_len -= data->res;
-                    iov.iov_base += data->res;
-                    break;
-                }
-            }
-            if (cur_op->send_list.sent >= cur_op->send_list.count)
-            {
-                // Done
-                if (cur_op->op_type == OSD_OP_IN)
-                {
-                    delete cur_op;
-                }
-                else
-                {
-                    cl.sent_ops[cl.write_op->req.hdr.id] = cl.write_op;
-                }
-                cl.write_op = NULL;
-                cl.write_state = cl.outbox.size() > 0 ? CL_WRITE_READY : 0;
-            }
-        }
-        if (cl.write_state != 0)
-        {
-            write_ready_clients.push_back(peer_fd);
-        }
-    }
-}

osd_test.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <sys/types.h>
 #include <sys/socket.h>
 #include <netinet/in.h>
@@ -19,6 +22,8 @@
 
 int connect_osd(const char *osd_address, int osd_port);
 
+uint64_t test_read(int connect_fd, uint64_t inode, uint64_t stripe, uint64_t version, uint64_t offset, uint64_t len);
+
 uint64_t test_write(int connect_fd, uint64_t inode, uint64_t stripe, uint64_t version, uint64_t pattern);
 
 void* test_primary_read(int connect_fd, uint64_t inode, uint64_t offset, uint64_t len);
@@ -105,7 +110,7 @@ int main3(int narg, char *args[])
     return 0;
 }
 
-int main(int narg, char *args[])
+int main4(int narg, char *args[])
 {
     int connect_fd;
     // Cluster write (sync not implemented yet)
@@ -117,6 +122,15 @@ int main(int narg, char *args[])
     return 0;
 }
 
+int main(int narg, char *args[])
+{
+    int connect_fd;
+    connect_fd = connect_osd("192.168.7.2", 43051);
+    test_read(connect_fd, 1, 1039663104, UINT64_MAX, 0, 128*1024);
+    close(connect_fd);
+    return 0;
+}
+
 int connect_osd(const char *osd_address, int osd_port)
 {
     struct sockaddr_in addr;
@@ -167,13 +181,73 @@ bool check_reply(int r, osd_any_op_t & op, osd_any_reply_t & reply, int expected
     return true;
 }
 
+uint64_t test_read(int connect_fd, uint64_t inode, uint64_t stripe, uint64_t version, uint64_t offset, uint64_t len)
+{
+    osd_any_op_t op;
+    osd_any_reply_t reply;
+    op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
+    op.hdr.id = 1;
+    op.hdr.opcode = OSD_OP_SEC_READ;
+    op.sec_rw.oid = {
+        .inode = inode,
+        .stripe = stripe,
+    };
+    op.sec_rw.version = version;
+    op.sec_rw.offset = offset;
+    op.sec_rw.len = len;
+    void *data = memalign(MEM_ALIGNMENT, op.sec_rw.len);
+    write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE);
+    int r = read_blocking(connect_fd, reply.buf, OSD_PACKET_SIZE);
+    if (!check_reply(r, op, reply, op.sec_rw.len))
+    {
+        free(data);
+        return 0;
+    }
+    r = read_blocking(connect_fd, data, len);
+    if (r != len)
+    {
+        free(data);
+        perror("read data");
+        return 0;
+    }
+    free(data);
+    printf("Read %lx:%lx v%lu = v%lu\n", inode, stripe, version, reply.sec_rw.version);
+    op.hdr.opcode = OSD_OP_SEC_LIST;
+    op.sec_list.list_pg = 1;
+    op.sec_list.pg_count = 1;
+    op.sec_list.pg_stripe_size = 4*1024*1024;
+    write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE);
+    r = read_blocking(connect_fd, reply.buf, OSD_PACKET_SIZE);
+    if (reply.hdr.retval < 0 || !check_reply(r, op, reply, reply.hdr.retval))
+    {
+        return 0;
+    }
+    data = memalign(MEM_ALIGNMENT, sizeof(obj_ver_id)*reply.hdr.retval);
+    r = read_blocking(connect_fd, data, sizeof(obj_ver_id)*reply.hdr.retval);
+    if (r != sizeof(obj_ver_id)*reply.hdr.retval)
+    {
+        free(data);
+        perror("read data");
+        return 0;
+    }
+    obj_ver_id *ov = (obj_ver_id*)data;
+    for (int i = 0; i < reply.hdr.retval; i++)
+    {
+        if (ov[i].oid.inode == inode && (ov[i].oid.stripe & ~(4096-1)) == (stripe & ~(4096-1)))
+        {
+            printf("list: %lx:%lx v%lu stable=%d\n", ov[i].oid.inode, ov[i].oid.stripe, ov[i].version, i < reply.sec_list.stable_count ? 1 : 0);
+        }
+    }
+    return 0;
+}
+
 uint64_t test_write(int connect_fd, uint64_t inode, uint64_t stripe, uint64_t version, uint64_t pattern)
 {
     osd_any_op_t op;
     osd_any_reply_t reply;
     op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
     op.hdr.id = 1;
-    op.hdr.opcode = OSD_OP_SECONDARY_WRITE;
+    op.hdr.opcode = OSD_OP_SEC_WRITE;
     op.sec_rw.oid = {
         .inode = inode,
         .stripe = stripe,
@@ -283,7 +357,7 @@ void test_list_stab(int connect_fd)
     osd_any_reply_t reply;
     op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
     op.hdr.id = 1;
-    op.hdr.opcode = OSD_OP_SECONDARY_LIST;
+    op.hdr.opcode = OSD_OP_SEC_LIST;
     op.sec_list.pg_count = 0;
     assert(write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE) == OSD_PACKET_SIZE);
     int r = read_blocking(connect_fd, reply.buf, OSD_PACKET_SIZE);
@@ -299,7 +373,7 @@ void test_list_stab(int connect_fd)
         // Stabilize in portions of 32 entries
         if (i - last_start >= 32 || i == total_count)
         {
-            op.hdr.opcode = OSD_OP_SECONDARY_STABILIZE;
+            op.hdr.opcode = OSD_OP_SEC_STABILIZE;
             op.sec_stab.len = sizeof(obj_ver_id) * (i - last_start);
             assert(write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE) == OSD_PACKET_SIZE);
             assert(write_blocking(connect_fd, data + last_start, op.sec_stab.len) == op.sec_stab.len);

pg_states.cpp

@@ -1,8 +1,11 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #include "pg_states.h"
 
-const int pg_state_bit_count = 13;
+const int pg_state_bit_count = 14;
 
-const int pg_state_bits[13] = {
+const int pg_state_bits[14] = {
     PG_STARTING,
     PG_PEERING,
     PG_INCOMPLETE,
@@ -14,10 +17,11 @@ const int pg_state_bits[13] = {
     PG_HAS_DEGRADED,
     PG_HAS_MISPLACED,
     PG_HAS_UNCLEAN,
+    PG_HAS_INVALID,
     PG_LEFT_ON_DEAD,
 };
 
-const char *pg_state_names[13] = {
+const char *pg_state_names[14] = {
     "starting",
     "peering",
     "incomplete",
@@ -29,5 +33,6 @@ const char *pg_state_names[13] = {
     "has_degraded",
     "has_misplaced",
     "has_unclean",
+    "has_invalid",
     "left_on_dead",
 };

pg_states.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #pragma once
 
 // Placement group states
@@ -15,9 +18,11 @@
 #define PG_HAS_DEGRADED (1<<8)
 #define PG_HAS_MISPLACED (1<<9)
 #define PG_HAS_UNCLEAN (1<<10)
-#define PG_LEFT_ON_DEAD (1<<11)
+#define PG_HAS_INVALID (1<<11)
+#define PG_LEFT_ON_DEAD (1<<12)
 
-// FIXME: Safe default that doesn't depend on pg_stripe_size or pg_block_size
+// Lower bits that represent object role (EC 0/1/2... or always 0 with replication)
+// 12 bits is a safe default that doesn't depend on pg_stripe_size or pg_block_size
 #define STRIPE_MASK ((uint64_t)4096 - 1)
 
 // OSD object states
@@ -26,7 +31,6 @@
 #define OBJ_MISPLACED 0x08
 #define OBJ_NEEDS_STABLE 0x10000
 #define OBJ_NEEDS_ROLLBACK 0x20000
-#define OBJ_BUGGY 0x80000
 
 extern const int pg_state_bits[];
 extern const char *pg_state_names[];

qemu_driver.c

@@ -0,0 +1,400 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+// QEMU block driver
+
+#define _GNU_SOURCE
+#include "qemu/osdep.h"
+#include "qemu/units.h"
+#include "block/block_int.h"
+#include "block/qdict.h"
+#include "qapi/error.h"
+#include "qapi/qmp/qdict.h"
+#include "qapi/qmp/qerror.h"
+#include "qemu/uri.h"
+#include "qemu/error-report.h"
+#include "qemu/module.h"
+#include "qemu/option.h"
+#include "qemu/cutils.h"
+
+#include "qemu_proxy.h"
+
+typedef struct VitastorClient
+{
+    void *proxy;
+    char *etcd_host;
+    char *etcd_prefix;
+    uint64_t inode;
+    uint64_t pool;
+    uint64_t size;
+    int readonly;
+    QemuMutex mutex;
+} VitastorClient;
+
+typedef struct VitastorRPC
+{
+    BlockDriverState *bs;
+    Coroutine *co;
+    QEMUIOVector *iov;
+    int ret;
+    int complete;
+} VitastorRPC;
+
+static char *qemu_rbd_next_tok(char *src, char delim, char **p)
+{
+    char *end;
+    *p = NULL;
+    for (end = src; *end; ++end)
+    {
+        if (*end == delim)
+            break;
+        if (*end == '\\' && end[1] != '\0')
+            end++;
+    }
+    if (*end == delim)
+    {
+        *p = end + 1;
+        *end = '\0';
+    }
+    return src;
+}
+
+static void qemu_rbd_unescape(char *src)
+{
+    char *p;
+    for (p = src; *src; ++src, ++p)
+    {
+        if (*src == '\\' && src[1] != '\0')
+            src++;
+        *p = *src;
+    }
+    *p = '\0';
+}
+
+// vitastor[:key=value]*
+// vitastor:etcd_host=127.0.0.1:inode=1:pool=1
+static void vitastor_parse_filename(const char *filename, QDict *options, Error **errp)
+{
+    const char *start;
+    char *p, *buf;
+
+    if (!strstart(filename, "vitastor:", &start))
+    {
+        error_setg(errp, "File name must start with 'vitastor:'");
+        return;
+    }
+
+    buf = g_strdup(start);
+    p = buf;
+
+    // The following are all key/value pairs
+    while (p)
+    {
+        char *name, *value;
+        name = qemu_rbd_next_tok(p, '=', &p);
+        if (!p)
+        {
+            error_setg(errp, "conf option %s has no value", name);
+            break;
+        }
+        qemu_rbd_unescape(name);
+        value = qemu_rbd_next_tok(p, ':', &p);
+        qemu_rbd_unescape(value);
+        if (!strcmp(name, "inode") || !strcmp(name, "pool") || !strcmp(name, "size"))
+        {
+            unsigned long long num_val;
+            if (parse_uint_full(value, &num_val, 0))
+            {
+                error_setg(errp, "Illegal %s: %s", name, value);
+                goto out;
+            }
+            qdict_put_int(options, name, num_val);
+        }
+        else
+        {
+            qdict_put_str(options, name, value);
+        }
+    }
+    if (!qdict_get_try_int(options, "inode", 0))
+    {
+        error_setg(errp, "inode is missing");
+        goto out;
+    }
+    if (!(qdict_get_try_int(options, "inode", 0) >> (64-POOL_ID_BITS)) &&
+        !qdict_get_try_int(options, "pool", 0))
+    {
+        error_setg(errp, "pool number is missing");
+        goto out;
+    }
+    if (!qdict_get_try_int(options, "size", 0))
+    {
+        error_setg(errp, "size is missing");
+        goto out;
+    }
+    if (!qdict_get_str(options, "etcd_host"))
+    {
+        error_setg(errp, "etcd_host is missing");
+        goto out;
+    }
+
+out:
+    g_free(buf);
+    return;
+}
+
+static int vitastor_file_open(BlockDriverState *bs, QDict *options, int flags, Error **errp)
+{
+    VitastorClient *client = bs->opaque;
+    int64_t ret = 0;
+    client->etcd_host = g_strdup(qdict_get_try_str(options, "etcd_host"));
+    client->etcd_prefix = g_strdup(qdict_get_try_str(options, "etcd_prefix"));
+    client->inode = qdict_get_int(options, "inode");
+    client->pool = qdict_get_int(options, "pool");
+    if (client->pool)
+        client->inode = (client->inode & ((1l << (64-POOL_ID_BITS)) - 1)) | (client->pool << (64-POOL_ID_BITS));
+    client->size = qdict_get_int(options, "size");
+    client->readonly = (flags & BDRV_O_RDWR) ? 1 : 0;
+    client->proxy = vitastor_proxy_create(bdrv_get_aio_context(bs), client->etcd_host, client->etcd_prefix);
+    //client->aio_context = bdrv_get_aio_context(bs);
+    bs->total_sectors = client->size / BDRV_SECTOR_SIZE;
+    qdict_del(options, "etcd_host");
+    qdict_del(options, "etcd_prefix");
+    qdict_del(options, "inode");
+    qdict_del(options, "pool");
+    qdict_del(options, "size");
+    qemu_mutex_init(&client->mutex);
+    return ret;
+}
+
+static void vitastor_close(BlockDriverState *bs)
+{
+    VitastorClient *client = bs->opaque;
+    vitastor_proxy_destroy(client->proxy);
+    qemu_mutex_destroy(&client->mutex);
+    g_free(client->etcd_host);
+    if (client->etcd_prefix)
+        g_free(client->etcd_prefix);
+}
+
+static int vitastor_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
+{
+    bsz->phys = 4096;
+    bsz->log = 4096;
+    return 0;
+}
+
+static int coroutine_fn vitastor_co_create_opts(
+#if QEMU_VERSION_MAJOR >= 4
+    BlockDriver *drv,
+#endif
+    const char *url, QemuOpts *opts, Error **errp)
+{
+    QDict *options;
+    int ret;
+
+    options = qdict_new();
+    vitastor_parse_filename(url, options, errp);
+    if (*errp)
+    {
+        ret = -1;
+        goto out;
+    }
+
+    // inodes don't require creation in Vitastor. FIXME: They will when there will be some metadata
+
+    ret = 0;
+out:
+    qobject_unref(options);
+    return ret;
+}
+
+static int coroutine_fn vitastor_co_truncate(BlockDriverState *bs, int64_t offset,
+#if QEMU_VERSION_MAJOR >= 4
+    bool exact,
+#endif
+    PreallocMode prealloc, Error **errp)
+{
+    VitastorClient *client = bs->opaque;
+
+    if (prealloc != PREALLOC_MODE_OFF)
+    {
+        error_setg(errp, "Unsupported preallocation mode '%s'", PreallocMode_str(prealloc));
+        return -ENOTSUP;
+    }
+
+    // TODO: Resize inode to <offset> bytes
+    client->size = offset / BDRV_SECTOR_SIZE;
+
+    return 0;
+}
+
+static int vitastor_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
+{
+    bdi->cluster_size = 4096;
+    return 0;
+}
+
+static int64_t vitastor_getlength(BlockDriverState *bs)
+{
+    VitastorClient *client = bs->opaque;
+    return client->size;
+}
+
+static void vitastor_refresh_limits(BlockDriverState *bs, Error **errp)
+{
+    bs->bl.request_alignment = 4096;
+    bs->bl.min_mem_alignment = 4096;
+    bs->bl.opt_mem_alignment = 4096;
+}
+
+static int64_t vitastor_get_allocated_file_size(BlockDriverState *bs)
+{
+    return 0;
+}
+
+static void vitastor_co_init_task(BlockDriverState *bs, VitastorRPC *task)
+{
+    *task = (VitastorRPC) {
+        .co     = qemu_coroutine_self(),
+        .bs     = bs,
+    };
+}
+
+static void vitastor_co_generic_bh_cb(int retval, void *opaque)
+{
+    VitastorRPC *task = opaque;
+    task->ret = retval;
+    task->complete = 1;
+    if (qemu_coroutine_self() != task->co)
+    {
+        aio_co_wake(task->co);
+    }
+}
+
+static int coroutine_fn vitastor_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes, QEMUIOVector *iov, int flags)
+{
+    VitastorClient *client = bs->opaque;
+    VitastorRPC task;
+    vitastor_co_init_task(bs, &task);
+    task.iov = iov;
+
+    qemu_mutex_lock(&client->mutex);
+    vitastor_proxy_rw(0, client->proxy, client->inode, offset, bytes, iov->iov, iov->niov, vitastor_co_generic_bh_cb, &task);
+    qemu_mutex_unlock(&client->mutex);
+
+    while (!task.complete)
+    {
+        qemu_coroutine_yield();
+    }
+
+    return task.ret;
+}
+
+static int coroutine_fn vitastor_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes, QEMUIOVector *iov, int flags)
+{
+    VitastorClient *client = bs->opaque;
+    VitastorRPC task;
+    vitastor_co_init_task(bs, &task);
+    task.iov = iov;
+
+    qemu_mutex_lock(&client->mutex);
+    vitastor_proxy_rw(1, client->proxy, client->inode, offset, bytes, iov->iov, iov->niov, vitastor_co_generic_bh_cb, &task);
+    qemu_mutex_unlock(&client->mutex);
+
+    while (!task.complete)
+    {
+        qemu_coroutine_yield();
+    }
+
+    return task.ret;
+}
+
+static int coroutine_fn vitastor_co_flush(BlockDriverState *bs)
+{
+    VitastorClient *client = bs->opaque;
+    VitastorRPC task;
+    vitastor_co_init_task(bs, &task);
+
+    qemu_mutex_lock(&client->mutex);
+    vitastor_proxy_sync(client->proxy, vitastor_co_generic_bh_cb, &task);
+    qemu_mutex_unlock(&client->mutex);
+
+    while (!task.complete)
+    {
+        qemu_coroutine_yield();
+    }
+
+    return task.ret;
+}
+
+static QemuOptsList vitastor_create_opts = {
+    .name = "vitastor-create-opts",
+    .head = QTAILQ_HEAD_INITIALIZER(vitastor_create_opts.head),
+    .desc = {
+        {
+            .name = BLOCK_OPT_SIZE,
+            .type = QEMU_OPT_SIZE,
+            .help = "Virtual disk size"
+        },
+        { /* end of list */ }
+    }
+};
+
+static const char *vitastor_strong_runtime_opts[] = {
+    "inode",
+    "pool",
+    "etcd_host",
+    "etcd_prefix",
+
+    NULL
+};
+
+static BlockDriver bdrv_vitastor = {
+    .format_name                    = "vitastor",
+    .protocol_name                  = "vitastor",
+
+    .instance_size                  = sizeof(VitastorClient),
+    .bdrv_parse_filename            = vitastor_parse_filename,
+
+    .bdrv_has_zero_init             = bdrv_has_zero_init_1,
+#if QEMU_VERSION_MAJOR >= 4
+    .bdrv_has_zero_init_truncate    = bdrv_has_zero_init_1,
+#endif
+    .bdrv_get_info                  = vitastor_get_info,
+    .bdrv_getlength                 = vitastor_getlength,
+    .bdrv_probe_blocksizes          = vitastor_probe_blocksizes,
+    .bdrv_refresh_limits            = vitastor_refresh_limits,
+
+    // FIXME: Implement it along with per-inode statistics
+    //.bdrv_get_allocated_file_size   = vitastor_get_allocated_file_size,
+
+    .bdrv_file_open                 = vitastor_file_open,
+    .bdrv_close                     = vitastor_close,
+
+    // Option list for the create operation
+    .create_opts                    = &vitastor_create_opts,
+
+    // For qmp_blockdev_create(), used by the qemu monitor / QAPI
+    // Requires patching QAPI IDL, thus unimplemented
+    //.bdrv_co_create                 = vitastor_co_create,
+
+    // For bdrv_create(), used by qemu-img
+    .bdrv_co_create_opts            = vitastor_co_create_opts,
+
+    .bdrv_co_truncate               = vitastor_co_truncate,
+
+    .bdrv_co_preadv                 = vitastor_co_preadv,
+    .bdrv_co_pwritev                = vitastor_co_pwritev,
+    .bdrv_co_flush_to_disk          = vitastor_co_flush,
+
+#if QEMU_VERSION_MAJOR >= 4
+    .strong_runtime_opts            = vitastor_strong_runtime_opts,
+#endif
+};
+
+static void vitastor_block_init(void)
+{
+    bdrv_register(&bdrv_vitastor);
+}
+
+block_init(vitastor_block_init);

qemu_proxy.cpp

@@ -0,0 +1,130 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+// C-C++ proxy for the QEMU driver
+// (QEMU headers don't compile with g++)
+
+#include <sys/epoll.h>
+
+#include "cluster_client.h"
+
+typedef void* AioContext;
+#include "qemu_proxy.h"
+
+extern "C"
+{
+    // QEMU
+    typedef void IOHandler(void *opaque);
+    void aio_set_fd_handler(AioContext *ctx, int fd, int is_external, IOHandler *fd_read, IOHandler *fd_write, void *poll_fn, void *opaque);
+}
+
+struct QemuProxyData
+{
+    int fd;
+    std::function<void(int, int)> callback;
+};
+
+class QemuProxy
+{
+    std::map<int, QemuProxyData> handlers;
+
+public:
+
+    timerfd_manager_t *tfd;
+    cluster_client_t *cli;
+    AioContext *ctx;
+
+    QemuProxy(AioContext *ctx, const char *etcd_host, const char *etcd_prefix)
+    {
+        this->ctx = ctx;
+        json11::Json cfg = json11::Json::object {
+            { "etcd_address", std::string(etcd_host) },
+            { "etcd_prefix", std::string(etcd_prefix ? etcd_prefix : "/vitastor") },
+        };
+        tfd = new timerfd_manager_t([this](int fd, bool wr, std::function<void(int, int)> callback) { set_fd_handler(fd, wr, callback); });
+        cli = new cluster_client_t(NULL, tfd, cfg);
+    }
+
+    ~QemuProxy()
+    {
+        cli->stop();
+        delete cli;
+        delete tfd;
+    }
+
+    void set_fd_handler(int fd, bool wr, std::function<void(int, int)> callback)
+    {
+        if (callback != NULL)
+        {
+            handlers[fd] = { .fd = fd, .callback = callback };
+            aio_set_fd_handler(ctx, fd, false, &QemuProxy::read_handler, wr ? &QemuProxy::write_handler : NULL, NULL, &handlers[fd]);
+        }
+        else
+        {
+            handlers.erase(fd);
+            aio_set_fd_handler(ctx, fd, false, NULL, NULL, NULL, NULL);
+        }
+    }
+
+    static void read_handler(void *opaque)
+    {
+        QemuProxyData *data = (QemuProxyData *)opaque;
+        data->callback(data->fd, EPOLLIN);
+    }
+
+    static void write_handler(void *opaque)
+    {
+        QemuProxyData *data = (QemuProxyData *)opaque;
+        data->callback(data->fd, EPOLLOUT);
+    }
+};
+
+extern "C" {
+
+void* vitastor_proxy_create(AioContext *ctx, const char *etcd_host, const char *etcd_prefix)
+{
+    QemuProxy *p = new QemuProxy(ctx, etcd_host, etcd_prefix);
+    return p;
+}
+
+void vitastor_proxy_destroy(void *client)
+{
+    QemuProxy *p = (QemuProxy*)client;
+    delete p;
+}
+
+void vitastor_proxy_rw(int write, void *client, uint64_t inode, uint64_t offset, uint64_t len,
+    iovec *iov, int iovcnt, VitastorIOHandler cb, void *opaque)
+{
+    QemuProxy *p = (QemuProxy*)client;
+    cluster_op_t *op = new cluster_op_t;
+    op->opcode = write ? OSD_OP_WRITE : OSD_OP_READ;
+    op->inode = inode;
+    op->offset = offset;
+    op->len = len;
+    for (int i = 0; i < iovcnt; i++)
+    {
+        op->iov.push_back(iov[i].iov_base, iov[i].iov_len);
+    }
+    op->callback = [cb, opaque](cluster_op_t *op)
+    {
+        cb(op->retval, opaque);
+        delete op;
+    };
+    p->cli->execute(op);
+}
+
+void vitastor_proxy_sync(void *client, VitastorIOHandler cb, void *opaque)
+{
+    QemuProxy *p = (QemuProxy*)client;
+    cluster_op_t *op = new cluster_op_t;
+    op->opcode = OSD_OP_SYNC;
+    op->callback = [cb, opaque](cluster_op_t *op)
+    {
+        cb(op->retval, opaque);
+        delete op;
+    };
+    p->cli->execute(op);
+}
+
+}

qemu_proxy.h

@@ -0,0 +1,29 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+#ifndef VITASTOR_QEMU_PROXY_H
+#define VITASTOR_QEMU_PROXY_H
+
+#ifndef POOL_ID_BITS
+#define POOL_ID_BITS 16
+#endif
+#include <stdint.h>
+#include <sys/uio.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Our exports
+typedef void VitastorIOHandler(int retval, void *opaque);
+void* vitastor_proxy_create(AioContext *ctx, const char *etcd_host, const char *etcd_prefix);
+void vitastor_proxy_destroy(void *client);
+void vitastor_proxy_rw(int write, void *client, uint64_t inode, uint64_t offset, uint64_t len,
+    struct iovec *iov, int iovcnt, VitastorIOHandler cb, void *opaque);
+void vitastor_proxy_sync(void *client, VitastorIOHandler cb, void *opaque);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

ringloop.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #include "ringloop.h"
 
 ring_loop_t::ring_loop_t(int qd)

ringloop.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #pragma once
 
 #ifndef _LARGEFILE64_SOURCE

rw_blocking.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #include <errno.h>
 #include <stdlib.h>
 #include <stdio.h>
@@ -51,6 +54,40 @@ int write_blocking(int fd, void *write_buf, size_t remaining)
     return done;
 }
 
+int readv_blocking(int fd, iovec *iov, int iovcnt)
+{
+    int v = 0;
+    int done = 0;
+    while (v < iovcnt)
+    {
+        ssize_t r = readv(fd, iov, iovcnt);
+        if (r < 0)
+        {
+            if (errno != EAGAIN && errno != EPIPE)
+            {
+                perror("writev");
+                exit(1);
+            }
+            continue;
+        }
+        while (v < iovcnt)
+        {
+            if (iov[v].iov_len > r)
+            {
+                iov[v].iov_len -= r;
+                iov[v].iov_base += r;
+                break;
+            }
+            else
+            {
+                v++;
+            }
+        }
+        done += r;
+    }
+    return done;
+}
+
 int writev_blocking(int fd, iovec *iov, int iovcnt)
 {
     int v = 0;

rw_blocking.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 #pragma once
 
 #include <unistd.h>
@@ -5,4 +8,5 @@
 
 int read_blocking(int fd, void *read_buf, size_t remaining);
 int write_blocking(int fd, void *write_buf, size_t remaining);
+int readv_blocking(int fd, iovec *iov, int iovcnt);
 int writev_blocking(int fd, iovec *iov, int iovcnt);

stub_bench.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 /**
  * Stub benchmarker
  */
@@ -25,20 +28,37 @@ int connect_stub(const char *server_address, int server_port);
 
 void run_bench(int peer_fd);
 
+static uint64_t read_sum = 0, read_count = 0;
 static uint64_t write_sum = 0, write_count = 0;
 static uint64_t sync_sum = 0, sync_count = 0;
 
 void handle_sigint(int sig)
 {
-    printf("4k randwrite: %lu us avg\n", write_sum/write_count);
+    printf("4k randread: %lu us avg\n", read_count ? read_sum/read_count : 0);
-    printf("sync: %lu us avg\n", sync_sum/sync_count);
+    printf("4k randwrite: %lu us avg\n", write_count ? write_sum/write_count : 0);
+    printf("sync: %lu us avg\n", sync_count ? sync_sum/sync_count : 0);
     exit(0);
 }
 
 int main(int narg, char *args[])
 {
+    if (narg < 2)
+    {
+        printf("USAGE: %s SERVER_IP [PORT]\n", args[0]);
+        return 1;
+    }
+    int port = 11203;
+    if (narg >= 3)
+    {
+        port = atoi(args[2]);
+        if (port <= 0 || port >= 65536)
+        {
+            printf("Bad port number\n");
+            return 1;
+        }
+    }
     signal(SIGINT, handle_sigint);
-    int peer_fd = connect_stub("127.0.0.1", 11203);
+    int peer_fd = connect_stub(args[1], port);
     run_bench(peer_fd);
     close(peer_fd);
     return 0;
@@ -98,14 +118,41 @@ void run_bench(int peer_fd)
     osd_any_reply_t reply;
     void *buf = NULL;
     int r;
+    iovec iov[2];
     timespec tv_begin, tv_end;
     clock_gettime(CLOCK_REALTIME, &tv_begin);
     while (1)
     {
+        // read
+        op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
+        op.hdr.id = 1;
+        op.hdr.opcode = OSD_OP_SEC_READ;
+        op.sec_rw.oid.inode = 3;
+        op.sec_rw.oid.stripe = (rand() << 17) % (1 << 29); // 512 MB
+        op.sec_rw.version = 0;
+        op.sec_rw.len = 4096;
+        op.sec_rw.offset = (rand() * op.sec_rw.len) % (1 << 17);
+        r = write_blocking(peer_fd, op.buf, OSD_PACKET_SIZE) == OSD_PACKET_SIZE;
+        if (!r)
+            break;
+        buf = malloc(op.sec_rw.len);
+        iov[0] = { reply.buf, OSD_PACKET_SIZE };
+        iov[1] = { buf, op.sec_rw.len };
+        r = readv_blocking(peer_fd, iov, 2) == (OSD_PACKET_SIZE + op.sec_rw.len);
+        free(buf);
+        if (!r || !check_reply(OSD_PACKET_SIZE, op, reply, op.sec_rw.len))
+            break;
+        clock_gettime(CLOCK_REALTIME, &tv_end);
+        read_count++;
+        read_sum += (
+            (tv_end.tv_sec - tv_begin.tv_sec)*1000000 +
+            tv_end.tv_nsec/1000 - tv_begin.tv_nsec/1000
+        );
+        tv_begin = tv_end;
         // write
         op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
         op.hdr.id = 1;
-        op.hdr.opcode = OSD_OP_SECONDARY_WRITE;
+        op.hdr.opcode = OSD_OP_SEC_WRITE;
         op.sec_rw.oid.inode = 3;
         op.sec_rw.oid.stripe = (rand() << 17) % (1 << 29); // 512 MB
         op.sec_rw.version = 0;
@@ -113,9 +160,9 @@ void run_bench(int peer_fd)
         op.sec_rw.offset = (rand() * op.sec_rw.len) % (1 << 17);
         buf = malloc(op.sec_rw.len);
         memset(buf, rand() % 255, op.sec_rw.len);
-        r = write_blocking(peer_fd, op.buf, OSD_PACKET_SIZE) == OSD_PACKET_SIZE;
+        iov[0] = { op.buf, OSD_PACKET_SIZE };
-        if (r)
+        iov[1] = { buf, op.sec_rw.len };
-            r = write_blocking(peer_fd, buf, op.sec_rw.len) == op.sec_rw.len;
+        r = writev_blocking(peer_fd, iov, 2) == (OSD_PACKET_SIZE + op.sec_rw.len);
         free(buf);
         if (!r)
             break;
@@ -128,6 +175,7 @@ void run_bench(int peer_fd)
             (tv_end.tv_sec - tv_begin.tv_sec)*1000000 +
             tv_end.tv_nsec/1000 - tv_begin.tv_nsec/1000
         );
+        tv_begin = tv_end;
         // sync/stab
         op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
         op.hdr.id = 1;
@@ -138,11 +186,12 @@ void run_bench(int peer_fd)
         r = read_blocking(peer_fd, reply.buf, OSD_PACKET_SIZE);
         if (!check_reply(r, op, reply, 0))
             break;
-        clock_gettime(CLOCK_REALTIME, &tv_begin);
+        clock_gettime(CLOCK_REALTIME, &tv_end);
         sync_count++;
         sync_sum += (
-            (tv_begin.tv_sec - tv_end.tv_sec)*1000000 +
+            (tv_end.tv_sec - tv_begin.tv_sec)*1000000 +
-            tv_begin.tv_nsec/1000 - tv_end.tv_nsec/1000
+            tv_end.tv_nsec/1000 - tv_begin.tv_nsec/1000
         );
+        tv_begin = tv_end;
     }
 }

stub_osd.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
 /**
  * Stub "OSD" to test & compare network performance with sync read/write and io_uring
  *
@@ -130,7 +133,7 @@ void run_stub(int peer_fd)
         reply.hdr.magic = SECONDARY_OSD_REPLY_MAGIC;
         reply.hdr.id = op.hdr.id;
         reply.hdr.opcode = op.hdr.opcode;
-        if (op.hdr.opcode == OSD_OP_SECONDARY_READ)
+        if (op.hdr.opcode == OSD_OP_SEC_READ)
         {
             reply.hdr.retval = op.sec_rw.len;
             buf = malloc(op.sec_rw.len);
@@ -141,7 +144,7 @@ void run_stub(int peer_fd)
             if (r < op.sec_rw.len)
                 break;
         }
-        else if (op.hdr.opcode == OSD_OP_SECONDARY_WRITE)
+        else if (op.hdr.opcode == OSD_OP_SEC_WRITE || op.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
         {
             buf = malloc(op.sec_rw.len);
             r = read_blocking(peer_fd, buf, op.sec_rw.len);

stub_uring_osd.cpp

@@ -0,0 +1,131 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+
+/**
+ * Stub "OSD" implemented on top of osd_messenger to test & compare
+ * network performance with sync read/write and io_uring
+ */
+
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <netinet/tcp.h>
+#include <arpa/inet.h>
+#include <string.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <stdlib.h>
+
+#include <stdexcept>
+
+#include "ringloop.h"
+#include "epoll_manager.h"
+#include "messenger.h"
+
+int bind_stub(const char *bind_address, int bind_port);
+
+void stub_exec_op(osd_messenger_t *msgr, osd_op_t *op);
+
+int main(int narg, char *args[])
+{
+    ring_consumer_t looper;
+    ring_loop_t *ringloop = new ring_loop_t(512);
+    epoll_manager_t *epmgr = new epoll_manager_t(ringloop);
+    osd_messenger_t *msgr = new osd_messenger_t();
+    msgr->osd_num = 1351;
+    msgr->tfd = epmgr->tfd;
+    msgr->ringloop = ringloop;
+    msgr->repeer_pgs = [](osd_num_t) {};
+    msgr->exec_op = [msgr](osd_op_t *op) { stub_exec_op(msgr, op); };
+    // Accept new connections
+    int listen_fd = bind_stub("0.0.0.0", 11203);
+    epmgr->set_fd_handler(listen_fd, false, [listen_fd, msgr](int fd, int events)
+    {
+        msgr->accept_connections(listen_fd);
+    });
+    looper.loop = [msgr, ringloop]()
+    {
+        msgr->read_requests();
+        msgr->send_replies();
+        ringloop->submit();
+    };
+    ringloop->register_consumer(&looper);
+    printf("stub_uring_osd: waiting for clients\n");
+    while (true)
+    {
+        ringloop->loop();
+        ringloop->wait();
+    }
+    delete msgr;
+    delete epmgr;
+    delete ringloop;
+    return 0;
+}
+
+int bind_stub(const char *bind_address, int bind_port)
+{
+    int listen_backlog = 128;
+
+    int listen_fd = socket(AF_INET, SOCK_STREAM, 0);
+    if (listen_fd < 0)
+    {
+        throw std::runtime_error(std::string("socket: ") + strerror(errno));
+    }
+    int enable = 1;
+    setsockopt(listen_fd, SOL_SOCKET, SO_REUSEADDR, &enable, sizeof(enable));
+
+    sockaddr_in addr;
+    int r;
+    if ((r = inet_pton(AF_INET, bind_address, &addr.sin_addr)) != 1)
+    {
+        close(listen_fd);
+        throw std::runtime_error("bind address "+std::string(bind_address)+(r == 0 ? " is not valid" : ": no ipv4 support"));
+    }
+    addr.sin_family = AF_INET;
+    addr.sin_port = htons(bind_port);
+
+    if (bind(listen_fd, (sockaddr*)&addr, sizeof(addr)) < 0)
+    {
+        close(listen_fd);
+        throw std::runtime_error(std::string("bind: ") + strerror(errno));
+    }
+
+    if (listen(listen_fd, listen_backlog) < 0)
+    {
+        close(listen_fd);
+        throw std::runtime_error(std::string("listen: ") + strerror(errno));
+    }
+
+    fcntl(listen_fd, F_SETFL, fcntl(listen_fd, F_GETFL, 0) | O_NONBLOCK);
+
+    return listen_fd;
+}
+
+void stub_exec_op(osd_messenger_t *msgr, osd_op_t *op)
+{
+    op->reply.hdr.magic = SECONDARY_OSD_REPLY_MAGIC;
+    op->reply.hdr.id = op->req.hdr.id;
+    op->reply.hdr.opcode = op->req.hdr.opcode;
+    if (op->req.hdr.opcode == OSD_OP_SEC_READ)
+    {
+        op->reply.hdr.retval = op->req.sec_rw.len;
+        op->buf = malloc(op->req.sec_rw.len);
+        op->iov.push_back(op->buf, op->req.sec_rw.len);
+    }
+    else if (op->req.hdr.opcode == OSD_OP_SEC_WRITE || op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
+    {
+        op->reply.hdr.retval = op->req.sec_rw.len;
+    }
+    else if (op->req.hdr.opcode == OSD_OP_TEST_SYNC_STAB_ALL)
+    {
+        op->reply.hdr.retval = 0;
+    }
+    else
+    {
+        printf("client %d: unsupported stub opcode: %lu\n", op->peer_fd, op->req.hdr.opcode);
+        op->reply.hdr.retval = -EINVAL;
+    }
+    msgr->outbox_push(op);
+}

test.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #define _LARGEFILE64_SOURCE
 #include <sys/types.h>
 #include <sys/ioctl.h>

test_allocator.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <stdio.h>
 #include "allocator.h"
 

test_blockstore.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #include <malloc.h>
 #include "timerfd_interval.h"
 #include "blockstore.h"

test_pattern.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.0 (see README.md for details)
+
 #pragma once
 
 #include <assert.h>