Release 0.5.10

The version seems to be stable after this bunch of fixes :)

- Fix delete & write operation ordering during rebalance to not lose objects in the immediate_commit=off mode
- Fix a possible crash caused by very high iodepths
- Re-distribute PG primaries over OSDs that come up after a short downtime
- Allow to specify etcd URLs for OSDs with http://, do not die with a strange error if -etcd option is missing for fio
- Fix a journal flushing deadlock which sometimes occurred in the immediate_commit=off mode
- Fix a bug where OSDs could hang if the data device filled up
- Fix an allocator bug where it was unable to allocate up to last (n%64) data device blocks
- Fix monitor crash that occurred on removal of some etcd keys
- Fix a bug where PGs could remain incomplete due to incorrect PG history with just zeroes in osd_sets

.dockerignore

@@ -0,0 +1,19 @@
+.git
+build
+packages
+mon/node_modules
+*.o
+*.so
+osd
+stub_osd
+stub_uring_osd
+stub_bench
+osd_test
+dump_journal
+nbd_proxy
+rm_inode
+fio
+qemu
+rpm/*.Dockerfile
+debian/*.Dockerfile
+Dockerfile

.gitignore

@@ -0,0 +1,18 @@
+*.o
+*.so
+package-lock.json
+fio
+qemu
+osd
+stub_osd
+stub_uring_osd
+stub_bench
+osd_test
+osd_peering_pg_test
+dump_journal
+nbd_proxy
+rm_inode
+test_allocator
+test_blockstore
+test_shit
+osd_rmw_test

CMakeLists.txt

@@ -0,0 +1,5 @@
+cmake_minimum_required(VERSION 2.8)
+
+project(vitastor)
+
+add_subdirectory(src)

Make-gen.pl

@@ -1,46 +0,0 @@
-#!/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

@@ -1,169 +0,0 @@
-BLOCKSTORE_OBJS := allocator.o blockstore.o blockstore_impl.o blockstore_init.o blockstore_open.o blockstore_journal.o blockstore_read.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: 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
-
-dump_journal: dump_journal.cpp crc32c.o blockstore_journal.h
-	g++ $(CXXFLAGS) -o $@ $< crc32c.o
-
-libblockstore.so: $(BLOCKSTORE_OBJS)
-	g++ $(CXXFLAGS) -o $@ -shared $(BLOCKSTORE_OBJS) -ltcmalloc_minimal -luring
-libfio_blockstore.so: ./libblockstore.so fio_engine.o json11.o
-	g++ $(CXXFLAGS) -shared -o $@ fio_engine.o json11.o ./libblockstore.so -ltcmalloc_minimal -luring
-
-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 messenger.o osd_cluster.o http_client.o osd_ops.o pg_states.o \
-	osd_rmw.o json11.o base64.o timerfd_manager.o epoll_manager.o
-osd: ./libblockstore.so osd_main.cpp osd.h osd_ops.h $(OSD_OBJS)
-	g++ $(CXXFLAGS) -o $@ osd_main.cpp $(OSD_OBJS) ./libblockstore.so -ltcmalloc_minimal -luring
-
-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.cpp rw_blocking.o -ltcmalloc_minimal
-osd_test: osd_test.cpp osd_ops.h rw_blocking.o
-	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.o rw_blocking.o
-	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
-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

@@ -16,12 +16,13 @@ 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)
+- Multiple redundancy schemes: Replication, XOR n+1, Reed-Solomon erasure codes
+  based on jerasure library with any number of data and parity drives in a group
 - 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
+  - Placement tree, OSD selection by tags (device classes) and placement root
   - Configurable failure domains
 - Recovery of degraded blocks
 - Rebalancing (data movement between OSDs)
@@ -31,24 +32,24 @@ breaking changes in the future. However, the following is implemented:
 - QEMU driver (built out-of-tree)
 - Loadable fio engine for benchmarks (also built out-of-tree)
 - NBD proxy for kernel mounts
+- Inode removal tool (vitastor-rm)
+- Packaging for Debian and CentOS
 
 ## 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
+- Scrubbing without checksums (verification of replicas)
 - Checksums
 - SSD+HDD optimizations, possibly including tiered storage and soft journal flushes
 - RDMA and NVDIMM support
+- Web GUI
 - Compression (possibly)
 - Read caching using system page cache (possibly)
 
@@ -80,7 +81,7 @@ Architectural differences from Ceph:
   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
+  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.
@@ -152,9 +153,9 @@ I use the following 6 commands with small variations to benchmark any storage:
   `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):
+- Random write latency (T1Q1, 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:
+- Random read latency (T1Q1):
   `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`
@@ -205,7 +206,7 @@ Hardware configuration: 4 nodes, each with:
 
 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.
+All of the results below apply to 4 KB blocks and random access (unless indicated otherwise).
 
 Raw drive performance:
 - T1Q1 write ~27000 iops (~0.037ms latency)
@@ -232,6 +233,8 @@ Vitastor:
 - 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
+- Linear write (4M T1Q32): 2800 MB/s
+- Linear read (4M T1Q32): 1500 MB/s
 
 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.
@@ -245,6 +248,24 @@ Vitastor was configured with: `--disable_data_fsync true --immediate_commit all
   --journal_no_same_sector_overwrites true --journal_sector_buffer_count 1024
   --journal_size 16777216`.
 
+### EC/XOR 2+1
+
+Vitastor:
+- T1Q1 write: 2808 iops (~0.355ms latency)
+- T1Q1 read: 6190 iops (~0.16ms latency)
+- T2Q64 write: 85500 iops, total CPU usage by OSDs about 3.4 virtual cores on each node
+- T8Q64 read: 812000 iops, total CPU usage by OSDs about 4.7 virtual cores on each node
+- Linear write (4M T1Q32): 3200 MB/s
+- Linear read (4M T1Q32): 1800 MB/s
+
+Ceph:
+- T1Q1 write: 730 iops (~1.37ms latency)
+- T1Q1 read: 1500 iops with cold cache (~0.66ms latency), 2300 iops after 2 minute metadata cache warmup (~0.435ms latency)
+- T4Q128 write (4 RBD images): 45300 iops, total CPU usage by OSDs about 30 virtual cores on each node
+- T8Q64 read (4 RBD images): 278600 iops, total CPU usage by OSDs about 40 virtual cores on each node
+- Linear write (4M T1Q32): 1950 MB/s before preallocation, 2500 MB/s after preallocation
+- Linear read (4M T1Q32): 2400 MB/s
+
 ### NBD
 
 NBD is currently required to mount Vitastor via kernel, but it imposes additional overhead
@@ -256,19 +277,50 @@ Vitastor with single-thread NBD on the same hardware:
 - 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)
+- Linear write (4M T1Q128): 1266 MB/s (compared to 2800 MB/s via fio)
+- Linear read (4M T1Q128): 975 MB/s (compared to 1500 MB/s via fio)
 
-## Building
+## Installation
 
-- Install Linux kernel 5.4 or newer for io_uring support.
+### Debian
+
+- Trust Vitastor package signing key:
+  `wget -q -O - https://vitastor.io/debian/pubkey | sudo apt-key add -`
+- Add Vitastor package repository to your /etc/apt/sources.list:
+  - Debian 11 (Bullseye/Sid): `deb https://vitastor.io/debian bullseye main`
+  - Debian 10 (Buster): `deb https://vitastor.io/debian buster main`
+- For Debian 10 (Buster) also enable backports repository:
+  `deb http://deb.debian.org/debian buster-backports main`
+- Install packages: `apt update; apt install vitastor lp-solve etcd linux-image-amd64`
+
+### CentOS
+
+- Add Vitastor package repository:
+  - CentOS 7: `yum install https://vitastor.io/rpms/centos/7/vitastor-release-1.0-1.el7.noarch.rpm`
+  - CentOS 8: `dnf install https://vitastor.io/rpms/centos/8/vitastor-release-1.0-1.el8.noarch.rpm`
+- Enable EPEL: `yum/dnf install epel-release`
+- Enable additional CentOS repositories:
+  - CentOS 7: `yum install centos-release-scl`
+  - CentOS 8: `dnf install centos-release-advanced-virtualization`
+- Enable elrepo-kernel:
+  - CentOS 7: `yum install https://www.elrepo.org/elrepo-release-7.el7.elrepo.noarch.rpm`
+  - CentOS 8: `dnf install https://www.elrepo.org/elrepo-release-8.el8.elrepo.noarch.rpm`
+- Install packages: `yum/dnf install vitastor lpsolve etcd kernel-ml qemu-kvm`
+
+### Building from Source
+
+- Install Linux kernel 5.4 or newer, for io_uring support. 5.8 or later is highly recommended because
+  there is at least one known io_uring hang with 5.4 and an HP SmartArray controller.
 - 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.
+- Install etcd. Attention: you need a fixed version from here: https://github.com/vitalif/etcd/,
+  branch release-3.4, because there is a bug in upstream etcd which makes Vitastor OSDs fail to
+  move PGs out of "starting" state if you have at least around ~500 PGs or so. The custom build
+  will be unnecessary when etcd merges the fix: https://github.com/etcd-io/etcd/pull/12402.
+- Install node.js 10 or newer.
+- Install gcc and g++ 8.x or newer.
 - 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:
+- Install QEMU 3.0+, 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
@@ -278,13 +330,14 @@ Vitastor with single-thread NBD on the same hardware:
       * 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`
+      * For QEMU 3.0+: `<qemu>/qapi` &rarr; `<vitastor>/qemu/b/qemu/qapi`
+      * For QEMU 2.0+: `<qemu>/qapi-types.h` &rarr; `<vitastor>/qemu/b/qemu/qapi-types.h`
    - `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.
+- You can also rebuild QEMU with a patch that makes LD_PRELOAD unnecessary to load vitastor driver.
+  See `qemu-*.*-vitastor.patch`.
+- Install fio 3.7 or later, get its source and symlink it into `<vitastor>/fio`.
+- Build & install Vitastor with `mkdir build && cd build && cmake .. && make -j8 && make install`.
+  Pay attention to the `QEMU_PLUGINDIR` cmake option - it must be set to `qemu-kvm` on RHEL.
 
 ## Running
 
@@ -295,56 +348,56 @@ and calculate disk offsets almost by hand. This will be fixed in near future.
   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:
+- Check `/usr/lib/vitastor/mon/make-units.sh` and `/usr/lib/vitastor/mon/make-osd.sh` and
+  put desired values into the variables at the top of these files.
+- Create systemd units for the monitor and etcd: `/usr/lib/vitastor/mon/make-units.sh`
+- Create systemd units for your OSDs: `/usr/lib/vitastor/mon/make-osd.sh /dev/disk/by-partuuid/XXX [/dev/disk/by-partuuid/YYY ...]`
+- You can edit the units and change OSD configuration. Notable configuration variables:
   - `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.
+  - `flusher_count 256` - flusher is a micro-thread that removes old data from the journal.
+    You don't have to worry about this parameter anymore, 256 is enough.
   - `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.
+    Most (99%) SSDs don't need this option. But Intel D3-4510 does because it doesn't like when you
+    overwrite the same sector twice in a short period of time. The setting forces Vitastor to never
+    overwrite the same journal sector twice in a row which makes D3-4510 almost happy. Not totally
+    happy, because overwrites of the same block can still happen in the metadata area... 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`.
+- Create global configuration in etcd: `etcdctl --endpoints=... put /vitastor/config/global '{"immediate_commit":"all"}'`
+  (if all your drives have capacitors).
+- Create pool configuration in etcd: `etcdctl --endpoints=... put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"replicated","pg_size":2,"pg_minsize":1,"pg_count":256,"failure_domain":"host"}}'`.
+  For jerasure pools the configuration should look like the following: `2:{"name":"ecpool","scheme":"jerasure","pg_size":4,"parity_chunks":2,"pg_minsize":2,"pg_count":256,"failure_domain":"host"}`.
 - 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`.
+- You can check PG states with `etcdctl --endpoints=... get --prefix /vitastor/pg/state`. All PGs should become 'active'.
+- Run tests with (for example): `fio -thread -ioengine=libfio_vitastor.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'
+  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'
   ```
+  Note that the command requires to be run with `LD_PRELOAD=/usr/lib/x86_64-linux-gnu/qemu/block-vitastor.so qemu-img ...`
+  if you use unmodified QEMU.
 - Run QEMU with (for example):
   ```
-  LD_PRELOAD=./qemu_driver.so qemu-system-x86_64 -enable-kvm -m 1024
+  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
   ```
+- Remove inode with (for example):
+  ```
+  vitastor-rm --etcd_address 10.115.0.10:2379/v3 --pool 1 --inode 1 --parallel_osds 16 --iodepth 32
+  ```
 
 ## 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.
+- Object deletion requests may currently lead to 'incomplete' objects if your OSDs crash during
+  deletion because proper handling of object cleanup in a cluster should be "three-phase"
+  and it's currently not implemented. Just to repeat the removal again in this case.
 
 ## Implementation Principles
 
@@ -366,22 +419,27 @@ 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
+Vitastor Network Public License 1.1 (VNPL 1.1), 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.
+through a computer network and expressly designed to be used in conjunction
+with it ("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.
 
+Please note that VNPL doesn't require you to open the code of proprietary
+software running inside a VM if it's not specially designed to be used with
+Vitastor.
+
 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
+VNPL 1.1 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).
+You can find the full text of VNPL-1.1 in the file [VNPL-1.1.txt](VNPL-1.1.txt).
 GPL 2.0 is also included in this repository as [GPL-2.0.txt](GPL-2.0.txt).

VNPL-1.1.txt

@@ -1,7 +1,7 @@
                      VITASTOR NETWORK PUBLIC LICENSE
-                       Version 1, 17 September 2020
+                     Version 1.1,  6 February 2021
 
- Copyright (C) 2020 Vitaliy Filippov <vitalif@yourcmc.ru>
+ Copyright (C) 2021 Vitaliy Filippov <vitalif@yourcmc.ru>
  Everyone is permitted to copy and distribute verbatim copies
  of this license document, but changing it is not allowed.
 
@@ -540,12 +540,15 @@ 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,
+  A "Proxy Program" means a separate program which is specially designed to
+be used in conjunction with the covered work and interacts with it directly
+or indirectly through any kind of API (application programming interfaces),
+a computer network, an imitation of such network, or another Proxy Program
+itself.
+
+  Notwithstanding any other provision of this License, if you provide any user
+with an opportunity to interact with the covered work through a computer
+network, an imitation of such network, or any number of "Proxy Programs",
 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

copy-fio-includes.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+
+gcc -I. -E -o fio_headers.i src/fio_headers.h
+
+rm -rf fio-copy
+for i in `grep -Po 'fio/[^"]+' fio_headers.i | sort | uniq`; do
+    j=${i##fio/}
+    p=$(dirname $j)
+    mkdir -p fio-copy/$p
+    cp $i fio-copy/$j
+done
+
+rm fio_headers.i

copy-qemu-includes.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+#cd qemu
+#debian/rules b/configure-stamp
+#cd b/qemu; make qapi
+
+gcc -I qemu/b/qemu `pkg-config glib-2.0 --cflags` \
+    -I qemu/include -E -o qemu_driver.i src/qemu_driver.c
+
+rm -rf qemu-copy
+for i in `grep -Po 'qemu/[^"]+' qemu_driver.i | sort | uniq`; do
+    j=${i##qemu/}
+    p=$(dirname $j)
+    mkdir -p qemu-copy/$p
+    cp $i qemu-copy/$j
+done
+
+rm qemu_driver.i

debian/build-vitastor-bullseye.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+
+sed 's/$REL/bullseye/g' < vitastor.Dockerfile > ../Dockerfile
+cd ..
+mkdir -p packages
+sudo podman build -v `pwd`/packages:/root/packages -f Dockerfile .
+rm Dockerfile

debian/build-vitastor-buster.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+
+sed 's/$REL/buster/g' < vitastor.Dockerfile > ../Dockerfile
+cd ..
+mkdir -p packages
+sudo podman build -v `pwd`/packages:/root/packages -f Dockerfile .
+rm Dockerfile

debian/changelog

@@ -0,0 +1,17 @@
+vitastor (0.5.10-1) unstable; urgency=medium
+
+  * Bugfixes
+
+ -- Vitaliy Filippov <vitalif@yourcmc.ru>  Tue, 02 Feb 2021 23:01:24 +0300
+
+vitastor (0.5.1-1) unstable; urgency=medium
+
+  * Add jerasure support
+
+ -- Vitaliy Filippov <vitalif@yourcmc.ru>  Sat, 05 Dec 2020 17:02:26 +0300
+
+vitastor (0.5-1) unstable; urgency=medium
+
+  * First packaging for Debian
+
+ -- Vitaliy Filippov <vitalif@yourcmc.ru>  Thu, 05 Nov 2020 02:20:59 +0300

debian/compat

@@ -0,0 +1 @@
+13

debian/control

@@ -0,0 +1,17 @@
+Source: vitastor
+Section: admin
+Priority: optional
+Maintainer: Vitaliy Filippov <vitalif@yourcmc.ru>
+Build-Depends: debhelper, liburing-dev (>= 0.6), g++ (>= 8), libstdc++6 (>= 8), linux-libc-dev, libgoogle-perftools-dev, libjerasure-dev, libgf-complete-dev
+Standards-Version: 4.5.0
+Homepage: https://vitastor.io/
+Rules-Requires-Root: no
+
+Package: vitastor
+Architecture: amd64
+Depends: ${shlibs:Depends}, ${misc:Depends}, fio (= ${dep:fio}), qemu (= ${dep:qemu}), nodejs (>= 10), node-sprintf-js, node-ws (>= 7), libjerasure2, lp-solve
+Description: Vitastor, a fast software-defined clustered block storage
+ 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).

debian/copyright

@@ -0,0 +1,21 @@
+Format: https://www.debian.org/doc/packaging-manuals/copyright-format/1.0/
+Upstream-Name: vitastor
+Upstream-Contact: Vitaliy Filippov <vitalif@yourcmc.ru>
+Source: https://vitastor.io
+
+Files: *
+Copyright: 2019+ Vitaliy Filippov <vitalif@yourcmc.ru>
+License: Multiple licenses VNPL-1.1 and/or GPL-2.0+
+ All server-side code (OSD, Monitor and so on) is licensed under the terms of
+ Vitastor Network Public License 1.1 (VNPL 1.1), 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 and expressly designed to be used in conjunction
+ with it ("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.
+ .
+ Client libraries (cluster_client and so on) are dual-licensed under the same
+ VNPL 1.1 and also GNU GPL 2.0 or later to allow for compatibility with GPLed
+ software like QEMU and fio.

debian/install

@@ -0,0 +1,3 @@
+VNPL-1.1.txt usr/share/doc/vitastor
+GPL-2.0.txt usr/share/doc/vitastor
+mon usr/lib/vitastor

debian/patched-qemu.Dockerfile

@@ -0,0 +1,44 @@
+# Build patched QEMU for Debian Buster or Bullseye/Sid inside a container
+# cd ..; podman build --build-arg REL=bullseye -v `pwd`/packages:/root/packages -f debian/patched-qemu.Dockerfile .
+
+FROM debian:$REL
+
+WORKDIR /root
+
+RUN if [ "$REL" = "buster" ]; then \
+        echo 'deb http://deb.debian.org/debian buster-backports main' >> /etc/apt/sources.list; \
+        echo >> /etc/apt/preferences; \
+        echo 'Package: *' >> /etc/apt/preferences; \
+        echo 'Pin: release a=buster-backports' >> /etc/apt/preferences; \
+        echo 'Pin-Priority: 500' >> /etc/apt/preferences; \
+    fi; \
+    grep '^deb ' /etc/apt/sources.list | perl -pe 's/^deb/deb-src/' >> /etc/apt/sources.list; \
+    echo 'APT::Install-Recommends false;' >> /etc/apt/apt.conf; \
+    echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
+
+RUN apt-get update
+RUN apt-get -y install qemu fio liburing1 liburing-dev libgoogle-perftools-dev devscripts
+RUN apt-get -y build-dep qemu
+RUN apt-get -y build-dep fio
+RUN apt-get --download-only source qemu
+RUN apt-get --download-only source fio
+
+ADD qemu-5.0-vitastor.patch qemu-5.1-vitastor.patch /root/vitastor/
+RUN set -e; \
+    mkdir -p /root/packages/qemu-$REL; \
+    rm -rf /root/packages/qemu-$REL/*; \
+    cd /root/packages/qemu-$REL; \
+    dpkg-source -x /root/qemu*.dsc; \
+    if [ -d /root/packages/qemu-$REL/qemu-5.0 ]; then \
+        cp /root/vitastor/qemu-5.0-vitastor.patch /root/packages/qemu-$REL/qemu-5.0/debian/patches; \
+        echo qemu-5.0-vitastor.patch >> /root/packages/qemu-$REL/qemu-5.0/debian/patches/series; \
+    else \
+        cp /root/vitastor/qemu-5.1-vitastor.patch /root/packages/qemu-$REL/qemu-*/debian/patches; \
+        P=`ls -d /root/packages/qemu-$REL/qemu-*/debian/patches`; \
+        echo qemu-5.1-vitastor.patch >> $P/series; \
+    fi; \
+    cd /root/packages/qemu-$REL/qemu-*/; \
+    V=$(head -n1 debian/changelog | perl -pe 's/^.*\((.*?)(~bpo[\d\+]*)?\).*$/$1/')+vitastor1; \
+    DEBFULLNAME="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v $V 'Plug Vitastor block driver'; \
+    DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \
+    rm -rf /root/packages/qemu-$REL/qemu-*/

debian/rules

@@ -0,0 +1,9 @@
+#!/usr/bin/make -f
+export DH_VERBOSE = 1
+
+%:
+	dh $@
+
+override_dh_installdeb:
+	cat debian/substvars >> debian/vitastor.substvars
+	dh_installdeb

debian/source/format

@@ -0,0 +1 @@
+3.0 (quilt)

debian/substvars

@@ -0,0 +1,2 @@
+dep:fio=3.16-1
+dep:qemu=1:5.1+dfsg-4+vitastor1

debian/vitastor.Dockerfile

@@ -0,0 +1,67 @@
+# Build Vitastor packages for Debian Buster or Bullseye/Sid inside a container
+# cd ..; podman build --build-arg REL=bullseye -v `pwd`/packages:/root/packages -f debian/vitastor.Dockerfile .
+
+FROM debian:$REL
+
+WORKDIR /root
+
+RUN if [ "$REL" = "buster" ]; then \
+        echo 'deb http://deb.debian.org/debian buster-backports main' >> /etc/apt/sources.list; \
+        echo >> /etc/apt/preferences; \
+        echo 'Package: *' >> /etc/apt/preferences; \
+        echo 'Pin: release a=buster-backports' >> /etc/apt/preferences; \
+        echo 'Pin-Priority: 500' >> /etc/apt/preferences; \
+    fi; \
+    grep '^deb ' /etc/apt/sources.list | perl -pe 's/^deb/deb-src/' >> /etc/apt/sources.list; \
+    echo 'APT::Install-Recommends false;' >> /etc/apt/apt.conf; \
+    echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
+
+RUN apt-get update
+RUN apt-get -y install qemu fio liburing1 liburing-dev libgoogle-perftools-dev devscripts
+RUN apt-get -y build-dep qemu
+RUN apt-get -y build-dep fio
+RUN apt-get --download-only source qemu
+RUN apt-get --download-only source fio
+RUN apt-get -y install libjerasure-dev cmake
+
+ADD . /root/vitastor
+RUN set -e -x; \
+    mkdir -p /root/fio-build/; \
+    cd /root/fio-build/; \
+    rm -rf /root/fio-build/*; \
+    dpkg-source -x /root/fio*.dsc; \
+    cd /root/packages/qemu-$REL/; \
+    rm -rf qemu*/; \
+    dpkg-source -x qemu*.dsc; \
+    cd /root/packages/qemu-$REL/qemu*/; \
+    debian/rules b/configure-stamp; \
+    cd b/qemu; \
+    make -j8 qapi/qapi-builtin-types.h; \
+    mkdir -p /root/packages/vitastor-$REL; \
+    rm -rf /root/packages/vitastor-$REL/*; \
+    cd /root/packages/vitastor-$REL; \
+    cp -r /root/vitastor vitastor-0.5.10; \
+    ln -s /root/packages/qemu-$REL/qemu-*/ vitastor-0.5.10/qemu; \
+    ln -s /root/fio-build/fio-*/ vitastor-0.5.10/fio; \
+    cd vitastor-0.5.10; \
+    FIO=$(head -n1 fio/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
+    QEMU=$(head -n1 qemu/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
+    sh copy-qemu-includes.sh; \
+    sh copy-fio-includes.sh; \
+    rm qemu fio; \
+    mkdir -p a b debian/patches; \
+    mv qemu-copy b/qemu; \
+    mv fio-copy b/fio; \
+    diff -NaurpbB a b > debian/patches/qemu-fio-headers.patch || true; \
+    echo qemu-fio-headers.patch >> debian/patches/series; \
+    rm -rf a b; \
+    rm -rf /root/packages/qemu-$REL/qemu*/; \
+    echo "dep:fio=$FIO" > debian/substvars; \
+    echo "dep:qemu=$QEMU" >> debian/substvars; \
+    cd /root/packages/vitastor-$REL; \
+    tar --sort=name --mtime='2020-01-01' --owner=0 --group=0 --exclude=debian -cJf vitastor_0.5.10.orig.tar.xz vitastor-0.5.10; \
+    cd vitastor-0.5.10; \
+    V=$(head -n1 debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
+    DEBFULLNAME="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v "$V""$REL" "Rebuild for $REL"; \
+    DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \
+    rm -rf /root/packages/vitastor-$REL/vitastor-*/

lambda_size.cpp

@@ -1,51 +0,0 @@
-// Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
-
-#include <iostream>
-#include <functional>
-#include <array>
-#include <cstdlib> // for malloc() and free()
-using namespace std;
-
-// replace operator new and delete to log allocations
-void* operator new(std::size_t n)
-{
-    cout << "Allocating " << n << " bytes" << endl;
-    return malloc(n);
-}
-
-void operator delete(void* p) throw()
-{
-    free(p);
-}
-
-class test
-{
-public:
-    std::string s;
-    void a(std::function<void()> & f, const char *str)
-    {
-        auto l = [this, str]() { cout << str << " ? " << s << " from this\n"; };
-        cout << "Assigning lambda3 of size " << sizeof(l) << endl;
-        f = l;
-    }
-};
-
-int main()
-{
-    std::array<char, 16> arr1;
-    auto lambda1 = [arr1](){};
-    cout << "Assigning lambda1 of size " << sizeof(lambda1) << endl;
-    std::function<void()> f1 = lambda1;
-
-    std::array<char, 17> arr2;
-    auto lambda2 = [arr2](){};
-    cout << "Assigning lambda2 of size " << sizeof(lambda2) << endl;
-    std::function<void()> f2 = lambda2;
-
-    test t;
-    std::function<void()> f3;
-    t.s = "str";
-    t.a(f3, "huyambda");
-    f3();
-}

mon/PGUtil.js

@@ -1,22 +1,59 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 module.exports = {
     scale_pg_count,
 };
 
+function add_pg_history(new_pg_history, new_pg, prev_pgs, prev_pg_history, old_pg)
+{
+    if (!new_pg_history[new_pg])
+    {
+        new_pg_history[new_pg] = {
+            osd_sets: {},
+            all_peers: {},
+            epoch: 0,
+        };
+    }
+    const nh = new_pg_history[new_pg], oh = prev_pg_history[old_pg];
+    nh.osd_sets[prev_pgs[old_pg].join(' ')] = prev_pgs[old_pg];
+    if (oh && oh.osd_sets && oh.osd_sets.length)
+    {
+        for (const pg of oh.osd_sets)
+        {
+            nh.osd_sets[pg.join(' ')] = pg;
+        }
+    }
+    if (oh && oh.all_peers && oh.all_peers.length)
+    {
+        for (const osd_num of oh.all_peers)
+        {
+            nh.all_peers[osd_num] = Number(osd_num);
+        }
+    }
+    if (oh && oh.epoch)
+    {
+        nh.epoch = nh.epoch < oh.epoch ? oh.epoch : nh.epoch;
+    }
+}
+
+function finish_pg_history(merged_history)
+{
+    merged_history.osd_sets = Object.values(merged_history.osd_sets);
+    merged_history.all_peers = Object.values(merged_history.all_peers);
+}
+
 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
+    // Add all possibly intersecting PGs to 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);
+        // New PG count is a multiple of 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]));
+            add_pg_history(new_pg_history, i, prev_pgs, prev_pg_history, i % old_pg_count);
+            finish_pg_history(new_pg_history[i]);
         }
     }
     else if (!(old_pg_count % new_pg_count))
@@ -25,68 +62,26 @@ function scale_pg_count(prev_pgs, prev_pg_history, new_pg_history, 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;
-                }
+                add_pg_history(new_pg_history, i, prev_pgs, prev_pg_history, i+j*new_pg_count);
             }
+            finish_pg_history(new_pg_history[i]);
         }
     }
     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)
+        let merged_history = {};
+        for (let i = 0; i < old_pg_count; i++)
         {
-            all_sets[pg.join(' ')] = pg;
+            add_pg_history(merged_history, 1, prev_pgs, prev_pg_history, i);
         }
-        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);
+        finish_pg_history(merged_history[1]);
         for (let i = 0; i < new_pg_count; i++)
         {
-            new_pg_history[i] = { osd_sets: all_sets, all_peers, epoch: max_epoch };
+            new_pg_history[i] = { ...merged_history[1] };
         }
     }
     // Mark history keys for removed PGs as removed
@@ -94,19 +89,16 @@ function scale_pg_count(prev_pgs, prev_pg_history, new_pg_history, new_pg_count)
     {
         new_pg_history[i] = null;
     }
+    // Just for the lp_solve optimizer - pick a "previous" PG for each "new" one
     if (old_pg_count < new_pg_count)
     {
-        for (let i = new_pg_count-1; i >= 0; i--)
+        for (let i = old_pg_count; i < new_pg_count; i++)
         {
-            prev_pgs[i] = prev_pgs[Math.floor(i/new_pg_count*old_pg_count)];
+            prev_pgs[i] = prev_pgs[i % 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

@@ -1,31 +1,16 @@
 // 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');
+// License: VNPL-1.1 (see https://yourcmc.ru/git/vitalif/vitastor/src/branch/master/README.md for details) or AGPL-3.0
+// Author: Vitaliy Filippov, 2020+
 
 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
@@ -56,93 +41,38 @@ function failure_rate_fullmesh(n, a, f)
 /******** 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
+// <k> replicas, <pgs> unique peer PGs per OSD (~50 for 100 PG-per-OSD in a big cluster)
 //
 // 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.
+// let them finish rebalance BEFORE replacing the failed drive (degraded_replacement=false).
 // At the same time, more peers per OSD increase probability of any of them to fail!
+// osd_rm=true means that failed OSDs' data is rebalanced over all other hosts,
+// not over the same host as it's in Ceph by default (dead OSDs are marked 'out').
 //
 // 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 })
+function cluster_afr({ n_hosts, n_drives, afr_drive, afr_host, capacity, speed, ec, ec_data, ec_parity, replicas, pgs = 1, osd_rm, degraded_replacement, down_out_interval = 600 })
 {
-    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');
+    const pg_size = (ec ? ec_data+ec_parity : replicas);
+    pgs = Math.min(pgs, (n_hosts-1)*n_drives/(pg_size-1));
+    const host_pgs = Math.min(pgs*n_drives, (n_hosts-1)*n_drives/(pg_size-1));
+    const resilver_disk = n_drives == 1 || osd_rm ? pgs : (n_drives-1);
+    const disk_heal_time = (down_out_interval + capacity/(degraded_replacement ? 1 : resilver_disk)/speed)/86400/365;
+    const host_heal_time = (down_out_interval + n_drives*capacity/pgs/speed)/86400/365;
+    const disk_heal_fail = ((afr_drive+afr_host/n_drives)*disk_heal_time);
+    const host_heal_fail = ((afr_drive+afr_host/n_drives)*host_heal_time);
+    const disk_pg_fail = ec
+        ? failure_rate_fullmesh(ec_data+ec_parity-1, disk_heal_fail, ec_parity)
+        : disk_heal_fail**(replicas-1);
+    const host_pg_fail = ec
+        ? failure_rate_fullmesh(ec_data+ec_parity-1, host_heal_fail, ec_parity)
+        : host_heal_fail**(replicas-1);
+    return 1 - ((1 - afr_drive * (1-(1-disk_pg_fail)**pgs)) ** (n_hosts*n_drives))
+        * ((1 - afr_host * (1-(1-host_pg_fail)**host_pgs)) ** n_hosts);
 }
 
 /******** UTILITY ********/

mon/afr_test.js

@@ -0,0 +1,28 @@
+const { sprintf } = require('sprintf-js');
+const { cluster_afr } = require('./afr.js');
+
+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, afr_host: 0, 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, afr_host: 0, capacity: 4000, speed: 0.1, ec: true, ec_data: 2, ec_parity: 1 });
+print_cluster_afr({ n_hosts: 4, n_drives: 3, afr_drive: 0.03, afr_host: 0.05, capacity: 4000, speed: 0.1, ec: true, ec_data: 2, ec_parity: 1 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0, 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, afr_host: 0, 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, afr_host: 0, 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 });
+
+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 ${config.ec_data}+${config.ec_parity}` : `, ${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');
+}

mon/lp-optimizer.js

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 // Data distribution optimizer using linear programming (lp_solve)
 
@@ -58,7 +58,7 @@ async function optimize_initial({ osd_tree, pg_count, pg_size = 3, pg_minsize =
     }
     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 all_pgs = Object.values(random_combinations(osd_tree, pg_size, max_combinations));
+    const all_pgs = Object.values(random_combinations(osd_tree, pg_size, max_combinations, parity_space > 1));
     const pg_per_osd = {};
     for (const pg of all_pgs)
     {
@@ -249,7 +249,7 @@ async function optimize_change({ prev_pgs: prev_int_pgs, osd_tree, pg_size = 3,
         }
     }
     // Get all combinations
-    let all_pgs = random_combinations(osd_tree, pg_size, max_combinations);
+    let all_pgs = random_combinations(osd_tree, pg_size, max_combinations, parity_space > 1);
     add_valid_previous(osd_tree, prev_weights, all_pgs);
     all_pgs = Object.values(all_pgs);
     const pg_per_osd = {};
@@ -275,6 +275,11 @@ async function optimize_change({ prev_pgs: prev_int_pgs, osd_tree, pg_size = 3,
     lp += 'max: '+all_pg_names.map(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';
+    lp += all_pg_names
+        .map(pg_name => (prev_weights[pg_name] ? `add_${pg_name} - del_${pg_name}` : `${pg_name}`))
+        .join(' + ')+' = '+(pg_count
+            - Object.keys(prev_weights).reduce((a, old_pg_name) => (a + (all_pgs_hash[old_pg_name] ? prev_weights[old_pg_name] : 0)), 0)
+        )+';\n';
     for (const osd in pg_per_osd)
     {
         if (osd !== NO_OSD)
@@ -488,7 +493,8 @@ function extract_osds(osd_tree, levels, osd_level, osds = {})
     return osds;
 }
 
-function random_combinations(osd_tree, pg_size, count)
+// ordered = don't treat (x,y) and (y,x) as equal
+function random_combinations(osd_tree, pg_size, count, ordered)
 {
     let seed = 0x5f020e43;
     let rng = () =>
@@ -516,25 +522,47 @@ function random_combinations(osd_tree, pg_size, count)
                 pg.push(osds[cur_hosts[next_host]][next_osd]);
                 cur_hosts.splice(next_host, 1);
             }
-            while (pg.length < pg_size)
+            const cyclic_pgs = [ pg ];
+            if (ordered)
             {
-                pg.push(NO_OSD);
+                for (let i = 1; i < pg.size; i++)
+                {
+                    cyclic_pgs.push([ ...pg.slice(i), ...pg.slice(0, i) ]);
+                }
+            }
+            for (const pg of cyclic_pgs)
+            {
+                while (pg.length < pg_size)
+                {
+                    pg.push(NO_OSD);
+                }
+                r['pg_'+pg.join('_')] = pg;
             }
-            r['pg_'+pg.join('_')] = pg;
         }
     }
     // Generate purely random combinations
-    restart: while (count > 0)
+    while (count > 0)
     {
         let host_idx = [];
-        for (let i = 0; i < pg_size && i < hosts.length; i++)
+        const cur_hosts = [ ...hosts.map((h, i) => i) ];
+        const max_hosts = pg_size < hosts.length ? pg_size : hosts.length;
+        if (ordered)
         {
-            let start = i > 0 ? host_idx[i-1]+1 : 0;
-            if (start >= hosts.length)
+            for (let i = 0; i < max_hosts; i++)
             {
-                continue restart;
+                const r = rng() % cur_hosts.length;
+                host_idx[i] = cur_hosts[r];
+                cur_hosts.splice(r, 1);
+            }
+        }
+        else
+        {
+            for (let i = 0; i < max_hosts; i++)
+            {
+                const r = rng() % (cur_hosts.length - (max_hosts - i - 1));
+                host_idx[i] = cur_hosts[r];
+                cur_hosts.splice(0, r+1);
             }
-            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)

mon/make-osd.sh

@@ -0,0 +1,76 @@
+#!/bin/bash
+# Very simple systemd unit generator for vitastor-osd services
+# Not the final solution yet, mostly for tests
+# Copyright (c) Vitaliy Filippov, 2019+
+# License: MIT
+
+# USAGE: ./make-osd.sh /dev/disk/by-partuuid/xxx [ /dev/disk/by-partuuid/yyy]...
+
+IP_SUBSTR="10.200.1."
+ETCD_HOSTS="etcd0=http://10.200.1.10:2380,etcd1=http://10.200.1.11:2380,etcd2=http://10.200.1.12:2380"
+
+set -e -x
+
+IP=`ip -json a s | jq -r '.[].addr_info[] | select(.local | startswith("'$IP_SUBSTR'")) | .local'`
+[ "$IP" != "" ] || exit 1
+ETCD_MON=$(echo $ETCD_HOSTS | perl -pe 's/:2380/:2379/g; s/etcd\d*=//g;')
+D=`dirname $0`
+
+# Create OSDs on all passed devices
+OSD_NUM=1
+for DEV in $*; do
+
+# Ugly :) -> node.js rework pending
+while true; do
+    ST=$(etcdctl --endpoints="$ETCD_MON" get --print-value-only /vitastor/osd/stats/$OSD_NUM)
+    if [ "$ST" = "" ]; then
+        break
+    fi
+    OSD_NUM=$((OSD_NUM+1))
+done
+etcdctl --endpoints="$ETCD_MON" put /vitastor/osd/stats/$OSD_NUM '{}'
+
+echo Creating OSD $OSD_NUM on $DEV
+
+OPT=`node $D/simple-offsets.js --device $DEV --format options | tr '\n' ' '`
+META=`echo $OPT | grep -Po '(?<=data_offset )\d+'`
+dd if=/dev/zero of=$DEV bs=1048576 count=$(((META+1048575)/1048576)) oflag=direct
+
+cat >/etc/systemd/system/vitastor-osd$OSD_NUM.service <<EOF
+[Unit]
+Description=Vitastor object storage daemon osd.$OSD_NUM
+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=/usr/bin/vitastor-osd \\
+    --etcd_address $IP:2379/v3 \\
+    --bind_address $IP \\
+    --osd_num $OSD_NUM \\
+    --disable_data_fsync 1 \\
+    --immediate_commit all \\
+    --flusher_count 256 \\
+    --disk_alignment 4096 --journal_block_size 4096 --meta_block_size 4096 \\
+    --journal_no_same_sector_overwrites true \\
+    --journal_sector_buffer_count 1024 \\
+    $OPT
+WorkingDirectory=/
+ExecStartPre=+chown vitastor:vitastor $DEV
+User=vitastor
+PrivateTmp=false
+TasksMax=infinity
+Restart=always
+StartLimitInterval=0
+RestartSec=10
+
+[Install]
+WantedBy=vitastor.target
+EOF
+
+systemctl enable vitastor-osd$OSD_NUM
+
+done

mon/make-units.sh

@@ -1,19 +1,59 @@
 #!/bin/bash
-# Example startup script generator
-# Of course this isn't a production solution yet, this is just for tests
+# Very simple systemd unit generator for etcd & vitastor-mon services
+# Not the final solution yet, mostly 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'`
+# USAGE: ./make-units.sh
 
+IP_SUBSTR="10.200.1."
+ETCD_HOSTS="etcd0=http://10.200.1.10:2380,etcd1=http://10.200.1.11:2380,etcd2=http://10.200.1.12:2380"
+
+# determine IP
+IP=`ip -json a s | jq -r '.[].addr_info[] | select(.local | startswith("'$IP_SUBSTR'")) | .local'`
 [ "$IP" != "" ] || exit 1
+ETCD_NUM=${ETCD_HOSTS/$IP*/}
+[ "$ETCD_NUM" != "$ETCD_HOSTS" ] || exit 1
+ETCD_NUM=$(echo $ETCD_NUM | tr -d -c , | wc -c)
+
+# etcd
+useradd etcd
+
+mkdir -p /var/lib/etcd$ETCD_NUM.etcd
+cat >/etc/systemd/system/etcd.service <<EOF
+[Unit]
+Description=etcd for vitastor
+After=network-online.target local-fs.target time-sync.target
+Wants=network-online.target local-fs.target time-sync.target
+
+[Service]
+Restart=always
+ExecStart=/usr/local/bin/etcd -name etcd$ETCD_NUM --data-dir /var/lib/etcd$ETCD_NUM.etcd \\
+    --advertise-client-urls http://$IP:2379 --listen-client-urls http://$IP:2379 \\
+    --initial-advertise-peer-urls http://$IP:2380 --listen-peer-urls http://$IP:2380 \\
+    --initial-cluster-token vitastor-etcd-1 --initial-cluster $ETCD_HOSTS \\
+    --initial-cluster-state new --max-txn-ops=100000 --auto-compaction-retention=10 --auto-compaction-mode=revision
+WorkingDirectory=/var/lib/etcd$ETCD_NUM.etcd
+ExecStartPre=+chown -R etcd /var/lib/etcd$ETCD_NUM.etcd
+User=etcd
+PrivateTmp=false
+TasksMax=infinity
+Restart=always
+StartLimitInterval=0
+RestartSec=10
+
+[Install]
+WantedBy=local.target
+EOF
+
+systemctl daemon-reload
+systemctl enable etcd
+systemctl start etcd
 
 useradd vitastor
 chmod 755 /root
 
-BASE=${IP/*./}
-BASE=$(((BASE-10)*12))
-
+# Vitastor target
 cat >/etc/systemd/system/vitastor.target <<EOF
 [Unit]
 Description=vitastor target
@@ -21,116 +61,25 @@ Description=vitastor target
 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
+# Monitor unit
+ETCD_MON=$(echo $ETCD_HOSTS | perl -pe 's/:2380/:2379/g; s/etcd\d*=//g;')
+cat >/etc/systemd/system/vitastor-mon.service <<EOF
 [Unit]
-Description=Vitastor object storage daemon osd.$((BASE+i))
+Description=Vitastor monitor
 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/$DEV
+Restart=always
+ExecStart=node /usr/lib/vitastor/mon/mon-main.js --etcd_url '$ETCD_MON' --etcd_prefix '/vitastor' --etcd_start_timeout 5
+WorkingDirectory=/
 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

@@ -1,7 +1,7 @@
 #!/usr/bin/node
 
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 const Mon = require('./mon.js');
 

mon/simple-offsets.js

@@ -4,6 +4,7 @@
 // Simple tool to calculate journal and metadata offsets for a single device
 // Will be replaced by smarter tools in the future
 
+const fs = require('fs').promises;
 const child_process = require('child_process');
 
 async function run()
@@ -15,6 +16,7 @@ async function run()
         device_block_size: 4096,
         journal_offset: 0,
         device_size: 0,
+        format: 'text',
     };
     for (let i = 2; i < process.argv.length; i++)
     {
@@ -24,7 +26,22 @@ async function run()
             i++;
         }
     }
-    const device_size = Number(options.device_size || await system("blockdev --getsize64 "+options.device));
+    if (!options.device)
+    {
+        process.stderr.write('USAGE: nodejs '+process.argv[1]+' --device /dev/sdXXX\n');
+        process.exit(1);
+    }
+    options.device_size = Number(options.device_size);
+    let device_size = options.device_size;
+    if (!device_size)
+    {
+        const st = await fs.stat(options.device);
+        options.device_block_size = st.blksize;
+        if (st.isBlockDevice())
+            device_size = Number(await system("/sbin/blockdev --getsize64 "+options.device))
+        else
+            device_size = st.size;
+    }
     if (!device_size)
     {
         process.stderr.write('Failed to get device size\n');
@@ -32,25 +49,45 @@ async function run()
     }
     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 entries_per_block = Math.floor(options.device_block_size / (24 + 2*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` : '')
-    );
+    if (options.format == 'text' || options.format == 'options')
+    {
+        if (options.format == 'text')
+        {
+            process.stderr.write(
+                `Metadata size: ${meta_size_fmt}\n`+
+                `Options for the OSD:\n`
+            );
+        }
+        process.stdout.write(
+            `    --data_device ${options.device}\n`+
+            `    --journal_offset ${options.journal_offset}\n`+
+            `    --meta_offset ${meta_offset}\n`+
+            `    --data_offset ${data_offset}\n`+
+            (options.device_size ? `    --data_size ${device_size-data_offset}\n` : '')
+        );
+    }
+    else if (options.format == 'env')
+    {
+        process.stdout.write(
+            `journal_offset=${options.journal_offset}\n`+
+            `meta_offset=${meta_offset}\n`+
+            `data_offset=${data_offset}\n`+
+            `data_size=${device_size-data_offset}\n`
+        );
+    }
+    else
+        process.stdout.write('Unknown format: '+options.format);
 }
 
 function system(cmd)
 {
-    return new Promise((ok, no) => child_process.exec(cmd, { maxBuffer: 64*1024*1024 }, (err, stdout, stderr) => (err ? no(err) : ok(stdout))));
+    return new Promise((ok, no) => child_process.exec(cmd, { maxBuffer: 64*1024*1024 }, (err, stdout, stderr) => (err ? no(err.message) : ok(stdout))));
 }
 
-run().catch(console.error);
+run().catch(err => { console.error(err); process.exit(1); });

mon/test-nonuniform.js

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (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,

mon/test-optimize-simple.js

@@ -0,0 +1,25 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+const LPOptimizer = require('./lp-optimizer.js');
+
+async function run()
+{
+    const osd_tree = { a: { 1: 1 }, b: { 2: 1 }, c: { 3: 1 } };
+    let res;
+
+    console.log('16 PGs, size=3');
+    res = await LPOptimizer.optimize_initial({ osd_tree, pg_size: 3, pg_count: 16 });
+    LPOptimizer.print_change_stats(res, false);
+
+    console.log('\nReduce PG size to 2');
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs.map(pg => pg.slice(0, 2)), osd_tree, pg_size: 2 });
+    LPOptimizer.print_change_stats(res, false);
+
+    console.log('\nRemove OSD 3');
+    delete osd_tree['c'];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2 });
+    LPOptimizer.print_change_stats(res, false);
+}
+
+run().catch(console.error);

mon/test-optimize-undersized.js

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 const LPOptimizer = require('./lp-optimizer.js');
 

mon/test-optimize.js

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 const LPOptimizer = require('./lp-optimizer.js');
 

msgr_send.cpp

@@ -1,186 +0,0 @@
-// 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);
-        }
-    }
-}

osd_rmw_test.cpp

@@ -1,363 +0,0 @@
-// 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"
-
-void dump_stripes(osd_rmw_stripe_t *stripes, int pg_size);
-void test1();
-void test4();
-void test5();
-void test6();
-void test7();
-void test8();
-void test9();
-
-/***
-
-Cases:
-
-1. split(offset=128K-4K, len=8K)
-   = [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 0 ] ]
-
-2. read(offset=128K-4K, len=8K, osd_set=[1,0,3])
-   = { read: [ [ 0, 128K ], [ 0, 4K ], [ 0, 4K ] ] }
-
-3. cover_read(0, 128K, { req: [ 128K-4K, 4K ] })
-   = { read: [ 0, 128K-4K ] }
-
-4. write(offset=128K-4K, len=8K, osd_set=[1,0,3])
-   = {
-     read: [ [ 0, 128K ], [ 4K, 128K ], [ 4K, 128K ] ],
-     write: [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2, read0, read1, read2 ],
-   }
-   + check write2 buffer
-
-5. write(offset=0, len=128K+64K, osd_set=[1,0,3])
-   = {
-     req: [ [ 0, 128K ], [ 0, 64K ], [ 0, 0 ] ],
-     read: [ [ 64K, 128K ], [ 64K, 128K ], [ 64K, 128K ] ],
-     write: [ [ 0, 128K ], [ 0, 64K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2, read0, read1, read2 ],
-   }
-
-6. write(offset=0, len=128K+64K, osd_set=[1,2,3])
-   = {
-     req: [ [ 0, 128K ], [ 0, 64K ], [ 0, 0 ] ],
-     read: [ [ 0, 0 ], [ 64K, 128K ], [ 0, 0 ] ],
-     write: [ [ 0, 128K ], [ 0, 64K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2, read1 ],
-   }
-
-7. calc_rmw(offset=128K-4K, len=8K, osd_set=[1,0,3], write_set=[1,2,3])
-   = {
-     read: [ [ 0, 128K ], [ 0, 128K ], [ 0, 128K ] ],
-     write: [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2, read0, read1, read2 ],
-   }
-   then, after calc_rmw_parity_xor(): {
-     write: [ [ 128K-4K, 128K ], [ 0, 128K ], [ 0, 128K ] ],
-     write1==read1,
-   }
-   + check write1 buffer
-   + check write2 buffer
-
-8. calc_rmw(offset=0, len=128K+4K, osd_set=[0,2,3], write_set=[1,2,3])
-   = {
-     read: [ [ 0, 0 ], [ 4K, 128K ], [ 0, 0 ] ],
-     write: [ [ 0, 128K ], [ 0, 4K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2, read1 ],
-   }
-   + check write2 buffer
-
-9. object recovery case:
-   calc_rmw(offset=0, len=0, read_osd_set=[0,2,3], write_osd_set=[1,2,3])
-   = {
-     read: [ [ 0, 128K ], [ 0, 128K ], [ 0, 128K ] ],
-     write: [ [ 0, 0 ], [ 0, 0 ], [ 0, 0 ] ],
-     input buffer: NULL,
-     rmw buffer: [ read0, read1, read2 ],
-   }
-   then, after calc_rmw_parity_xor(): {
-     write: [ [ 0, 128K ], [ 0, 0 ], [ 0, 0 ] ],
-     write0==read0,
-   }
-   + check write0 buffer
-
-***/
-
-int main(int narg, char *args[])
-{
-    // Test 1
-    test1();
-    // Test 4
-    test4();
-    // Test 5
-    test5();
-    // Test 6
-    test6();
-    // Test 7
-    test7();
-    // Test 8
-    test8();
-    // Test 9
-    test9();
-    // End
-    printf("all ok\n");
-    return 0;
-}
-
-void dump_stripes(osd_rmw_stripe_t *stripes, int pg_size)
-{
-    printf("request");
-    for (int i = 0; i < pg_size; i++)
-    {
-        printf(" {%uK-%uK}", stripes[i].req_start/1024, stripes[i].req_end/1024);
-    }
-    printf("\n");
-    printf("read");
-    for (int i = 0; i < pg_size; i++)
-    {
-        printf(" {%uK-%uK}", stripes[i].read_start/1024, stripes[i].read_end/1024);
-    }
-    printf("\n");
-    printf("write");
-    for (int i = 0; i < pg_size; i++)
-    {
-        printf(" {%uK-%uK}", stripes[i].write_start/1024, stripes[i].write_end/1024);
-    }
-    printf("\n");
-}
-
-void test1()
-{
-    osd_num_t osd_set[3] = { 1, 0, 3 };
-    osd_rmw_stripe_t stripes[3] = { 0 };
-    // Test 1.1
-    split_stripes(2, 128*1024, 128*1024-4096, 8192, stripes);
-    assert(stripes[0].req_start == 128*1024-4096 && stripes[0].req_end == 128*1024);
-    assert(stripes[1].req_start == 0 && stripes[1].req_end == 4096);
-    assert(stripes[2].req_end == 0);
-    // Test 1.2
-    for (int i = 0; i < 3; i++)
-    {
-        stripes[i].read_start = stripes[i].req_start;
-        stripes[i].read_end = stripes[i].req_end;
-    }
-    assert(extend_missing_stripes(stripes, osd_set, 2, 3) == 0);
-    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
-    assert(stripes[2].read_start == 0 && stripes[2].read_end == 4096);
-    // Test 1.3
-    stripes[0] = { .req_start = 128*1024-4096, .req_end = 128*1024 };
-    cover_read(0, 128*1024, stripes[0]);
-    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024-4096);
-}
-
-void test4()
-{
-    osd_num_t osd_set[3] = { 1, 0, 3 };
-    osd_rmw_stripe_t stripes[3] = { 0 };
-    // Test 4.1
-    split_stripes(2, 128*1024, 128*1024-4096, 8192, stripes);
-    void* write_buf = malloc(8192);
-    void* rmw_buf = calc_rmw(write_buf, stripes, osd_set, 3, 2, 2, osd_set, 128*1024);
-    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
-    assert(stripes[1].read_start == 4096 && stripes[1].read_end == 128*1024);
-    assert(stripes[2].read_start == 4096 && stripes[2].read_end == 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 == 4096);
-    assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024);
-    assert(stripes[0].read_buf == rmw_buf+128*1024);
-    assert(stripes[1].read_buf == rmw_buf+128*1024*2);
-    assert(stripes[2].read_buf == rmw_buf+128*1024*3-4096);
-    assert(stripes[0].write_buf == write_buf);
-    assert(stripes[1].write_buf == write_buf+4096);
-    assert(stripes[2].write_buf == rmw_buf);
-    // Test 4.2
-    set_pattern(write_buf, 8192, PATTERN0);
-    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_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
-    free(rmw_buf);
-    free(write_buf);
-}
-
-void test5()
-{
-    osd_num_t osd_set[3] = { 1, 0, 3 };
-    osd_rmw_stripe_t stripes[3] = { 0 };
-    // Test 5.1
-    split_stripes(2, 128*1024, 0, 64*1024*3, stripes);
-    assert(stripes[0].req_start == 0 && stripes[0].req_end == 128*1024);
-    assert(stripes[1].req_start == 0 && stripes[1].req_end == 64*1024);
-    assert(stripes[2].req_end == 0);
-    // Test 5.2
-    void *write_buf = malloc(64*1024*3);
-    void *rmw_buf = calc_rmw(write_buf, stripes, osd_set, 3, 2, 2, osd_set, 128*1024);
-    assert(stripes[0].read_start == 64*1024 && stripes[0].read_end == 128*1024);
-    assert(stripes[1].read_start == 64*1024 && stripes[1].read_end == 128*1024);
-    assert(stripes[2].read_start == 64*1024 && stripes[2].read_end == 128*1024);
-    assert(stripes[0].write_start == 0 && stripes[0].write_end == 128*1024);
-    assert(stripes[1].write_start == 0 && stripes[1].write_end == 64*1024);
-    assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024);
-    assert(stripes[0].read_buf == rmw_buf+128*1024);
-    assert(stripes[1].read_buf == rmw_buf+64*3*1024);
-    assert(stripes[2].read_buf == rmw_buf+64*4*1024);
-    assert(stripes[0].write_buf == write_buf);
-    assert(stripes[1].write_buf == write_buf+128*1024);
-    assert(stripes[2].write_buf == rmw_buf);
-    free(rmw_buf);
-    free(write_buf);
-}
-
-void test6()
-{
-    osd_num_t osd_set[3] = { 1, 2, 3 };
-    osd_rmw_stripe_t stripes[3] = { 0 };
-    // Test 6.1
-    split_stripes(2, 128*1024, 0, 64*1024*3, stripes);
-    void *write_buf = malloc(64*1024*3);
-    void *rmw_buf = calc_rmw(write_buf, stripes, osd_set, 3, 2, 3, osd_set, 128*1024);
-    assert(stripes[0].read_end == 0);
-    assert(stripes[1].read_start == 64*1024 && stripes[1].read_end == 128*1024);
-    assert(stripes[2].read_end == 0);
-    assert(stripes[0].write_start == 0 && stripes[0].write_end == 128*1024);
-    assert(stripes[1].write_start == 0 && stripes[1].write_end == 64*1024);
-    assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024);
-    assert(stripes[0].read_buf == 0);
-    assert(stripes[1].read_buf == rmw_buf+128*1024);
-    assert(stripes[2].read_buf == 0);
-    assert(stripes[0].write_buf == write_buf);
-    assert(stripes[1].write_buf == write_buf+128*1024);
-    assert(stripes[2].write_buf == rmw_buf);
-    free(rmw_buf);
-    free(write_buf);
-}
-
-void test7()
-{
-    osd_num_t osd_set[3] = { 1, 0, 3 };
-    osd_num_t write_osd_set[3] = { 1, 2, 3 };
-    osd_rmw_stripe_t stripes[3] = { 0 };
-    // Test 7.1
-    split_stripes(2, 128*1024, 128*1024-4096, 8192, stripes);
-    void *write_buf = malloc(8192);
-    void *rmw_buf = calc_rmw(write_buf, stripes, osd_set, 3, 2, 2, write_osd_set, 128*1024);
-    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
-    assert(stripes[1].read_start == 0 && stripes[1].read_end == 128*1024);
-    assert(stripes[2].read_start == 0 && stripes[2].read_end == 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 == 4096);
-    assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024);
-    assert(stripes[0].read_buf == rmw_buf+128*1024);
-    assert(stripes[1].read_buf == rmw_buf+128*1024*2);
-    assert(stripes[2].read_buf == rmw_buf+128*1024*3);
-    assert(stripes[0].write_buf == write_buf);
-    assert(stripes[1].write_buf == write_buf+4096);
-    assert(stripes[2].write_buf == rmw_buf);
-    // Test 7.2
-    set_pattern(write_buf, 8192, PATTERN0);
-    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_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);
-    assert(stripes[1].write_buf == stripes[1].read_buf);
-    check_pattern(stripes[1].write_buf, 4096, PATTERN0);
-    check_pattern(stripes[1].write_buf+4096, 128*1024-4096, PATTERN1);
-    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
-    free(rmw_buf);
-    free(write_buf);
-}
-
-void test8()
-{
-    osd_num_t osd_set[3] = { 0, 2, 3 };
-    osd_num_t write_osd_set[3] = { 1, 2, 3 };
-    osd_rmw_stripe_t stripes[3] = { 0 };
-    // Test 8.1
-    split_stripes(2, 128*1024, 0, 128*1024+4096, stripes);
-    void *write_buf = malloc(128*1024+4096);
-    void *rmw_buf = calc_rmw(write_buf, stripes, osd_set, 3, 2, 2, write_osd_set, 128*1024);
-    assert(stripes[0].read_start == 0 && stripes[0].read_end == 0);
-    assert(stripes[1].read_start == 4096 && stripes[1].read_end == 128*1024);
-    assert(stripes[2].read_start == 0 && stripes[2].read_end == 0);
-    assert(stripes[0].write_start == 0 && stripes[0].write_end == 128*1024);
-    assert(stripes[1].write_start == 0 && stripes[1].write_end == 4096);
-    assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024);
-    assert(stripes[0].read_buf == NULL);
-    assert(stripes[1].read_buf == rmw_buf+128*1024);
-    assert(stripes[2].read_buf == NULL);
-    assert(stripes[0].write_buf == write_buf);
-    assert(stripes[1].write_buf == write_buf+128*1024);
-    assert(stripes[2].write_buf == rmw_buf);
-    // Test 8.2
-    set_pattern(write_buf, 128*1024+4096, PATTERN0);
-    set_pattern(stripes[1].read_buf, 128*1024-4096, PATTERN1);
-    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
-    assert(stripes[0].write_buf == write_buf);                               // recheck again
-    assert(stripes[1].write_buf == write_buf+128*1024);                      // recheck again
-    assert(stripes[2].write_buf == rmw_buf);                                 // recheck again
-    check_pattern(stripes[2].write_buf, 4096, 0); // new parity
-    check_pattern(stripes[2].write_buf+4096, 128*1024-4096, PATTERN0^PATTERN1); // new parity
-    free(rmw_buf);
-    free(write_buf);
-}
-
-void test9()
-{
-    osd_num_t osd_set[3] = { 0, 2, 3 };
-    osd_num_t write_osd_set[3] = { 1, 2, 3 };
-    osd_rmw_stripe_t stripes[3] = { 0 };
-    // Test 9.0
-    split_stripes(2, 128*1024, 64*1024, 0, stripes);
-    assert(stripes[0].req_start == 0 && stripes[0].req_end == 0);
-    assert(stripes[1].req_start == 0 && stripes[1].req_end == 0);
-    assert(stripes[2].req_start == 0 && stripes[2].req_end == 0);
-    // Test 9.1
-    void *write_buf = NULL;
-    void *rmw_buf = calc_rmw(write_buf, stripes, osd_set, 3, 2, 3, write_osd_set, 128*1024);
-    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
-    assert(stripes[1].read_start == 0 && stripes[1].read_end == 128*1024);
-    assert(stripes[2].read_start == 0 && stripes[2].read_end == 128*1024);
-    assert(stripes[0].write_start == 0 && stripes[0].write_end == 0);
-    assert(stripes[1].write_start == 0 && stripes[1].write_end == 0);
-    assert(stripes[2].write_start == 0 && stripes[2].write_end == 0);
-    assert(stripes[0].read_buf == rmw_buf);
-    assert(stripes[1].read_buf == rmw_buf+128*1024);
-    assert(stripes[2].read_buf == rmw_buf+128*1024*2);
-    assert(stripes[0].write_buf == NULL);
-    assert(stripes[1].write_buf == NULL);
-    assert(stripes[2].write_buf == NULL);
-    // Test 9.2
-    set_pattern(stripes[1].read_buf, 128*1024, 0);
-    set_pattern(stripes[2].read_buf, 128*1024, PATTERN1);
-    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);
-    assert(stripes[0].write_buf == rmw_buf);
-    assert(stripes[1].write_buf == NULL);
-    assert(stripes[2].write_buf == NULL);
-    check_pattern(stripes[0].read_buf, 128*1024, PATTERN1);
-    check_pattern(stripes[0].write_buf, 128*1024, PATTERN1);
-    free(rmw_buf);
-}

qemu-3.1-vitastor.patch

@@ -0,0 +1,84 @@
+Index: qemu-3.1+dfsg/qapi/block-core.json
+===================================================================
+--- qemu-3.1+dfsg.orig/qapi/block-core.json
++++ qemu-3.1+dfsg/qapi/block-core.json
+@@ -2617,7 +2617,7 @@
+ ##
+ { 'enum': 'BlockdevDriver',
+   'data': [ 'blkdebug', 'blklogwrites', 'blkverify', 'bochs', 'cloop',
+-            'copy-on-read', 'dmg', 'file', 'ftp', 'ftps', 'gluster',
++            'copy-on-read', 'dmg', 'file', 'ftp', 'ftps', 'gluster', 'vitastor',
+             'host_cdrom', 'host_device', 'http', 'https', 'iscsi', 'luks',
+             'nbd', 'nfs', 'null-aio', 'null-co', 'nvme', 'parallels', 'qcow',
+             'qcow2', 'qed', 'quorum', 'raw', 'rbd', 'replication', 'sheepdog',
+@@ -3367,6 +3367,24 @@
+             '*tag': 'str' } }
+ 
+ ##
++# @BlockdevOptionsVitastor:
++#
++# Driver specific block device options for vitastor
++#
++# @inode:       Inode number
++# @pool:        Pool ID
++# @size:        Desired image size in bytes
++# @etcd_host:   etcd connection address
++# @etcd_prefix: etcd key/value prefix
++##
++{ 'struct': 'BlockdevOptionsVitastor',
++  'data': { 'inode': 'uint64',
++            'pool': 'uint64',
++            'size': 'uint64',
++            'etcd_host': 'str',
++            '*etcd_prefix': 'str' } }
++
++##
+ # @ReplicationMode:
+ #
+ # An enumeration of replication modes.
+@@ -3713,6 +3731,7 @@
+       'rbd':        'BlockdevOptionsRbd',
+       'replication':'BlockdevOptionsReplication',
+       'sheepdog':   'BlockdevOptionsSheepdog',
++      'vitastor':   'BlockdevOptionsVitastor',
+       'ssh':        'BlockdevOptionsSsh',
+       'throttle':   'BlockdevOptionsThrottle',
+       'vdi':        'BlockdevOptionsGenericFormat',
+@@ -4158,6 +4177,17 @@
+             '*block-state-zero':    'bool' } }
+ 
+ ##
++# @BlockdevCreateOptionsVitastor:
++#
++# Driver specific image creation options for Vitastor.
++#
++# @size: Size of the virtual disk in bytes
++##
++{ 'struct': 'BlockdevCreateOptionsVitastor',
++  'data': { 'location':         'BlockdevOptionsVitastor',
++            'size':             'size' } }
++
++##
+ # @BlockdevVpcSubformat:
+ #
+ # @dynamic: Growing image file
+@@ -4212,6 +4242,7 @@
+       'qed':            'BlockdevCreateOptionsQed',
+       'rbd':            'BlockdevCreateOptionsRbd',
+       'sheepdog':       'BlockdevCreateOptionsSheepdog',
++      'vitastor':       'BlockdevCreateOptionsVitastor',
+       'ssh':            'BlockdevCreateOptionsSsh',
+       'vdi':            'BlockdevCreateOptionsVdi',
+       'vhdx':           'BlockdevCreateOptionsVhdx',
+Index: qemu-3.1+dfsg/scripts/modules/module_block.py
+===================================================================
+--- qemu-3.1+dfsg.orig/scripts/modules/module_block.py
++++ qemu-3.1+dfsg/scripts/modules/module_block.py
+@@ -88,6 +88,7 @@ def print_bottom(fheader):
+ output_file = sys.argv[1]
+ with open(output_file, 'w') as fheader:
+     print_top(fheader)
++    add_module(fheader, "vitastor", "vitastor", "vitastor")
+ 
+     for filename in sys.argv[2:]:
+         if os.path.isfile(filename):

qemu-4.2-vitastor.patch

@@ -0,0 +1,84 @@
+Index: qemu/qapi/block-core.json
+===================================================================
+--- qemu.orig/qapi/block-core.json	2020-11-07 22:57:38.932613674 +0000
++++ qemu.orig/qapi/block-core.json	2020-11-07 22:59:49.890722862 +0000
+@@ -2907,7 +2907,7 @@
+             'nbd', 'nfs', 'null-aio', 'null-co', 'nvme', 'parallels', 'qcow',
+             'qcow2', 'qed', 'quorum', 'raw', 'rbd',
+             { 'name': 'replication', 'if': 'defined(CONFIG_REPLICATION)' },
+-            'sheepdog',
++            'sheepdog', 'vitastor',
+             'ssh', 'throttle', 'vdi', 'vhdx', 'vmdk', 'vpc', 'vvfat', 'vxhs' ] }
+ 
+ ##
+@@ -3725,6 +3725,24 @@
+             '*tag': 'str' } }
+ 
+ ##
++# @BlockdevOptionsVitastor:
++#
++# Driver specific block device options for vitastor
++#
++# @inode:       Inode number
++# @pool:        Pool ID
++# @size:        Desired image size in bytes
++# @etcd_host:   etcd connection address
++# @etcd_prefix: etcd key/value prefix
++##
++{ 'struct': 'BlockdevOptionsVitastor',
++  'data': { 'inode': 'uint64',
++            'pool': 'uint64',
++            'size': 'uint64',
++            'etcd_host': 'str',
++            '*etcd_prefix': 'str' } }
++
++##
+ # @ReplicationMode:
+ #
+ # An enumeration of replication modes.
+@@ -4084,6 +4102,7 @@
+       'replication': { 'type': 'BlockdevOptionsReplication',
+                        'if': 'defined(CONFIG_REPLICATION)' },
+       'sheepdog':   'BlockdevOptionsSheepdog',
++      'vitastor':   'BlockdevOptionsVitastor',
+       'ssh':        'BlockdevOptionsSsh',
+       'throttle':   'BlockdevOptionsThrottle',
+       'vdi':        'BlockdevOptionsGenericFormat',
+@@ -4461,6 +4480,17 @@
+             '*cluster-size' :   'size' } }
+ 
+ ##
++# @BlockdevCreateOptionsVitastor:
++#
++# Driver specific image creation options for Vitastor.
++#
++# @size: Size of the virtual disk in bytes
++##
++{ 'struct': 'BlockdevCreateOptionsVitastor',
++  'data': { 'location':         'BlockdevOptionsVitastor',
++            'size':             'size' } }
++
++##
+ # @BlockdevVmdkSubformat:
+ #
+ # Subformat options for VMDK images
+@@ -4722,6 +4752,7 @@
+       'qed':            'BlockdevCreateOptionsQed',
+       'rbd':            'BlockdevCreateOptionsRbd',
+       'sheepdog':       'BlockdevCreateOptionsSheepdog',
++      'vitastor':       'BlockdevCreateOptionsVitastor',
+       'ssh':            'BlockdevCreateOptionsSsh',
+       'vdi':            'BlockdevCreateOptionsVdi',
+       'vhdx':           'BlockdevCreateOptionsVhdx',
+Index: qemu/scripts/modules/module_block.py
+===================================================================
+--- qemu.orig/scripts/modules/module_block.py	2020-11-07 22:57:38.936613739 +0000
++++ qemu/scripts/modules/module_block.py	2020-11-07 22:59:49.890722862 +0000
+@@ -86,6 +86,7 @@ def print_bottom(fheader):
+ output_file = sys.argv[1]
+ with open(output_file, 'w') as fheader:
+     print_top(fheader)
++    add_module(fheader, "vitastor", "vitastor", "vitastor")
+ 
+     for filename in sys.argv[2:]:
+         if os.path.isfile(filename):

qemu-5.0-vitastor.patch

@@ -0,0 +1,84 @@
+Index: qemu/qapi/block-core.json
+===================================================================
+--- qemu.orig/qapi/block-core.json
++++ qemu/qapi/block-core.json
+@@ -2798,7 +2798,7 @@
+             'luks', 'nbd', 'nfs', 'null-aio', 'null-co', 'nvme', 'parallels',
+             'qcow', 'qcow2', 'qed', 'quorum', 'raw', 'rbd',
+             { 'name': 'replication', 'if': 'defined(CONFIG_REPLICATION)' },
+-            'sheepdog',
++            'sheepdog', 'vitastor',
+             'ssh', 'throttle', 'vdi', 'vhdx', 'vmdk', 'vpc', 'vvfat', 'vxhs' ] }
+ 
+ ##
+@@ -3635,6 +3635,24 @@
+             '*tag': 'str' } }
+ 
+ ##
++# @BlockdevOptionsVitastor:
++#
++# Driver specific block device options for vitastor
++#
++# @inode:       Inode number
++# @pool:        Pool ID
++# @size:        Desired image size in bytes
++# @etcd_host:   etcd connection address
++# @etcd_prefix: etcd key/value prefix
++##
++{ 'struct': 'BlockdevOptionsVitastor',
++  'data': { 'inode': 'uint64',
++            'pool': 'uint64',
++            'size': 'uint64',
++            'etcd_host': 'str',
++            '*etcd_prefix': 'str' } }
++
++##
+ # @ReplicationMode:
+ #
+ # An enumeration of replication modes.
+@@ -3995,6 +4013,7 @@
+       'replication': { 'type': 'BlockdevOptionsReplication',
+                        'if': 'defined(CONFIG_REPLICATION)' },
+       'sheepdog':   'BlockdevOptionsSheepdog',
++      'vitastor':   'BlockdevOptionsVitastor',
+       'ssh':        'BlockdevOptionsSsh',
+       'throttle':   'BlockdevOptionsThrottle',
+       'vdi':        'BlockdevOptionsGenericFormat',
+@@ -4365,6 +4384,17 @@
+             '*cluster-size' :   'size' } }
+ 
+ ##
++# @BlockdevCreateOptionsVitastor:
++#
++# Driver specific image creation options for Vitastor.
++#
++# @size: Size of the virtual disk in bytes
++##
++{ 'struct': 'BlockdevCreateOptionsVitastor',
++  'data': { 'location':         'BlockdevOptionsVitastor',
++            'size':             'size' } }
++
++##
+ # @BlockdevVmdkSubformat:
+ #
+ # Subformat options for VMDK images
+@@ -4626,6 +4656,7 @@
+       'qed':            'BlockdevCreateOptionsQed',
+       'rbd':            'BlockdevCreateOptionsRbd',
+       'sheepdog':       'BlockdevCreateOptionsSheepdog',
++      'vitastor':       'BlockdevCreateOptionsVitastor',
+       'ssh':            'BlockdevCreateOptionsSsh',
+       'vdi':            'BlockdevCreateOptionsVdi',
+       'vhdx':           'BlockdevCreateOptionsVhdx',
+Index: qemu/scripts/modules/module_block.py
+===================================================================
+--- qemu.orig/scripts/modules/module_block.py
++++ qemu/scripts/modules/module_block.py
+@@ -85,6 +85,7 @@ def print_bottom(fheader):
+ output_file = sys.argv[1]
+ with open(output_file, 'w') as fheader:
+     print_top(fheader)
++    add_module(fheader, "vitastor", "vitastor", "vitastor")
+ 
+     for filename in sys.argv[2:]:
+         if os.path.isfile(filename):

qemu-5.1-vitastor.patch

@@ -0,0 +1,84 @@
+Index: qemu-5.1+dfsg/qapi/block-core.json
+===================================================================
+--- qemu-5.1+dfsg.orig/qapi/block-core.json
++++ qemu-5.1+dfsg/qapi/block-core.json
+@@ -2807,7 +2807,7 @@
+             'luks', 'nbd', 'nfs', 'null-aio', 'null-co', 'nvme', 'parallels',
+             'qcow', 'qcow2', 'qed', 'quorum', 'raw', 'rbd',
+             { 'name': 'replication', 'if': 'defined(CONFIG_REPLICATION)' },
+-            'sheepdog',
++            'sheepdog', 'vitastor',
+             'ssh', 'throttle', 'vdi', 'vhdx', 'vmdk', 'vpc', 'vvfat' ] }
+ 
+ ##
+@@ -3644,6 +3644,24 @@
+             '*tag': 'str' } }
+ 
+ ##
++# @BlockdevOptionsVitastor:
++#
++# Driver specific block device options for vitastor
++#
++# @inode:       Inode number
++# @pool:        Pool ID
++# @size:        Desired image size in bytes
++# @etcd_host:   etcd connection address
++# @etcd_prefix: etcd key/value prefix
++##
++{ 'struct': 'BlockdevOptionsVitastor',
++  'data': { 'inode': 'uint64',
++            'pool': 'uint64',
++            'size': 'uint64',
++            'etcd_host': 'str',
++            '*etcd_prefix': 'str' } }
++
++##
+ # @ReplicationMode:
+ #
+ # An enumeration of replication modes.
+@@ -3988,6 +4006,7 @@
+       'replication': { 'type': 'BlockdevOptionsReplication',
+                        'if': 'defined(CONFIG_REPLICATION)' },
+       'sheepdog':   'BlockdevOptionsSheepdog',
++      'vitastor':   'BlockdevOptionsVitastor',
+       'ssh':        'BlockdevOptionsSsh',
+       'throttle':   'BlockdevOptionsThrottle',
+       'vdi':        'BlockdevOptionsGenericFormat',
+@@ -4376,6 +4395,17 @@
+             '*cluster-size' :   'size' } }
+ 
+ ##
++# @BlockdevCreateOptionsVitastor:
++#
++# Driver specific image creation options for Vitastor.
++#
++# @size: Size of the virtual disk in bytes
++##
++{ 'struct': 'BlockdevCreateOptionsVitastor',
++  'data': { 'location':         'BlockdevOptionsVitastor',
++            'size':             'size' } }
++
++##
+ # @BlockdevVmdkSubformat:
+ #
+ # Subformat options for VMDK images
+@@ -4637,6 +4667,7 @@
+       'qed':            'BlockdevCreateOptionsQed',
+       'rbd':            'BlockdevCreateOptionsRbd',
+       'sheepdog':       'BlockdevCreateOptionsSheepdog',
++      'vitastor':       'BlockdevCreateOptionsVitastor',
+       'ssh':            'BlockdevCreateOptionsSsh',
+       'vdi':            'BlockdevCreateOptionsVdi',
+       'vhdx':           'BlockdevCreateOptionsVhdx',
+Index: qemu-5.1+dfsg/scripts/modules/module_block.py
+===================================================================
+--- qemu-5.1+dfsg.orig/scripts/modules/module_block.py
++++ qemu-5.1+dfsg/scripts/modules/module_block.py
+@@ -86,6 +86,7 @@ if __name__ == '__main__':
+     output_file = sys.argv[1]
+     with open(output_file, 'w') as fheader:
+         print_top(fheader)
++        add_module(fheader, "vitastor", "vitastor", "vitastor")
+ 
+         for filename in sys.argv[2:]:
+             if os.path.isfile(filename):

rpm/build-tarball.sh

@@ -0,0 +1,51 @@
+#!/bin/bash
+# Vitastor depends on QEMU and FIO headers, but QEMU and FIO don't have -devel packages
+# So we have to copy their headers into the source tarball
+
+set -e
+
+VITASTOR=$(dirname $0)
+VITASTOR=$(realpath "$VITASTOR/..")
+
+if [ -d /opt/rh/gcc-toolset-9 ]; then
+    # CentOS 8
+    EL=8
+    . /opt/rh/gcc-toolset-9/enable
+else
+    # CentOS 7
+    EL=7
+    . /opt/rh/devtoolset-9/enable
+fi
+cd ~/rpmbuild/SPECS
+rpmbuild -bp fio.spec
+perl -i -pe 's/^make V=1/exit 0; make V=1/' qemu*.spec
+rpmbuild -bc qemu*.spec
+perl -i -pe 's/^exit 0; make V=1/make V=1/' qemu*.spec
+cd ~/rpmbuild/BUILD/qemu*/
+rm -rf $VITASTOR/qemu $VITASTOR/fio
+mkdir -p $VITASTOR/qemu/b/qemu
+make -j8 config-host.h
+cp config-host.h $VITASTOR/qemu/b/qemu
+cp -r include $VITASTOR/qemu
+if [ -f qapi-schema.json ]; then
+    # QEMU 2.0
+    make qapi-types.h
+    cp qapi-types.h $VITASTOR/qemu/b/qemu
+else
+    # QEMU 3.0+
+    make qapi
+    cp -r qapi $VITASTOR/qemu/b/qemu
+fi
+cd $VITASTOR
+sh copy-qemu-includes.sh
+rm -rf qemu
+mv qemu-copy qemu
+ln -s ~/rpmbuild/BUILD/fio*/ fio
+sh copy-fio-includes.sh
+rm fio
+mv fio-copy fio
+FIO=`rpm -qi fio | perl -e 'while(<>) { /^Epoch[\s:]+(\S+)/ && print "$1:"; /^Version[\s:]+(\S+)/ && print $1; /^Release[\s:]+(\S+)/ && print "-$1"; }'`
+QEMU=`rpm -qi qemu qemu-kvm | perl -e 'while(<>) { /^Epoch[\s:]+(\S+)/ && print "$1:"; /^Version[\s:]+(\S+)/ && print $1; /^Release[\s:]+(\S+)/ && print "-$1"; }'`
+perl -i -pe 's/(Requires:\s*fio)([^\n]+)?/$1 = '$FIO'/' $VITASTOR/rpm/vitastor-el$EL.spec
+perl -i -pe 's/(Requires:\s*qemu(?:-kvm)?)([^\n]+)?/$1 = '$QEMU'/' $VITASTOR/rpm/vitastor-el$EL.spec
+tar --transform 's#^#vitastor-0.5.10/#' --exclude 'rpm/*.rpm' -czf $VITASTOR/../vitastor-0.5.10$(rpm --eval '%dist').tar.gz *

rpm/qemu-el8.Dockerfile

@@ -0,0 +1,31 @@
+# Build packages for CentOS 8 inside a container
+# cd ..; podman build -t qemu-el8 -v `pwd`/packages:/root/packages -f rpm/qemu-el8.Dockerfile .
+
+FROM centos:8
+
+WORKDIR /root
+
+RUN rm -f /etc/yum.repos.d/CentOS-Media.repo
+RUN dnf -y install centos-release-advanced-virtualization epel-release dnf-plugins-core rpm-build
+RUN rm -rf /var/lib/dnf/*; dnf download --disablerepo='*' --enablerepo='centos-advanced-virtualization-source' --source qemu-kvm
+RUN rpm --nomd5 -i qemu*.src.rpm
+RUN cd ~/rpmbuild/SPECS && dnf builddep -y --enablerepo=PowerTools --spec qemu-kvm.spec
+
+ADD qemu-*-vitastor.patch /root/vitastor/
+
+RUN set -e; \
+    mkdir -p /root/packages/qemu-el8; \
+    rm -rf /root/packages/qemu-el8/*; \
+    rpm --nomd5 -i /root/qemu*.src.rpm; \
+    cd ~/rpmbuild/SPECS; \
+    PN=$(grep ^Patch qemu-kvm.spec | tail -n1 | perl -pe 's/Patch(\d+).*/$1/'); \
+    csplit qemu-kvm.spec "/^Patch$PN/"; \
+    cat xx00 > qemu-kvm.spec; \
+    head -n 1 xx01 >> qemu-kvm.spec; \
+    echo "Patch$((PN+1)): qemu-4.2-vitastor.patch" >> qemu-kvm.spec; \
+    tail -n +2 xx01 >> qemu-kvm.spec; \
+    perl -i -pe 's/(^Release:\s*\d+)/$1.vitastor/' qemu-kvm.spec; \
+    cp /root/vitastor/qemu-4.2-vitastor.patch ~/rpmbuild/SOURCES; \
+    rpmbuild --nocheck -ba qemu-kvm.spec; \
+    cp ~/rpmbuild/RPMS/*/*qemu* /root/packages/qemu-el8/; \
+    cp ~/rpmbuild/SRPMS/*qemu* /root/packages/qemu-el8/

rpm/qemu-kvm-el7.spec.patch

@@ -0,0 +1,257 @@
+--- qemu-kvm.spec.orig	2020-11-09 23:41:03.000000000 +0000
++++ qemu-kvm.spec	2020-12-06 10:44:24.207640963 +0000
+@@ -2,7 +2,7 @@
+ %global SLOF_gittagcommit 899d9883
+ 
+ %global have_usbredir 1
+-%global have_spice    1
++%global have_spice    0
+ %global have_opengl   1
+ %global have_fdt      0
+ %global have_gluster  1
+@@ -56,7 +56,7 @@ Requires: %{name}-block-curl = %{epoch}:
+ Requires: %{name}-block-gluster = %{epoch}:%{version}-%{release} \
+ %endif                                                           \
+ Requires: %{name}-block-iscsi = %{epoch}:%{version}-%{release}   \
+-Requires: %{name}-block-rbd = %{epoch}:%{version}-%{release}     \
++#Requires: %{name}-block-rbd = %{epoch}:%{version}-%{release}     \
+ Requires: %{name}-block-ssh = %{epoch}:%{version}-%{release}
+ 
+ # Macro to properly setup RHEL/RHEV conflict handling
+@@ -67,7 +67,7 @@ Obsoletes: %1-rhev
+ Summary: QEMU is a machine emulator and virtualizer
+ Name: qemu-kvm
+ Version: 4.2.0
+-Release: 29.vitastor%{?dist}.6
++Release: 30.vitastor%{?dist}.6
+ # Epoch because we pushed a qemu-1.0 package. AIUI this can't ever be dropped
+ Epoch: 15
+ License: GPLv2 and GPLv2+ and CC-BY
+@@ -99,8 +99,8 @@ Source30: kvm-s390x.conf
+ Source31: kvm-x86.conf
+ Source32: qemu-pr-helper.service
+ Source33: qemu-pr-helper.socket
+-Source34: 81-kvm-rhel.rules
+-Source35: udev-kvm-check.c
++#Source34: 81-kvm-rhel.rules
++#Source35: udev-kvm-check.c
+ Source36: README.tests
+ 
+ 
+@@ -825,7 +825,9 @@ Patch331: kvm-Drop-bogus-IPv6-messages.p
+ Patch333: kvm-virtiofsd-Whitelist-fchmod.patch
+ # For bz#1883869 - virtiofsd core dump in KATA Container [rhel-8.2.1.z]
+ Patch334: kvm-virtiofsd-avoid-proc-self-fd-tempdir.patch
+-Patch335: qemu-4.2-vitastor.patch
++Patch335: qemu-use-sphinx-1.2.patch
++Patch336: qemu-config-tcmalloc-warning.patch
++Patch337: qemu-4.2-vitastor.patch
+ 
+ BuildRequires: wget
+ BuildRequires: rpm-build
+@@ -842,7 +844,8 @@ BuildRequires: pciutils-devel
+ BuildRequires: libiscsi-devel
+ BuildRequires: ncurses-devel
+ BuildRequires: libattr-devel
+-BuildRequires: libusbx-devel >= 1.0.22
++BuildRequires: gperftools-devel
++BuildRequires: libusbx-devel >= 1.0.21
+ %if %{have_usbredir}
+ BuildRequires: usbredir-devel >= 0.7.1
+ %endif
+@@ -856,12 +859,12 @@ BuildRequires: virglrenderer-devel
+ # For smartcard NSS support
+ BuildRequires: nss-devel
+ %endif
+-BuildRequires: libseccomp-devel >= 2.4.0
++#Requires: libseccomp >= 2.4.0
+ # For network block driver
+ BuildRequires: libcurl-devel
+ BuildRequires: libssh-devel
+-BuildRequires: librados-devel
+-BuildRequires: librbd-devel
++#BuildRequires: librados-devel
++#BuildRequires: librbd-devel
+ %if %{have_gluster}
+ # For gluster block driver
+ BuildRequires: glusterfs-api-devel
+@@ -955,25 +958,25 @@ hardware for a full system such as a PC
+ 
+ %package -n qemu-kvm-core
+ Summary: qemu-kvm core components
++Requires: gperftools-libs
+ Requires: qemu-img = %{epoch}:%{version}-%{release}
+ %ifarch %{ix86} x86_64
+ Requires: seabios-bin >= 1.10.2-1
+ Requires: sgabios-bin
+-Requires: edk2-ovmf
+ %endif
+ %ifarch aarch64
+ Requires: edk2-aarch64
+ %endif
+ 
+ %ifnarch aarch64 s390x
+-Requires: seavgabios-bin >= 1.12.0-3
+-Requires: ipxe-roms-qemu >= 20170123-1
++Requires: seavgabios-bin >= 1.11.0-1
++Requires: ipxe-roms-qemu >= 20181214-1
++Requires: /usr/share/ipxe.efi
+ %endif
+ %ifarch %{power64}
+ Requires: SLOF >= %{SLOF_gittagdate}-1.git%{SLOF_gittagcommit}
+ %endif
+ Requires: %{name}-common = %{epoch}:%{version}-%{release}
+-Requires: libseccomp >= 2.4.0
+ # For compressed guest memory dumps
+ Requires: lzo snappy
+ %if %{have_kvm_setup}
+@@ -1085,15 +1088,15 @@ This package provides the additional iSC
+ Install this package if you want to access iSCSI volumes.
+ 
+ 
+-%package  block-rbd
+-Summary: QEMU Ceph/RBD block driver
+-Requires: %{name}-common%{?_isa} = %{epoch}:%{version}-%{release}
+-
+-%description block-rbd
+-This package provides the additional Ceph/RBD block driver for QEMU.
+-
+-Install this package if you want to access remote Ceph volumes
+-using the rbd protocol.
++#%package  block-rbd
++#Summary: QEMU Ceph/RBD block driver
++#Requires: %{name}-common%{?_isa} = %{epoch}:%{version}-%{release}
++#
++#%description block-rbd
++#This package provides the additional Ceph/RBD block driver for QEMU.
++#
++#Install this package if you want to access remote Ceph volumes
++#using the rbd protocol.
+ 
+ 
+ %package  block-ssh
+@@ -1117,12 +1120,14 @@ the Secure Shell (SSH) protocol.
+ # --build-id option is used for giving info to the debug packages.
+ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ 
+-%global block_drivers_list qcow2,raw,file,host_device,nbd,iscsi,rbd,blkdebug,luks,null-co,nvme,copy-on-read,throttle
++#%global block_drivers_list qcow2,raw,file,host_device,nbd,iscsi,rbd,blkdebug,luks,null-co,nvme,copy-on-read,throttle
++%global block_drivers_list qcow2,raw,file,host_device,nbd,iscsi,blkdebug,luks,null-co,nvme,copy-on-read,throttle
+ 
+ %if 0%{have_gluster}
+     %global block_drivers_list %{block_drivers_list},gluster
+ %endif
+ 
++[ -e /usr/bin/sphinx-build ] || ln -s sphinx-build-3 /usr/bin/sphinx-build
+ ./configure  \
+  --prefix="%{_prefix}" \
+  --libdir="%{_libdir}" \
+@@ -1152,15 +1157,15 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %else
+   --disable-numa \
+ %endif
+-  --enable-rbd \
++  --disable-rbd \
+ %if 0%{have_librdma}
+   --enable-rdma \
+ %else
+   --disable-rdma \
+ %endif
+   --disable-pvrdma \
+-  --enable-seccomp \
+-%if 0%{have_spice}
++  --disable-seccomp \
++%if %{have_spice}
+   --enable-spice \
+   --enable-smartcard \
+   --enable-virglrenderer \
+@@ -1179,7 +1184,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %else
+   --disable-usb-redir \
+ %endif
+-  --disable-tcmalloc \
++  --enable-tcmalloc \
+ %ifarch x86_64
+   --enable-libpmem \
+ %else
+@@ -1193,9 +1198,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %endif
+   --python=%{__python3} \
+   --target-list="%{buildarch}" \
+-  --block-drv-rw-whitelist=%{block_drivers_list} \
+   --audio-drv-list= \
+-  --block-drv-ro-whitelist=vmdk,vhdx,vpc,https,ssh \
+   --with-coroutine=ucontext \
+   --tls-priority=NORMAL \
+   --disable-bluez \
+@@ -1262,7 +1265,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+   --disable-sanitizers \
+   --disable-hvf \
+   --disable-whpx \
+-  --enable-malloc-trim \
++  --disable-malloc-trim \
+   --disable-membarrier \
+   --disable-vhost-crypto \
+   --disable-libxml2 \
+@@ -1308,7 +1311,7 @@ make V=1 %{?_smp_mflags} $buildldflags
+ cp -a %{kvm_target}-softmmu/qemu-system-%{kvm_target} qemu-kvm
+ 
+ gcc %{SOURCE6} $RPM_OPT_FLAGS $RPM_LD_FLAGS -o ksmctl
+-gcc %{SOURCE35} $RPM_OPT_FLAGS $RPM_LD_FLAGS -o udev-kvm-check
++#gcc %{SOURCE35} $RPM_OPT_FLAGS $RPM_LD_FLAGS -o udev-kvm-check
+ 
+ %install
+ %define _udevdir %(pkg-config --variable=udevdir udev)
+@@ -1343,8 +1346,8 @@ mkdir -p $RPM_BUILD_ROOT%{testsdir}/test
+ mkdir -p $RPM_BUILD_ROOT%{testsdir}/tests/qemu-iotests
+ mkdir -p $RPM_BUILD_ROOT%{testsdir}/scripts/qmp
+ 
+-install -p -m 0755 udev-kvm-check $RPM_BUILD_ROOT%{_udevdir}
+-install -p -m 0644 %{SOURCE34} $RPM_BUILD_ROOT%{_udevrulesdir}
++#install -p -m 0755 udev-kvm-check $RPM_BUILD_ROOT%{_udevdir}
++#install -p -m 0644 %{SOURCE34} $RPM_BUILD_ROOT%{_udevrulesdir}
+ 
+ install -m 0644 scripts/dump-guest-memory.py \
+                 $RPM_BUILD_ROOT%{_datadir}/%{name}
+@@ -1562,6 +1565,8 @@ rm -rf $RPM_BUILD_ROOT%{qemudocdir}/inte
+ # Remove spec
+ rm -rf $RPM_BUILD_ROOT%{qemudocdir}/specs
+ 
++%global __os_install_post %(echo '%{__os_install_post}' | sed -e 's!/usr/lib[^[:space:]]*/brp-python-bytecompile[[:space:]].*$!!g')
++
+ %check
+ export DIFF=diff; make check V=1
+ 
+@@ -1645,8 +1650,8 @@ useradd -r -u 107 -g qemu -G kvm -d / -s
+ %config(noreplace) %{_sysconfdir}/sysconfig/ksm
+ %{_unitdir}/ksmtuned.service
+ %{_sbindir}/ksmtuned
+-%{_udevdir}/udev-kvm-check
+-%{_udevrulesdir}/81-kvm-rhel.rules
++#%{_udevdir}/udev-kvm-check
++#%{_udevrulesdir}/81-kvm-rhel.rules
+ %ghost %{_sysconfdir}/kvm
+ %config(noreplace) %{_sysconfdir}/ksmtuned.conf
+ %dir %{_sysconfdir}/%{name}
+@@ -1711,8 +1716,8 @@ useradd -r -u 107 -g qemu -G kvm -d / -s
+ %{_libexecdir}/vhost-user-gpu
+ %{_datadir}/%{name}/vhost-user/50-qemu-gpu.json
+ %endif
+-%{_libexecdir}/virtiofsd
+-%{_datadir}/%{name}/vhost-user/50-qemu-virtiofsd.json
++#%{_libexecdir}/virtiofsd
++#%{_datadir}/%{name}/vhost-user/50-qemu-virtiofsd.json
+ 
+ %files -n qemu-img
+ %defattr(-,root,root)
+@@ -1748,8 +1753,8 @@ useradd -r -u 107 -g qemu -G kvm -d / -s
+ %files block-iscsi
+ %{_libdir}/qemu-kvm/block-iscsi.so
+ 
+-%files block-rbd
+-%{_libdir}/qemu-kvm/block-rbd.so
++#%files block-rbd
++#%{_libdir}/qemu-kvm/block-rbd.so
+ 
+ %files block-ssh
+ %{_libdir}/qemu-kvm/block-ssh.so

rpm/qemu-kvm.spec.patch

@@ -0,0 +1,29 @@
+--- qemu-kvm.spec	2020-12-05 13:13:54.388623517 +0000
++++ qemu-kvm.spec	2020-12-05 13:13:58.728696598 +0000
+@@ -67,7 +67,7 @@ Obsoletes: %1-rhev
+ Summary: QEMU is a machine emulator and virtualizer
+ Name: qemu-kvm
+ Version: 4.2.0
+-Release: 29%{?dist}.6
++Release: 29.vitastor%{?dist}.6
+ # Epoch because we pushed a qemu-1.0 package. AIUI this can't ever be dropped
+ Epoch: 15
+ License: GPLv2 and GPLv2+ and CC-BY
+@@ -825,6 +825,7 @@ Patch331: kvm-Drop-bogus-IPv6-messages.p
+ Patch333: kvm-virtiofsd-Whitelist-fchmod.patch
+ # For bz#1883869 - virtiofsd core dump in KATA Container [rhel-8.2.1.z]
+ Patch334: kvm-virtiofsd-avoid-proc-self-fd-tempdir.patch
++Patch335: qemu-4.2-vitastor.patch
+ 
+ BuildRequires: wget
+ BuildRequires: rpm-build
+@@ -1192,9 +1193,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %endif
+   --python=%{__python3} \
+   --target-list="%{buildarch}" \
+-  --block-drv-rw-whitelist=%{block_drivers_list} \
+   --audio-drv-list= \
+-  --block-drv-ro-whitelist=vmdk,vhdx,vpc,https,ssh \
+   --with-coroutine=ucontext \
+   --tls-priority=NORMAL \
+   --disable-bluez \

rpm/vitastor-el7.Dockerfile

@@ -0,0 +1,47 @@
+# Build packages for CentOS 7 inside a container
+# cd ..; podman build -t vitastor-el7 -v `pwd`/packages:/root/packages -f rpm/vitastor-el7.Dockerfile .
+# localedef -i ru_RU -f UTF-8 ru_RU.UTF-8
+
+FROM centos:7
+
+WORKDIR /root
+
+RUN rm -f /etc/yum.repos.d/CentOS-Media.repo
+RUN yum -y --enablerepo=extras install centos-release-scl epel-release yum-utils rpm-build
+RUN yum -y install https://vitastor.io/rpms/centos/7/vitastor-release-1.0-1.el7.noarch.rpm
+RUN yum -y install devtoolset-9-gcc-c++ devtoolset-9-libatomic-devel gperftools-devel qemu-kvm fio rh-nodejs12 jerasure-devel gf-complete-devel
+RUN yumdownloader --disablerepo=centos-sclo-rh --source qemu-kvm
+RUN yumdownloader --disablerepo=centos-sclo-rh --source fio
+RUN rpm --nomd5 -i qemu*.src.rpm
+RUN rpm --nomd5 -i fio*.src.rpm
+RUN rm -f /etc/yum.repos.d/CentOS-Media.repo
+RUN cd ~/rpmbuild/SPECS && yum-builddep -y --enablerepo='*' --disablerepo=centos-sclo-rh --disablerepo=centos-sclo-rh-source --disablerepo=centos-sclo-sclo-testing qemu-kvm.spec
+RUN cd ~/rpmbuild/SPECS && yum-builddep -y --enablerepo='*' --disablerepo=centos-sclo-rh --disablerepo=centos-sclo-rh-source --disablerepo=centos-sclo-sclo-testing fio.spec
+
+ADD https://vitastor.io/rpms/liburing-el7/liburing-0.7-2.el7.src.rpm /root
+
+RUN set -e; \
+    rpm -i liburing*.src.rpm; \
+    cd ~/rpmbuild/SPECS/; \
+    . /opt/rh/devtoolset-9/enable; \
+    rpmbuild -ba liburing.spec; \
+    mkdir -p /root/packages/liburing-el7; \
+    rm -rf /root/packages/liburing-el7/*; \
+    cp ~/rpmbuild/RPMS/*/liburing* /root/packages/liburing-el7/; \
+    cp ~/rpmbuild/SRPMS/liburing* /root/packages/liburing-el7/
+
+RUN rpm -i `ls /root/packages/liburing-el7/liburing-*.x86_64.rpm | grep -v debug`
+
+ADD . /root/vitastor
+
+RUN set -e; \
+    cd /root/vitastor/rpm; \
+    sh build-tarball.sh; \
+    cp /root/vitastor-0.5.10.el7.tar.gz ~/rpmbuild/SOURCES; \
+    cp vitastor-el7.spec ~/rpmbuild/SPECS/vitastor.spec; \
+    cd ~/rpmbuild/SPECS/; \
+    rpmbuild -ba vitastor.spec; \
+    mkdir -p /root/packages/vitastor-el7; \
+    rm -rf /root/packages/vitastor-el7/*; \
+    cp ~/rpmbuild/RPMS/*/vitastor* /root/packages/vitastor-el7/; \
+    cp ~/rpmbuild/SRPMS/vitastor* /root/packages/vitastor-el7/

rpm/vitastor-el7.spec

@@ -0,0 +1,69 @@
+Name:           vitastor
+Version:        0.5.10
+Release:        1%{?dist}
+Summary:        Vitastor, a fast software-defined clustered block storage
+
+License:        Vitastor Network Public License 1.1
+URL:            https://vitastor.io/
+Source0:        vitastor-0.5.10.el7.tar.gz
+
+BuildRequires:  liburing-devel >= 0.6
+BuildRequires:  gperftools-devel
+BuildRequires:  devtoolset-9-gcc-c++
+BuildRequires:  rh-nodejs12
+BuildRequires:  rh-nodejs12-npm
+BuildRequires:  jerasure-devel
+BuildRequires:  gf-complete-devel
+BuildRequires:  cmake
+Requires:       fio = 3.7-1.el7
+Requires:       qemu-kvm = 2.0.0-1.el7.6
+Requires:       rh-nodejs12
+Requires:       rh-nodejs12-npm
+Requires:       liburing >= 0.6
+Requires:       libJerasure2
+Requires:       lpsolve
+
+%description
+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).
+
+
+%prep
+%setup -q
+
+
+%build
+. /opt/rh/devtoolset-9/enable
+%cmake . -DQEMU_PLUGINDIR=qemu-kvm
+%make_build
+
+
+%install
+rm -rf $RPM_BUILD_ROOT
+%make_install
+. /opt/rh/rh-nodejs12/enable
+cd mon
+npm install
+cd ..
+mkdir -p %buildroot/usr/lib/vitastor
+cp -r mon %buildroot/usr/lib/vitastor/mon
+
+
+%files
+%doc
+%_bindir/vitastor-dump-journal
+%_bindir/vitastor-nbd
+%_bindir/vitastor-osd
+%_bindir/vitastor-rm
+%_libdir/qemu-kvm/block-vitastor.so
+%_libdir/libfio_vitastor.so
+%_libdir/libfio_vitastor_blk.so
+%_libdir/libfio_vitastor_sec.so
+%_libdir/libvitastor_blk.so
+%_libdir/libvitastor_client.so
+/usr/lib/vitastor
+
+
+%changelog

rpm/vitastor-el8.Dockerfile

@@ -0,0 +1,45 @@
+# Build packages for CentOS 8 inside a container
+# cd ..; podman build -t vitastor-el8 -v `pwd`/packages:/root/packages -f rpm/vitastor-el8.Dockerfile .
+
+FROM centos:8
+
+WORKDIR /root
+
+RUN rm -f /etc/yum.repos.d/CentOS-Media.repo
+RUN dnf -y install centos-release-advanced-virtualization epel-release dnf-plugins-core
+RUN yum -y install https://vitastor.io/rpms/centos/8/vitastor-release-1.0-1.el8.noarch.rpm
+RUN dnf --enablerepo='centos-advanced-virtualization' -y install gcc-toolset-9 gcc-toolset-9-gcc-c++ gperftools-devel qemu-kvm fio nodejs rpm-build jerasure-devel gf-complete-devel
+RUN rm -rf /var/lib/dnf/*; dnf download --disablerepo='*' --enablerepo='vitastor' --source qemu-kvm
+RUN dnf download --source fio
+RUN rpm --nomd5 -i qemu*.src.rpm
+RUN rpm --nomd5 -i fio*.src.rpm
+RUN cd ~/rpmbuild/SPECS && dnf builddep -y --enablerepo=powertools --spec qemu-kvm.spec
+RUN cd ~/rpmbuild/SPECS && dnf builddep -y --enablerepo=powertools --spec fio.spec && dnf install -y cmake
+
+ADD https://vitastor.io/rpms/liburing-el7/liburing-0.7-2.el7.src.rpm /root
+
+RUN set -e; \
+    rpm -i liburing*.src.rpm; \
+    cd ~/rpmbuild/SPECS/; \
+    . /opt/rh/gcc-toolset-9/enable; \
+    rpmbuild -ba liburing.spec; \
+    mkdir -p /root/packages/liburing-el8; \
+    rm -rf /root/packages/liburing-el8/*; \
+    cp ~/rpmbuild/RPMS/*/liburing* /root/packages/liburing-el8/; \
+    cp ~/rpmbuild/SRPMS/liburing* /root/packages/liburing-el8/
+
+RUN rpm -i `ls /root/packages/liburing-el7/liburing-*.x86_64.rpm | grep -v debug`
+
+ADD . /root/vitastor
+
+RUN set -e; \
+    cd /root/vitastor/rpm; \
+    sh build-tarball.sh; \
+    cp /root/vitastor-0.5.10.el8.tar.gz ~/rpmbuild/SOURCES; \
+    cp vitastor-el8.spec ~/rpmbuild/SPECS/vitastor.spec; \
+    cd ~/rpmbuild/SPECS/; \
+    rpmbuild -ba vitastor.spec; \
+    mkdir -p /root/packages/vitastor-el8; \
+    rm -rf /root/packages/vitastor-el8/*; \
+    cp ~/rpmbuild/RPMS/*/vitastor* /root/packages/vitastor-el8/; \
+    cp ~/rpmbuild/SRPMS/vitastor* /root/packages/vitastor-el8/

rpm/vitastor-el8.spec

@@ -0,0 +1,66 @@
+Name:           vitastor
+Version:        0.5.10
+Release:        1%{?dist}
+Summary:        Vitastor, a fast software-defined clustered block storage
+
+License:        Vitastor Network Public License 1.1
+URL:            https://vitastor.io/
+Source0:        vitastor-0.5.10.el8.tar.gz
+
+BuildRequires:  liburing-devel >= 0.6
+BuildRequires:  gperftools-devel
+BuildRequires:  gcc-toolset-9-gcc-c++
+BuildRequires:  nodejs >= 10
+BuildRequires:  jerasure-devel
+BuildRequires:  gf-complete-devel
+BuildRequires:  cmake
+Requires:       fio = 3.7-3.el8
+Requires:       qemu-kvm = 4.2.0-29.el8.6
+Requires:       nodejs >= 10
+Requires:       liburing >= 0.6
+Requires:       libJerasure2
+Requires:       lpsolve
+
+%description
+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).
+
+
+%prep
+%setup -q
+
+
+%build
+. /opt/rh/gcc-toolset-9/enable
+%cmake . -DQEMU_PLUGINDIR=qemu-kvm
+%make_build
+
+
+%install
+rm -rf $RPM_BUILD_ROOT
+%make_install
+cd mon
+npm install
+cd ..
+mkdir -p %buildroot/usr/lib/vitastor
+cp -r mon %buildroot/usr/lib/vitastor
+
+
+%files
+%doc
+%_bindir/vitastor-dump-journal
+%_bindir/vitastor-nbd
+%_bindir/vitastor-osd
+%_bindir/vitastor-rm
+%_libdir/qemu-kvm/block-vitastor.so
+%_libdir/libfio_vitastor.so
+%_libdir/libfio_vitastor_blk.so
+%_libdir/libfio_vitastor_sec.so
+%_libdir/libvitastor_blk.so
+%_libdir/libvitastor_client.so
+/usr/lib/vitastor
+
+
+%changelog

src/CMakeLists.txt

@@ -0,0 +1,188 @@
+cmake_minimum_required(VERSION 2.8)
+
+project(vitastor)
+
+include(GNUInstallDirs)
+
+set(QEMU_PLUGINDIR qemu CACHE STRING "QEMU plugin directory suffix (qemu-kvm on RHEL)")
+set(WITH_ASAN false CACHE BOOL "Build with AddressSanitizer")
+if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
+	if(EXISTS "/etc/debian_version")
+		set(CMAKE_INSTALL_LIBDIR "lib/${CMAKE_LIBRARY_ARCHITECTURE}")
+	endif()
+	set(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBDIR}")
+endif()
+
+add_definitions(-DVERSION="0.6-dev")
+add_definitions(-Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith)
+if (${WITH_ASAN})
+	add_definitions(-fsanitize=address -fno-omit-frame-pointer)
+	add_link_options(-fsanitize=address -fno-omit-frame-pointer)
+endif (${WITH_ASAN})
+
+set(CMAKE_BUILD_TYPE RelWithDebInfo)
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE}")
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_CXX_FLAGS_MINSIZEREL "${CMAKE_CXX_FLAGS_MINSIZEREL}")
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_CXX_FLAGS_MINSIZEREL "${CMAKE_CXX_FLAGS_MINSIZEREL}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO}")
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE}")
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_C_FLAGS_MINSIZEREL "${CMAKE_C_FLAGS_MINSIZEREL}")
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_C_FLAGS_RELWITHDEBINFO "${CMAKE_C_FLAGS_RELWITHDEBINFO}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_C_FLAGS_MINSIZEREL "${CMAKE_C_FLAGS_MINSIZEREL}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_C_FLAGS_RELWITHDEBINFO "${CMAKE_C_FLAGS_RELWITHDEBINFO}")
+
+find_package(PkgConfig)
+pkg_check_modules(LIBURING REQUIRED liburing)
+pkg_check_modules(GLIB REQUIRED glib-2.0)
+
+include_directories(
+	../
+	/usr/include/jerasure
+	${LIBURING_INCLUDE_DIRS}
+)
+
+# libvitastor_blk.so
+add_library(vitastor_blk SHARED
+	allocator.cpp blockstore.cpp blockstore_impl.cpp blockstore_init.cpp blockstore_open.cpp blockstore_journal.cpp blockstore_read.cpp
+	blockstore_write.cpp blockstore_sync.cpp blockstore_stable.cpp blockstore_rollback.cpp blockstore_flush.cpp crc32c.c ringloop.cpp
+)
+target_link_libraries(vitastor_blk
+	${LIBURING_LIBRARIES}
+	tcmalloc_minimal
+)
+
+# libfio_vitastor_blk.so
+add_library(fio_vitastor_blk SHARED
+	fio_engine.cpp
+	../json11/json11.cpp
+)
+target_link_libraries(fio_vitastor_blk
+	vitastor_blk
+)
+
+# vitastor-osd
+add_executable(vitastor-osd
+	osd_main.cpp osd.cpp osd_secondary.cpp msgr_receive.cpp msgr_send.cpp osd_peering.cpp osd_flush.cpp osd_peering_pg.cpp
+	osd_primary.cpp osd_primary_subops.cpp etcd_state_client.cpp messenger.cpp osd_cluster.cpp http_client.cpp osd_ops.cpp pg_states.cpp
+	osd_rmw.cpp base64.cpp timerfd_manager.cpp epoll_manager.cpp ../json11/json11.cpp
+)
+target_link_libraries(vitastor-osd
+	vitastor_blk
+	Jerasure
+)
+
+# libfio_vitastor_sec.so
+add_library(fio_vitastor_sec SHARED
+	fio_sec_osd.cpp
+	rw_blocking.cpp
+)
+target_link_libraries(fio_vitastor_sec
+	tcmalloc_minimal
+)
+
+# libvitastor_client.so
+add_library(vitastor_client SHARED
+	cluster_client.cpp epoll_manager.cpp etcd_state_client.cpp
+	messenger.cpp msgr_send.cpp msgr_receive.cpp ringloop.cpp ../json11/json11.cpp
+	http_client.cpp osd_ops.cpp pg_states.cpp timerfd_manager.cpp base64.cpp
+)
+target_link_libraries(vitastor_client
+	tcmalloc_minimal
+	${LIBURING_LIBRARIES}
+)
+
+# libfio_vitastor.so
+add_library(fio_vitastor SHARED
+	fio_cluster.cpp
+)
+target_link_libraries(fio_vitastor
+	vitastor_client
+)
+
+# vitastor-nbd
+add_executable(vitastor-nbd
+	nbd_proxy.cpp
+)
+target_link_libraries(vitastor-nbd
+	vitastor_client
+)
+
+# vitastor-rm
+add_executable(vitastor-rm
+	rm_inode.cpp
+)
+target_link_libraries(vitastor-rm
+	vitastor_client
+)
+
+# vitastor-dump-journal
+add_executable(vitastor-dump-journal
+	dump_journal.cpp crc32c.c
+)
+
+# qemu_driver.so
+add_library(qemu_proxy STATIC qemu_proxy.cpp)
+target_compile_options(qemu_proxy PUBLIC -fPIC)
+target_include_directories(qemu_proxy PUBLIC
+	../qemu/b/qemu
+	../qemu/include
+	${GLIB_INCLUDE_DIRS}
+)
+target_link_libraries(qemu_proxy
+	vitastor_client
+)
+add_library(qemu_vitastor SHARED
+	qemu_driver.c
+)
+target_link_libraries(qemu_vitastor
+	qemu_proxy
+)
+set_target_properties(qemu_vitastor PROPERTIES
+	PREFIX ""
+	OUTPUT_NAME "block-vitastor"
+)
+
+### Test stubs
+
+# stub_osd, stub_bench, osd_test
+add_executable(stub_osd stub_osd.cpp rw_blocking.cpp)
+target_link_libraries(stub_osd tcmalloc_minimal)
+add_executable(stub_bench stub_bench.cpp rw_blocking.cpp)
+target_link_libraries(stub_bench tcmalloc_minimal)
+add_executable(osd_test osd_test.cpp rw_blocking.cpp)
+target_link_libraries(osd_test tcmalloc_minimal)
+
+# osd_rmw_test
+add_executable(osd_rmw_test osd_rmw_test.cpp allocator.cpp)
+target_link_libraries(osd_rmw_test Jerasure tcmalloc_minimal)
+
+# stub_uring_osd
+add_executable(stub_uring_osd
+	stub_uring_osd.cpp epoll_manager.cpp messenger.cpp msgr_send.cpp msgr_receive.cpp ringloop.cpp timerfd_manager.cpp ../json11/json11.cpp
+)
+target_link_libraries(stub_uring_osd
+	${LIBURING_LIBRARIES}
+	tcmalloc_minimal
+)
+
+# osd_peering_pg_test
+add_executable(osd_peering_pg_test osd_peering_pg_test.cpp osd_peering_pg.cpp)
+target_link_libraries(osd_peering_pg_test tcmalloc_minimal)
+
+# test_allocator
+add_executable(test_allocator test_allocator.cpp allocator.cpp)
+
+## test_blockstore, test_shit
+#add_executable(test_blockstore test_blockstore.cpp timerfd_interval.cpp)
+#target_link_libraries(test_blockstore blockstore)
+#add_executable(test_shit test_shit.cpp osd_peering_pg.cpp)
+#target_link_libraries(test_shit ${LIBURING_LIBRARIES} m)
+
+### Install
+
+install(TARGETS vitastor-osd vitastor-dump-journal vitastor-nbd vitastor-rm RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
+install(TARGETS fio_vitastor fio_vitastor_blk fio_vitastor_sec vitastor_blk vitastor_client LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR})
+install(TARGETS qemu_vitastor LIBRARY DESTINATION /usr/${CMAKE_INSTALL_LIBDIR}/${QEMU_PLUGINDIR})

src/allocator.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include <stdexcept>
 #include "allocator.h"
@@ -13,19 +13,19 @@ allocator::allocator(uint64_t blocks)
     {
         throw std::invalid_argument("blocks");
     }
-    uint64_t p2 = 1, total = 1;
+    uint64_t p2 = 1;
+    total = 0;
     while (p2 * 64 < blocks)
     {
-        p2 = p2 * 64;
         total += p2;
+        p2 = p2 * 64;
     }
-    total -= p2;
     total += (blocks+63) / 64;
-    mask = new uint64_t[2 + total];
+    mask = new uint64_t[total];
     size = free = blocks;
     last_one_mask = (blocks % 64) == 0
         ? UINT64_MAX
-        : ~(UINT64_MAX << (64 - blocks % 64));
+        : ((1l << (blocks % 64)) - 1);
     for (uint64_t i = 0; i < total; i++)
     {
         mask[i] = 0;
@@ -99,6 +99,10 @@ uint64_t allocator::find_free()
     uint64_t p2 = 1, offset = 0, addr = 0, f, i;
     while (p2 < size)
     {
+        if (offset+addr >= total)
+        {
+            return UINT64_MAX;
+        }
         uint64_t m = mask[offset + addr];
         for (i = 0, f = 1; i < 64; i++, f <<= 1)
         {
@@ -113,11 +117,6 @@ uint64_t allocator::find_free()
             return UINT64_MAX;
         }
         addr = (addr * 64) | i;
-        if (addr >= size)
-        {
-            // No space
-            return UINT64_MAX;
-        }
         offset += p2;
         p2 = p2 * 64;
     }

src/allocator.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #pragma once
 
@@ -8,6 +8,7 @@
 // Hierarchical bitmap allocator
 class allocator
 {
+    uint64_t total;
     uint64_t size;
     uint64_t free;
     uint64_t last_one_mask;

src/base64.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "base64.h"
 

src/base64.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #pragma once
 #include <string>

src/blockstore.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "blockstore_impl.h"
 
@@ -35,12 +35,7 @@ bool blockstore_t::is_safe_to_stop()
 
 void blockstore_t::enqueue_op(blockstore_op_t *op)
 {
-    impl->enqueue_op(op, false);
-}
-
-void blockstore_t::enqueue_op_first(blockstore_op_t *op)
-{
-    impl->enqueue_op(op, true);
+    impl->enqueue_op(op);
 }
 
 std::unordered_map<object_id, uint64_t> & blockstore_t::get_unstable_writes()

src/blockstore.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #pragma once
 
@@ -9,6 +9,7 @@
 
 #include <stdint.h>
 
+#include <string>
 #include <map>
 #include <unordered_map>
 #include <functional>
@@ -174,10 +175,6 @@ public:
     // Submission
     void enqueue_op(blockstore_op_t *op);
 
-    // Insert operation into the beginning of the queue
-    // Intended for the OSD syncer "thread" to be able to stabilize something when the journal is full
-    void enqueue_op_first(blockstore_op_t *op);
-
     // Unstable writes are added here (map of object_id -> version)
     std::unordered_map<object_id, uint64_t> & get_unstable_writes();
 

src/blockstore_flush.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "blockstore_impl.h"
 
@@ -7,12 +7,17 @@ journal_flusher_t::journal_flusher_t(int flusher_count, blockstore_impl_t *bs)
 {
     this->bs = bs;
     this->flusher_count = flusher_count;
+    this->cur_flusher_count = 1;
+    this->target_flusher_count = 1;
     dequeuing = false;
+    trimming = false;
     active_flushers = 0;
     syncing_flushers = 0;
+    // FIXME: allow to configure flusher_start_threshold and journal_trim_interval
     flusher_start_threshold = bs->journal_block_size / sizeof(journal_entry_stable);
-    journal_trim_interval = flusher_start_threshold;
+    journal_trim_interval = 512;
     journal_trim_counter = 0;
+    trim_wanted = 0;
     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++)
@@ -65,14 +70,31 @@ bool journal_flusher_t::is_active()
 
 void journal_flusher_t::loop()
 {
-    for (int i = 0; (active_flushers > 0 || dequeuing) && i < flusher_count; i++)
+    target_flusher_count = bs->write_iodepth*2;
+    if (target_flusher_count <= 0)
+        target_flusher_count = 1;
+    else if (target_flusher_count > flusher_count)
+        target_flusher_count = flusher_count;
+    if (target_flusher_count > cur_flusher_count)
+        cur_flusher_count = target_flusher_count;
+    else if (target_flusher_count < cur_flusher_count)
     {
-        co[i].loop();
+        while (target_flusher_count < cur_flusher_count)
+        {
+            if (co[cur_flusher_count-1].wait_state)
+                break;
+            cur_flusher_count--;
+        }
     }
+    for (int i = 0; (active_flushers > 0 || dequeuing) && i < cur_flusher_count; i++)
+        co[i].loop();
 }
 
 void journal_flusher_t::enqueue_flush(obj_ver_id ov)
 {
+#ifdef BLOCKSTORE_DEBUG
+    printf("enqueue_flush %lx:%lx v%lu\n", ov.oid.inode, ov.oid.stripe, ov.version);
+#endif
     auto it = flush_versions.find(ov.oid);
     if (it != flush_versions.end())
     {
@@ -84,15 +106,18 @@ 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() >= flusher_start_threshold)
+    if (!dequeuing && (flush_queue.size() >= flusher_start_threshold || trim_wanted > 0))
     {
         dequeuing = true;
         bs->ringloop->wakeup();
     }
 }
 
-void journal_flusher_t::unshift_flush(obj_ver_id ov)
+void journal_flusher_t::unshift_flush(obj_ver_id ov, bool force)
 {
+#ifdef BLOCKSTORE_DEBUG
+    printf("unshift_flush %lx:%lx v%lu\n", ov.oid.inode, ov.oid.stripe, ov.version);
+#endif
     auto it = flush_versions.find(ov.oid);
     if (it != flush_versions.end())
     {
@@ -102,15 +127,38 @@ void journal_flusher_t::unshift_flush(obj_ver_id ov)
     else
     {
         flush_versions[ov.oid] = ov.version;
-        flush_queue.push_front(ov.oid);
+        if (!force)
+            flush_queue.push_front(ov.oid);
     }
-    if (!dequeuing && flush_queue.size() >= flusher_start_threshold)
+    if (force)
+        flush_queue.push_front(ov.oid);
+    if (force || !dequeuing && (flush_queue.size() >= flusher_start_threshold || trim_wanted > 0))
     {
         dequeuing = true;
         bs->ringloop->wakeup();
     }
 }
 
+void journal_flusher_t::remove_flush(object_id oid)
+{
+#ifdef BLOCKSTORE_DEBUG
+    printf("undo_flush %lx:%lx\n", oid.inode, oid.stripe);
+#endif
+    auto v_it = flush_versions.find(oid);
+    if (v_it != flush_versions.end())
+    {
+        flush_versions.erase(v_it);
+        for (auto q_it = flush_queue.begin(); q_it != flush_queue.end(); q_it++)
+        {
+            if (*q_it == oid)
+            {
+                flush_queue.erase(q_it);
+                break;
+            }
+        }
+    }
+}
+
 void journal_flusher_t::request_trim()
 {
     dequeuing = true;
@@ -118,6 +166,16 @@ void journal_flusher_t::request_trim()
     bs->ringloop->wakeup();
 }
 
+void journal_flusher_t::mark_trim_possible()
+{
+    if (trim_wanted > 0)
+    {
+        dequeuing = true;
+        journal_trim_counter++;
+        bs->ringloop->wakeup();
+    }
+}
+
 void journal_flusher_t::release_trim()
 {
     trim_wanted--;
@@ -172,9 +230,22 @@ bool journal_flusher_co::loop()
         goto resume_17;
     else if (wait_state == 18)
         goto resume_18;
+    else if (wait_state == 19)
+        goto resume_19;
+    else if (wait_state == 20)
+        goto resume_20;
+    else if (wait_state == 21)
+        goto resume_21;
 resume_0:
     if (!flusher->flush_queue.size() || !flusher->dequeuing)
     {
+stop_flusher:
+        if (flusher->trim_wanted > 0 && flusher->journal_trim_counter > 0)
+        {
+            // Attempt forced trim
+            flusher->active_flushers++;
+            goto trim_journal;
+        }
         flusher->dequeuing = false;
         wait_state = 0;
         return true;
@@ -273,9 +344,7 @@ resume_0:
 #ifdef BLOCKSTORE_DEBUG
                     printf("No older flushes, stopping\n");
 #endif
-                    flusher->dequeuing = false;
-                    wait_state = 0;
-                    return true;
+                    goto stop_flusher;
                 }
             }
         }
@@ -294,12 +363,12 @@ resume_1:
             return false;
         }
         // 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)
+        assert(!has_writes || !has_delete);
+        if (!has_writes && !has_delete || has_delete && old_clean_loc == UINT64_MAX)
         {
             // Nothing to flush
             bs->erase_dirty(dirty_start, std::next(dirty_end), clean_loc);
-            goto trim_journal;
+            goto release_oid;
         }
         if (clean_loc == UINT64_MAX)
         {
@@ -418,7 +487,12 @@ resume_1:
         else
         {
             clean_disk_entry *new_entry = (clean_disk_entry*)(meta_new.buf + meta_new.pos*bs->clean_entry_size);
-            assert(new_entry->oid.inode == 0 || new_entry->oid == cur.oid);
+            if (new_entry->oid.inode != 0 && new_entry->oid != cur.oid)
+            {
+                printf("Fatal error (metadata corruption or bug): tried to overwrite non-zero metadata entry %lu (%lx:%lx) with %lx:%lx\n",
+                    clean_loc >> bs->block_order, new_entry->oid.inode, new_entry->oid.stripe, cur.oid.inode, cur.oid.stripe);
+                exit(1);
+            }
             new_entry->oid = cur.oid;
             new_entry->version = cur.version;
             if (!bs->inmemory_meta)
@@ -474,14 +548,35 @@ resume_1:
         }
         // Update clean_db and dirty_db, free old data locations
         update_clean_db();
+#ifdef BLOCKSTORE_DEBUG
+        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
+    release_oid:
+        repeat_it = flusher->sync_to_repeat.find(cur.oid);
+        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 }, false);
+        }
+        flusher->sync_to_repeat.erase(repeat_it);
     trim_journal:
         // Clear unused part of the journal every <journal_trim_interval> flushes
         if (!((++flusher->journal_trim_counter) % flusher->journal_trim_interval) || flusher->trim_wanted > 0)
         {
             flusher->journal_trim_counter = 0;
-            if (bs->journal.trim())
+            new_trim_pos = bs->journal.get_trim_pos();
+            if (new_trim_pos != bs->journal.used_start)
             {
-                // Update journal "superblock"
+            resume_19:
+                // Wait for other coroutines trimming the journal, if any
+                if (flusher->trimming)
+                {
+                    wait_state = 19;
+                    return false;
+                }
+                flusher->trimming = true;
+                // First update journal "superblock" and only then update <used_start> in memory
                 await_sqe(12);
                 *((journal_entry_start*)flusher->journal_superblock) = {
                     .crc32 = 0,
@@ -489,7 +584,7 @@ resume_1:
                     .type = JE_START,
                     .size = sizeof(journal_entry_start),
                     .reserved = 0,
-                    .journal_start = bs->journal.used_start,
+                    .journal_start = new_trim_pos,
                 };
                 ((journal_entry_start*)flusher->journal_superblock)->crc32 = je_crc32((journal_entry*)flusher->journal_superblock);
                 data->iov = (struct iovec){ flusher->journal_superblock, bs->journal_block_size };
@@ -502,21 +597,28 @@ resume_1:
                     wait_state = 13;
                     return false;
                 }
+                if (!bs->disable_journal_fsync)
+                {
+                    await_sqe(20);
+                    my_uring_prep_fsync(sqe, bs->journal.fd, IORING_FSYNC_DATASYNC);
+                    data->iov = { 0 };
+                    data->callback = simple_callback_w;
+                resume_21:
+                    if (wait_count > 0)
+                    {
+                        wait_state = 21;
+                        return false;
+                    }
+                }
+                bs->journal.used_start = new_trim_pos;
+#ifdef BLOCKSTORE_DEBUG
+                printf("Journal trimmed to %08lx (next_free=%08lx)\n", bs->journal.used_start, bs->journal.next_free);
+#endif
+                flusher->trimming = false;
             }
         }
         // All done
-#ifdef BLOCKSTORE_DEBUG
-        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);
-        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;
     }
@@ -544,7 +646,7 @@ bool journal_flusher_co::scan_dirty(int wait_base)
         {
             char err[1024];
             snprintf(
-                err, 1024, "BUG: Unexpected dirty_entry %lx:%lx v%lu state during flush: %d",
+                err, 1024, "BUG: Unexpected dirty_entry %lx:%lx v%lu unstable 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);
@@ -721,31 +823,34 @@ bool journal_flusher_co::fsync_batch(bool fsync_meta, int wait_base)
     sync_found:
         cur_sync->ready_count++;
         flusher->syncing_flushers++;
-        if (flusher->syncing_flushers >= flusher->flusher_count || !flusher->flush_queue.size())
+    resume_1:
+        if (!cur_sync->state)
         {
-            // Sync batch is ready. Do it.
-            await_sqe(0);
-            data->iov = { 0 };
-            data->callback = simple_callback_w;
-            my_uring_prep_fsync(sqe, fsync_meta ? bs->meta_fd : bs->data_fd, IORING_FSYNC_DATASYNC);
-            cur_sync->state = 1;
-            wait_count++;
-        resume_1:
-            if (wait_count > 0)
+            if (flusher->syncing_flushers >= flusher->cur_flusher_count || !flusher->flush_queue.size())
             {
+                // Sync batch is ready. Do it.
+                await_sqe(0);
+                data->iov = { 0 };
+                data->callback = simple_callback_w;
+                my_uring_prep_fsync(sqe, fsync_meta ? bs->meta_fd : bs->data_fd, IORING_FSYNC_DATASYNC);
+                cur_sync->state = 1;
+                wait_count++;
+            resume_2:
+                if (wait_count > 0)
+                {
+                    wait_state = 2;
+                    return false;
+                }
+                // Sync completed. All previous coroutines waiting for it must be resumed
+                cur_sync->state = 2;
+                bs->ringloop->wakeup();
+            }
+            else
+            {
+                // Wait until someone else sends and completes a sync.
                 wait_state = 1;
                 return false;
             }
-            // Sync completed. All previous coroutines waiting for it must be resumed
-            cur_sync->state = 2;
-            bs->ringloop->wakeup();
-        }
-        // Wait until someone else sends and completes a sync.
-    resume_2:
-        if (!cur_sync->state)
-        {
-            wait_state = 2;
-            return false;
         }
         flusher->syncing_flushers--;
         cur_sync->ready_count--;

src/blockstore_flush.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 struct copy_buffer_t
 {
@@ -59,6 +59,8 @@ class journal_flusher_co
     uint64_t clean_bitmap_offset, clean_bitmap_len;
     void *new_clean_bitmap;
 
+    uint64_t new_trim_pos;
+
     // local: scan_dirty()
     uint64_t offset, end_offset, submit_offset, submit_len;
 
@@ -78,13 +80,14 @@ class journal_flusher_t
 {
     int trim_wanted = 0;
     bool dequeuing;
-    int flusher_count;
+    int flusher_count, cur_flusher_count, target_flusher_count;
     int flusher_start_threshold;
     journal_flusher_co *co;
     blockstore_impl_t *bs;
     friend class journal_flusher_co;
 
     int journal_trim_counter, journal_trim_interval;
+    bool trimming;
     void* journal_superblock;
 
     int active_flushers;
@@ -100,8 +103,10 @@ public:
     ~journal_flusher_t();
     void loop();
     bool is_active();
+    void mark_trim_possible();
     void request_trim();
     void release_trim();
     void enqueue_flush(obj_ver_id oid);
-    void unshift_flush(obj_ver_id oid);
+    void unshift_flush(obj_ver_id oid, bool force);
+    void remove_flush(object_id oid);
 };

src/blockstore_impl.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "blockstore_impl.h"
 
@@ -101,26 +101,14 @@ void blockstore_impl_t::loop()
     {
         // try to submit ops
         unsigned initial_ring_space = ringloop->space_left();
-        // FIXME: rework this "sync polling"
-        auto cur_sync = in_progress_syncs.begin();
-        while (cur_sync != in_progress_syncs.end())
+        // has_writes == 0 - no writes before the current queue item
+        // has_writes == 1 - some writes in progress
+        // has_writes == 2 - tried to submit some writes, but failed
+        int has_writes = 0, op_idx = 0, new_idx = 0;
+        for (; op_idx < submit_queue.size(); op_idx++, new_idx++)
         {
-            if (continue_sync(*cur_sync) != 2)
-            {
-                // List is unmodified
-                cur_sync++;
-            }
-            else
-            {
-                cur_sync = in_progress_syncs.begin();
-            }
-        }
-        auto cur = submit_queue.begin();
-        int has_writes = 0;
-        while (cur != submit_queue.end())
-        {
-            auto op_ptr = cur;
-            auto op = *(cur++);
+            auto op = submit_queue[op_idx];
+            submit_queue[new_idx] = op;
             // FIXME: This needs some simplification
             // Writes should not block reads if the ring is not full and reads don't depend on them
             // In all other cases we should stop submission
@@ -142,30 +130,33 @@ void blockstore_impl_t::loop()
             }
             unsigned ring_space = ringloop->space_left();
             unsigned prev_sqe_pos = ringloop->save();
-            bool dequeue_op = false;
+            // 0 = can't submit
+            // 1 = in progress
+            // 2 = can be removed from queue
+            int wr_st = 0;
             if (op->opcode == BS_OP_READ)
             {
-                dequeue_op = dequeue_read(op);
+                wr_st = dequeue_read(op);
             }
             else if (op->opcode == BS_OP_WRITE || op->opcode == BS_OP_WRITE_STABLE)
             {
                 if (has_writes == 2)
                 {
-                    // Some writes could not be submitted
-                    break;
+                    // Some writes already could not be submitted
+                    continue;
                 }
-                dequeue_op = dequeue_write(op);
-                has_writes = dequeue_op ? 1 : 2;
+                wr_st = dequeue_write(op);
+                has_writes = wr_st > 0 ? 1 : 2;
             }
             else if (op->opcode == BS_OP_DELETE)
             {
                 if (has_writes == 2)
                 {
-                    // Some writes could not be submitted
-                    break;
+                    // Some writes already could not be submitted
+                    continue;
                 }
-                dequeue_op = dequeue_del(op);
-                has_writes = dequeue_op ? 1 : 2;
+                wr_st = dequeue_del(op);
+                has_writes = wr_st > 0 ? 1 : 2;
             }
             else if (op->opcode == BS_OP_SYNC)
             {
@@ -178,43 +169,31 @@ void blockstore_impl_t::loop()
                     // Can't submit SYNC before previous writes
                     continue;
                 }
-                dequeue_op = dequeue_sync(op);
+                wr_st = continue_sync(op, false);
+                if (wr_st != 2)
+                {
+                    has_writes = wr_st > 0 ? 1 : 2;
+                }
             }
             else if (op->opcode == BS_OP_STABLE)
             {
-                if (has_writes == 2)
-                {
-                    // Don't submit additional flushes before completing previous LISTs
-                    break;
-                }
-                dequeue_op = dequeue_stable(op);
+                wr_st = dequeue_stable(op);
             }
             else if (op->opcode == BS_OP_ROLLBACK)
             {
-                if (has_writes == 2)
-                {
-                    // Don't submit additional flushes before completing previous LISTs
-                    break;
-                }
-                dequeue_op = dequeue_rollback(op);
+                wr_st = dequeue_rollback(op);
             }
             else if (op->opcode == BS_OP_LIST)
             {
-                // Block LIST operation by previous modifications,
-                // so it always returns a consistent state snapshot
-                if (has_writes == 2 || inflight_writes > 0)
-                    has_writes = 2;
-                else
-                {
-                    process_list(op);
-                    dequeue_op = true;
-                }
+                // LIST doesn't need to be blocked by previous modifications
+                process_list(op);
+                wr_st = 2;
             }
-            if (dequeue_op)
+            if (wr_st == 2)
             {
-                submit_queue.erase(op_ptr);
+                new_idx--;
             }
-            else
+            if (wr_st == 0)
             {
                 ringloop->restore(prev_sqe_pos);
                 if (PRIV(op)->wait_for == WAIT_SQE)
@@ -225,6 +204,14 @@ void blockstore_impl_t::loop()
                 }
             }
         }
+        if (op_idx != new_idx)
+        {
+            while (op_idx < submit_queue.size())
+            {
+                submit_queue[new_idx++] = submit_queue[op_idx++];
+            }
+            submit_queue.resize(new_idx);
+        }
         if (!readonly)
         {
             flusher->loop();
@@ -247,7 +234,7 @@ bool blockstore_impl_t::is_safe_to_stop()
 {
     // It's safe to stop blockstore when there are no in-flight operations,
     // no in-progress syncs and flusher isn't doing anything
-    if (submit_queue.size() > 0 || in_progress_syncs.size() > 0 || !readonly && flusher->is_active())
+    if (submit_queue.size() > 0 || !readonly && flusher->is_active())
     {
         return false;
     }
@@ -301,7 +288,7 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
     else if (PRIV(op)->wait_for == WAIT_JOURNAL_BUFFER)
     {
         int next = ((journal.cur_sector + 1) % journal.sector_count);
-        if (journal.sector_info[next].usage_count > 0 ||
+        if (journal.sector_info[next].flush_count > 0 ||
             journal.sector_info[next].dirty)
         {
             // do not submit
@@ -314,7 +301,7 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
     }
     else if (PRIV(op)->wait_for == WAIT_FREE)
     {
-        if (!data_alloc->get_free_count() && !flusher->is_active())
+        if (!data_alloc->get_free_count() && flusher->is_active())
         {
 #ifdef BLOCKSTORE_DEBUG
             printf("Still waiting for free space on the data device\n");
@@ -329,7 +316,7 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
     }
 }
 
-void blockstore_impl_t::enqueue_op(blockstore_op_t *op, bool first)
+void blockstore_impl_t::enqueue_op(blockstore_op_t *op)
 {
     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_WRITE_STABLE) && (
@@ -337,8 +324,7 @@ void blockstore_impl_t::enqueue_op(blockstore_op_t *op, bool first)
             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 || op->opcode == BS_OP_WRITE_STABLE))
+        readonly && op->opcode != BS_OP_READ && op->opcode != BS_OP_LIST)
     {
         // Basic verification not passed
         op->retval = -EINVAL;
@@ -388,25 +374,12 @@ void blockstore_impl_t::enqueue_op(blockstore_op_t *op, bool first)
         std::function<void (blockstore_op_t*)>(op->callback)(op);
         return;
     }
-    if (op->opcode == BS_OP_SYNC && immediate_commit == IMMEDIATE_ALL)
-    {
-        op->retval = 0;
-        std::function<void (blockstore_op_t*)>(op->callback)(op);
-        return;
-    }
     // Call constructor without allocating memory. We'll call destructor before returning op back
     new ((void*)op->private_data) blockstore_op_private_t;
     PRIV(op)->wait_for = 0;
     PRIV(op)->op_state = 0;
     PRIV(op)->pending_ops = 0;
-    if (!first)
-    {
-        submit_queue.push_back(op);
-    }
-    else
-    {
-        submit_queue.push_front(op);
-    }
+    submit_queue.push_back(op);
     ringloop->wakeup();
 }
 
@@ -531,7 +504,7 @@ void blockstore_impl_t::process_list(blockstore_op_t *op)
                     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)
+                        if (stable_count > 0 && stable[stable_count-1].oid == dirty_it->first.oid)
                         {
                             stable[stable_count-1].version = dirty_it->first.version;
                         }

src/blockstore_impl.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #pragma once
 
@@ -30,12 +30,13 @@
 #define BS_ST_BIG_WRITE 0x02
 #define BS_ST_DELETE 0x03
 
-#define BS_ST_WAIT_BIG 0x10
-#define BS_ST_IN_FLIGHT 0x20
-#define BS_ST_SUBMITTED 0x30
-#define BS_ST_WRITTEN 0x40
-#define BS_ST_SYNCED 0x50
-#define BS_ST_STABLE 0x60
+#define BS_ST_WAIT_DEL 0x10
+#define BS_ST_WAIT_BIG 0x20
+#define BS_ST_IN_FLIGHT 0x30
+#define BS_ST_SUBMITTED 0x40
+#define BS_ST_WRITTEN 0x50
+#define BS_ST_SYNCED 0x60
+#define BS_ST_STABLE 0x70
 
 #define BS_ST_INSTANT 0x100
 
@@ -153,12 +154,12 @@ struct blockstore_op_private_t
 
     // Write
     struct iovec iov_zerofill[3];
+    // Warning: must not have a default value here because it's written to before calling constructor in blockstore_write.cpp O_o
+    uint64_t real_version;
 
     // Sync
     std::vector<obj_ver_id> sync_big_writes, sync_small_writes;
     int sync_small_checked, sync_big_checked;
-    std::list<blockstore_op_t*>::iterator in_progress_ptr;
-    int prev_sync_count;
 };
 
 // https://github.com/algorithm-ninja/cpp-btree
@@ -196,7 +197,10 @@ class blockstore_impl_t
     // Suitable only for server SSDs with capacitors, requires disabled data and journal fsyncs
     int immediate_commit = IMMEDIATE_NONE;
     bool inmemory_meta = false;
-    int flusher_count;
+    // Maximum flusher count
+    unsigned flusher_count;
+    // Maximum queue depth
+    unsigned max_write_iodepth = 128;
     /******* END OF OPTIONS *******/
 
     struct ring_consumer_t ring_consumer;
@@ -204,9 +208,8 @@ class blockstore_impl_t
     blockstore_clean_db_t clean_db;
     uint8_t *clean_bitmap = NULL;
     blockstore_dirty_db_t dirty_db;
-    std::list<blockstore_op_t*> submit_queue; // FIXME: funny thing is that vector is better here
+    std::vector<blockstore_op_t*> submit_queue;
     std::vector<obj_ver_id> unsynced_big_writes, unsynced_small_writes;
-    std::list<blockstore_op_t*> in_progress_syncs; // ...and probably here, too
     allocator *data_alloc = NULL;
     uint8_t *zero_object;
 
@@ -223,10 +226,10 @@ class blockstore_impl_t
 
     struct journal_t journal;
     journal_flusher_t *flusher;
+    int write_iodepth = 0;
 
     bool live = false, queue_stall = false;
     ring_loop_t *ringloop;
-    int inflight_writes = 0;
 
     bool stop_sync_submitted;
 
@@ -265,6 +268,7 @@ class blockstore_impl_t
 
     // Write
     bool enqueue_write(blockstore_op_t *op);
+    void cancel_all_writes(blockstore_op_t *op, blockstore_dirty_db_t::iterator dirty_it, int retval);
     int dequeue_write(blockstore_op_t *op);
     int dequeue_del(blockstore_op_t *op);
     int continue_write(blockstore_op_t *op);
@@ -272,11 +276,9 @@ class blockstore_impl_t
     void handle_write_event(ring_data_t *data, blockstore_op_t *op);
 
     // Sync
-    int dequeue_sync(blockstore_op_t *op);
+    int continue_sync(blockstore_op_t *op, bool queue_has_in_progress_sync);
     void handle_sync_event(ring_data_t *data, blockstore_op_t *op);
-    int continue_sync(blockstore_op_t *op);
-    void ack_one_sync(blockstore_op_t *op);
-    int ack_sync(blockstore_op_t *op);
+    void ack_sync(blockstore_op_t *op);
 
     // Stabilize
     int dequeue_stable(blockstore_op_t *op);
@@ -316,7 +318,7 @@ public:
     bool is_stalled();
 
     // Submission
-    void enqueue_op(blockstore_op_t *op, bool first = false);
+    void enqueue_op(blockstore_op_t *op);
 
     // Unstable writes are added here (map of object_id -> version)
     std::unordered_map<object_id, uint64_t> unstable_writes;

src/blockstore_init.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "blockstore_impl.h"
 
@@ -111,7 +111,7 @@ void blockstore_init_meta::handle_entries(void* entries, unsigned count, int blo
                 {
                     // free the previous block
 #ifdef BLOCKSTORE_DEBUG
-                    printf("Free block %lu (new location is %lu)\n", clean_it->second.location >> block_order, done_cnt+i >> block_order);
+                    printf("Free block %lu (new location is %lu)\n", clean_it->second.location >> block_order, done_cnt+i);
 #endif
                     bs->data_alloc->set(clean_it->second.location >> block_order, false);
                 }
@@ -399,8 +399,7 @@ resume_1:
             }
         }
     }
-    // Trim journal on start so we don't stall when all entries are older
-    bs->journal.trim();
+    bs->flusher->mark_trim_possible();
     bs->journal.dirty_start = bs->journal.next_free;
     printf(
         "Journal entries loaded: %lu, free journal space: %lu bytes (%08lx..%08lx is used), free blocks: %lu / %lu\n",
@@ -560,9 +559,54 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                 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
+                    je->big_write.oid.inode, je->big_write.oid.stripe, je->big_write.version, je->big_write.location >> bs->block_order
                 );
 #endif
+                auto dirty_it = bs->dirty_db.upper_bound((obj_ver_id){
+                    .oid = je->big_write.oid,
+                    .version = UINT64_MAX,
+                });
+                if (dirty_it != bs->dirty_db.begin() && bs->dirty_db.size() > 0)
+                {
+                    dirty_it--;
+                    if (dirty_it->first.oid == je->big_write.oid &&
+                        dirty_it->first.version >= je->big_write.version &&
+                        (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_DELETE)
+                    {
+                        // It is allowed to overwrite a deleted object with a
+                        // version number smaller than deletion version number,
+                        // because the presence of a BIG_WRITE entry means that
+                        // its data and metadata are already flushed.
+                        // We don't know if newer versions are flushed, but
+                        // the previous delete definitely is.
+                        // So we flush previous dirty entries, but retain the clean one.
+                        // This feature is required for writes happening shortly
+                        // after deletes.
+                        auto dirty_end = dirty_it;
+                        dirty_end++;
+                        while (1)
+                        {
+                            if (dirty_it == bs->dirty_db.begin())
+                            {
+                                break;
+                            }
+                            dirty_it--;
+                            if (dirty_it->first.oid != je->big_write.oid)
+                            {
+                                dirty_it++;
+                                break;
+                            }
+                        }
+                        auto clean_it = bs->clean_db.find(je->big_write.oid);
+                        bs->erase_dirty(
+                            dirty_it, dirty_end,
+                            clean_it != bs->clean_db.end() ? clean_it->second.location : UINT64_MAX
+                        );
+                        // Remove it from the flusher's queue, too
+                        // Otherwise it may end up referring to a small unstable write after reading the rest of the journal
+                        bs->flusher->remove_flush(je->big_write.oid);
+                    }
+                }
                 auto clean_it = bs->clean_db.find(je->big_write.oid);
                 if (clean_it == bs->clean_db.end() ||
                     clean_it->second.version < je->big_write.version)
@@ -585,6 +629,12 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
 #endif
                     bs->data_alloc->set(je->big_write.location >> bs->block_order, true);
                     bs->journal.used_sectors[proc_pos]++;
+#ifdef BLOCKSTORE_DEBUG
+                    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_BIG_WRITE_INSTANT)

src/blockstore_init.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #pragma once
 

src/blockstore_journal.cpp

@@ -1,12 +1,12 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "blockstore_impl.h"
 
 blockstore_journal_check_t::blockstore_journal_check_t(blockstore_impl_t *bs)
 {
     this->bs = bs;
-    sectors_required = 0;
+    sectors_to_write = 0;
     next_pos = bs->journal.next_free;
     next_sector = bs->journal.cur_sector;
     first_sector = -1;
@@ -20,23 +20,26 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
     int required = entries_required;
     while (1)
     {
-        int fits = bs->journal.no_same_sector_overwrites && bs->journal.sector_info[next_sector].written
+        int fits = bs->journal.no_same_sector_overwrites && next_pos == bs->journal.next_free && bs->journal.sector_info[next_sector].written
             ? 0
             : (bs->journal.block_size - next_in_pos) / size;
         if (fits > 0)
         {
+            if (fits > required)
+            {
+                fits = required;
+            }
             if (first_sector == -1)
             {
                 first_sector = next_sector;
             }
             required -= fits;
             next_in_pos += fits * size;
-            sectors_required++;
+            sectors_to_write++;
         }
         else if (bs->journal.sector_info[next_sector].dirty)
         {
-            // sectors_required is more like "sectors to write"
-            sectors_required++;
+            sectors_to_write++;
         }
         if (required <= 0)
         {
@@ -59,7 +62,7 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
                 " is too small for a batch of "+std::to_string(entries_required)+" entries of "+std::to_string(size)+" bytes"
             );
         }
-        if (bs->journal.sector_info[next_sector].usage_count > 0 ||
+        if (bs->journal.sector_info[next_sector].flush_count > 0 ||
             bs->journal.sector_info[next_sector].dirty)
         {
             // No memory buffer available. Wait for it.
@@ -71,17 +74,18 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
                     dirty++;
                     used++;
                 }
-                if (bs->journal.sector_info[i].usage_count > 0)
+                if (bs->journal.sector_info[i].flush_count > 0)
                 {
                     used++;
                 }
             }
             // In fact, it's even more rare than "ran out of journal space", so print a warning
             printf(
-                "Ran out of journal sector buffers: %d/%lu buffers used (%d dirty), next buffer (%ld) is %s and flushed %lu times\n",
+                "Ran out of journal sector buffers: %d/%lu buffers used (%d dirty), next buffer (%ld)"
+                " is %s and flushed %lu times. Consider increasing \'journal_sector_buffer_count\'\n",
                 used, bs->journal.sector_count, dirty, next_sector,
                 bs->journal.sector_info[next_sector].dirty ? "dirty" : "not dirty",
-                bs->journal.sector_info[next_sector].usage_count
+                bs->journal.sector_info[next_sector].flush_count
             );
             PRIV(op)->wait_for = WAIT_JOURNAL_BUFFER;
             return 0;
@@ -100,10 +104,8 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
     {
         // No space in the journal. Wait until used_start changes.
         printf(
-            "Ran out of journal space (free space: %lu bytes)\n",
-            (bs->journal.next_free >= bs->journal.used_start
-                ? bs->journal.len-bs->journal.block_size - (bs->journal.next_free-bs->journal.used_start)
-                : bs->journal.used_start - bs->journal.next_free)
+            "Ran out of journal space (used_start=%08lx, next_free=%08lx, dirty_start=%08lx)\n",
+            bs->journal.used_start, bs->journal.next_free, bs->journal.dirty_start
         );
         PRIV(op)->wait_for = WAIT_JOURNAL;
         bs->flusher->request_trim();
@@ -115,22 +117,21 @@ 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 ||
-        journal.no_same_sector_overwrites && journal.sector_info[journal.cur_sector].written)
+    if (!journal.entry_fits(size))
     {
         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)
+        if (journal.sector_info[journal.cur_sector].flush_count > 0)
         {
             // Also select next sector buffer in memory
             journal.cur_sector = ((journal.cur_sector + 1) % journal.sector_count);
-            assert(!journal.sector_info[journal.cur_sector].usage_count);
+            assert(!journal.sector_info[journal.cur_sector].flush_count);
         }
         else
         {
             journal.dirty_start = journal.next_free;
         }
+        journal.sector_info[journal.cur_sector].written = false;
         journal.sector_info[journal.cur_sector].offset = journal.next_free;
         journal.in_sector_pos = 0;
         journal.next_free = (journal.next_free+journal.block_size) < journal.len ? journal.next_free + journal.block_size : journal.block_size;
@@ -156,7 +157,7 @@ void prepare_journal_sector_write(journal_t & journal, int cur_sector, io_uring_
 {
     journal.sector_info[cur_sector].dirty = false;
     journal.sector_info[cur_sector].written = true;
-    journal.sector_info[cur_sector].usage_count++;
+    journal.sector_info[cur_sector].flush_count++;
     ring_data_t *data = ((ring_data_t*)sqe->user_data);
     data->iov = (struct iovec){
         (journal.inmemory
@@ -183,7 +184,7 @@ journal_t::~journal_t()
     buffer = NULL;
 }
 
-bool journal_t::trim()
+uint64_t journal_t::get_trim_pos()
 {
     auto journal_used_it = used_sectors.lower_bound(used_start);
 #ifdef BLOCKSTORE_DEBUG
@@ -201,26 +202,19 @@ bool journal_t::trim()
         if (journal_used_it == used_sectors.end())
         {
             // Journal is empty
-            used_start = next_free;
+            return next_free;
         }
         else
         {
-            used_start = journal_used_it->first;
-            // next_free does not need updating here
+            // next_free does not need updating during trim
+            return journal_used_it->first;
         }
     }
     else if (journal_used_it->first > used_start)
     {
         // Journal is cleared up to <journal_used_it>
-        used_start = journal_used_it->first;
+        return journal_used_it->first;
     }
-    else
-    {
-        // Can't trim journal
-        return false;
-    }
-#ifdef BLOCKSTORE_DEBUG
-    printf("Journal trimmed to %08lx (next_free=%08lx)\n", used_start, next_free);
-#endif
-    return true;
+    // Can't trim journal
+    return used_start;
 }

src/blockstore_journal.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #pragma once
 
@@ -10,6 +10,8 @@
 #define JOURNAL_BUFFER_SIZE 4*1024*1024
 
 // We reserve some extra space for future stabilize requests during writes
+// FIXME: This value should be dynamic i.e. Blockstore ideally shouldn't allow
+// writing more than can be stabilized afterwards
 #define JOURNAL_STABILIZE_RESERVATION 65536
 
 // Journal entries
@@ -131,7 +133,7 @@ inline uint32_t je_crc32(journal_entry *je)
 struct journal_sector_info_t
 {
     uint64_t offset;
-    uint64_t usage_count;
+    uint64_t flush_count;
     bool written;
     bool dirty;
 };
@@ -167,13 +169,19 @@ struct journal_t
 
     ~journal_t();
     bool trim();
+    uint64_t get_trim_pos();
+    inline bool entry_fits(int size)
+    {
+        return !(block_size - in_sector_pos < size ||
+            no_same_sector_overwrites && sector_info[cur_sector].written);
+    }
 };
 
 struct blockstore_journal_check_t
 {
     blockstore_impl_t *bs;
     uint64_t next_pos, next_sector, next_in_pos;
-    int sectors_required, first_sector;
+    int sectors_to_write, first_sector;
     bool right_dir; // writing to the end or the beginning of the ring buffer
 
     blockstore_journal_check_t(blockstore_impl_t *bs);

src/blockstore_open.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include <sys/file.h>
 #include "blockstore_impl.h"
@@ -70,6 +70,7 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
     meta_block_size = strtoull(config["meta_block_size"].c_str(), NULL, 10);
     bitmap_granularity = strtoull(config["bitmap_granularity"].c_str(), NULL, 10);
     flusher_count = strtoull(config["flusher_count"].c_str(), NULL, 10);
+    max_write_iodepth = strtoull(config["max_write_iodepth"].c_str(), NULL, 10);
     // Validate
     if (!block_size)
     {
@@ -83,6 +84,10 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
     {
         flusher_count = 32;
     }
+    if (!max_write_iodepth)
+    {
+        max_write_iodepth = 128;
+    }
     if (!disk_alignment)
     {
         disk_alignment = 4096;

src/blockstore_read.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "blockstore_impl.h"
 
@@ -112,7 +112,7 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
         read_op->version = 0;
         read_op->retval = read_op->len;
         FINISH_OP(read_op);
-        return 1;
+        return 2;
     }
     uint64_t fulfilled = 0;
     PRIV(read_op)->pending_ops = 0;
@@ -191,8 +191,8 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
                     if (bmp_end > bmp_start)
                     {
                         // fill with zeroes
-                        fulfill_read(read_op, fulfilled, bmp_start * bitmap_granularity,
-                            bmp_end * bitmap_granularity, (BS_ST_DELETE | BS_ST_STABLE), 0, 0);
+                        assert(fulfill_read(read_op, fulfilled, bmp_start * bitmap_granularity,
+                            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)
@@ -218,7 +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, (BS_ST_DELETE | BS_ST_STABLE), 0, 0);
+        assert(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;
@@ -232,10 +232,10 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
         }
         read_op->retval = read_op->len;
         FINISH_OP(read_op);
-        return 1;
+        return 2;
     }
     read_op->retval = 0;
-    return 1;
+    return 2;
 }
 
 void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op)

src/blockstore_rollback.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "blockstore_impl.h"
 
@@ -9,10 +9,14 @@ int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
     {
         return continue_rollback(op);
     }
-    obj_ver_id* v;
+    obj_ver_id *v, *nv;
     int i, todo = op->len;
-    for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
+    for (i = 0, v = (obj_ver_id*)op->buf, nv = (obj_ver_id*)op->buf; i < op->len; i++, v++, nv++)
     {
+        if (nv != v)
+        {
+            *nv = *v;
+        }
         // Check that there are some versions greater than v->version (which may be zero),
         // check that they're unstable, synced, and not currently written to
         auto dirty_it = dirty_db.lower_bound((obj_ver_id){
@@ -21,31 +25,32 @@ int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
         });
         if (dirty_it == dirty_db.begin())
         {
-            if (v->version == 0)
-            {
-                // Already rolled back
-                // FIXME Skip this object version
-            }
-        bad_op:
-            op->retval = -ENOENT;
-            FINISH_OP(op);
-            return 1;
+skip_ov:
+            // Already rolled back, skip this object version
+            todo--;
+            nv--;
+            continue;
         }
         else
         {
             dirty_it--;
             if (dirty_it->first.oid != v->oid || dirty_it->first.version < v->version)
             {
-                goto bad_op;
+                goto skip_ov;
             }
             while (dirty_it->first.oid == v->oid && dirty_it->first.version > v->version)
             {
-                if (!IS_SYNCED(dirty_it->second.state) ||
+                if (IS_IN_FLIGHT(dirty_it->second.state))
+                {
+                    // Object write is still in progress. Wait until the write request completes
+                    return 0;
+                }
+                else if (!IS_SYNCED(dirty_it->second.state) ||
                     IS_STABLE(dirty_it->second.state))
                 {
                     op->retval = -EBUSY;
                     FINISH_OP(op);
-                    return 1;
+                    return 2;
                 }
                 if (dirty_it == dirty_db.begin())
                 {
@@ -55,6 +60,14 @@ int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
             }
         }
     }
+    op->len = todo;
+    if (!todo)
+    {
+        // Already rolled back
+        op->retval = 0;
+        FINISH_OP(op);
+        return 2;
+    }
     // Check journal space
     blockstore_journal_check_t space_check(this);
     if (!space_check.check_available(op, todo, sizeof(journal_entry_rollback), 0))
@@ -62,43 +75,38 @@ int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
         return 0;
     }
     // There is sufficient space. Get SQEs
-    struct io_uring_sqe *sqe[space_check.sectors_required];
-    for (i = 0; i < space_check.sectors_required; i++)
+    struct io_uring_sqe *sqe[space_check.sectors_to_write];
+    for (i = 0; i < space_check.sectors_to_write; i++)
     {
         BS_SUBMIT_GET_SQE_DECL(sqe[i]);
     }
     // Prepare and submit journal entries
     auto cb = [this, op](ring_data_t *data) { handle_rollback_event(data, op); };
     int s = 0, cur_sector = -1;
-    if ((journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_rollback) &&
-        journal.sector_info[journal.cur_sector].dirty)
-    {
-        if (cur_sector == -1)
-            PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-        cur_sector = journal.cur_sector;
-        prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-    }
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
     {
+        if (!journal.entry_fits(sizeof(journal_entry_rollback)) &&
+            journal.sector_info[journal.cur_sector].dirty)
+        {
+            if (cur_sector == -1)
+                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+            prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+            cur_sector = journal.cur_sector;
+        }
         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;
         je->oid = v->oid;
         je->version = v->version;
         je->crc32 = je_crc32((journal_entry*)je);
         journal.crc32_last = je->crc32;
-        if (cur_sector != journal.cur_sector)
-        {
-            if (cur_sector == -1)
-                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-            cur_sector = journal.cur_sector;
-            prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-        }
     }
+    prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+    assert(s == space_check.sectors_to_write);
+    if (cur_sector == -1)
+        PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
     PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
     PRIV(op)->pending_ops = s;
     PRIV(op)->op_state = 1;
-    inflight_writes++;
     return 1;
 }
 
@@ -118,11 +126,8 @@ resume_2:
 resume_3:
     if (!disable_journal_fsync)
     {
-        io_uring_sqe *sqe = get_sqe();
-        if (!sqe)
-        {
-            return 0;
-        }
+        io_uring_sqe *sqe;
+        BS_SUBMIT_GET_SQE_DECL(sqe);
         ring_data_t *data = ((ring_data_t*)sqe->user_data);
         my_uring_prep_fsync(sqe, journal.fd, IORING_FSYNC_DATASYNC);
         data->iov = { 0 };
@@ -139,12 +144,11 @@ resume_5:
     {
         mark_rolled_back(*v);
     }
-    journal.trim();
-    inflight_writes--;
+    flusher->mark_trim_possible();
     // Acknowledge op
     op->retval = 0;
     FINISH_OP(op);
-    return 1;
+    return 2;
 }
 
 void blockstore_impl_t::mark_rolled_back(const obj_ver_id & ov)
@@ -200,7 +204,6 @@ void blockstore_impl_t::handle_rollback_event(ring_data_t *data, blockstore_op_t
     live = true;
     if (data->res != data->iov.iov_len)
     {
-        inflight_writes--;
         throw std::runtime_error(
             "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
             "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
@@ -210,19 +213,44 @@ void blockstore_impl_t::handle_rollback_event(ring_data_t *data, blockstore_op_t
     if (PRIV(op)->pending_ops == 0)
     {
         PRIV(op)->op_state++;
-        if (!continue_rollback(op))
-        {
-            submit_queue.push_front(op);
-        }
+        ringloop->wakeup();
     }
 }
 
 void blockstore_impl_t::erase_dirty(blockstore_dirty_db_t::iterator dirty_start, blockstore_dirty_db_t::iterator dirty_end, uint64_t clean_loc)
 {
-    auto dirty_it = dirty_end;
-    while (dirty_it != dirty_start)
+    if (dirty_end == dirty_start)
     {
+        return;
+    }
+    auto dirty_it = dirty_end;
+    dirty_it--;
+    if (IS_DELETE(dirty_it->second.state))
+    {
+        object_id oid = dirty_it->first.oid;
+#ifdef BLOCKSTORE_DEBUG
+        printf("Unblock writes-after-delete %lx:%lx v%lx\n", oid.inode, oid.stripe, dirty_it->first.version);
+#endif
+        dirty_it = dirty_end;
+        // Unblock operations blocked by delete flushing
+        uint32_t next_state = BS_ST_IN_FLIGHT;
+        while (dirty_it != dirty_db.end() && dirty_it->first.oid == oid)
+        {
+            if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_DEL)
+            {
+                dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | next_state;
+                if (IS_BIG_WRITE(dirty_it->second.state))
+                {
+                    next_state = BS_ST_WAIT_BIG;
+                }
+            }
+            dirty_it++;
+        }
+        dirty_it = dirty_end;
         dirty_it--;
+    }
+    while (1)
+    {
         if (IS_BIG_WRITE(dirty_it->second.state) && dirty_it->second.location != clean_loc)
         {
 #ifdef BLOCKSTORE_DEBUG
@@ -241,6 +269,11 @@ void blockstore_impl_t::erase_dirty(blockstore_dirty_db_t::iterator dirty_start,
         {
             journal.used_sectors.erase(dirty_it->second.journal_sector);
         }
+        if (dirty_it == dirty_start)
+        {
+            break;
+        }
+        dirty_it--;
     }
     dirty_db.erase(dirty_start, dirty_end);
 }

src/blockstore_stable.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "blockstore_impl.h"
 
@@ -60,19 +60,24 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
                 // No such object version
                 op->retval = -ENOENT;
                 FINISH_OP(op);
-                return 1;
+                return 2;
             }
             else
             {
                 // Already stable
             }
         }
+        else if (IS_IN_FLIGHT(dirty_it->second.state))
+        {
+            // Object write is still in progress. Wait until the write request completes
+            return 0;
+        }
         else if (!IS_SYNCED(dirty_it->second.state))
         {
             // Object not synced yet. Caller must sync it first
             op->retval = -EBUSY;
             FINISH_OP(op);
-            return 1;
+            return 2;
         }
         else if (!IS_STABLE(dirty_it->second.state))
         {
@@ -84,7 +89,7 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
         // Already stable
         op->retval = 0;
         FINISH_OP(op);
-        return 1;
+        return 2;
     }
     // Check journal space
     blockstore_journal_check_t space_check(this);
@@ -93,44 +98,39 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
         return 0;
     }
     // There is sufficient space. Get SQEs
-    struct io_uring_sqe *sqe[space_check.sectors_required];
-    for (i = 0; i < space_check.sectors_required; i++)
+    struct io_uring_sqe *sqe[space_check.sectors_to_write];
+    for (i = 0; i < space_check.sectors_to_write; i++)
     {
         BS_SUBMIT_GET_SQE_DECL(sqe[i]);
     }
     // Prepare and submit journal entries
     auto cb = [this, op](ring_data_t *data) { handle_stable_event(data, op); };
     int s = 0, cur_sector = -1;
-    if ((journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_stable) &&
-        journal.sector_info[journal.cur_sector].dirty)
-    {
-        if (cur_sector == -1)
-            PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-        cur_sector = journal.cur_sector;
-        prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-    }
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
     {
         // FIXME: Only stabilize versions that aren't stable yet
+        if (!journal.entry_fits(sizeof(journal_entry_stable)) &&
+            journal.sector_info[journal.cur_sector].dirty)
+        {
+            if (cur_sector == -1)
+                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+            prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+            cur_sector = journal.cur_sector;
+        }
         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;
         je->oid = v->oid;
         je->version = v->version;
         je->crc32 = je_crc32((journal_entry*)je);
         journal.crc32_last = je->crc32;
-        if (cur_sector != journal.cur_sector)
-        {
-            if (cur_sector == -1)
-                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-            cur_sector = journal.cur_sector;
-            prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-        }
     }
+    prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+    assert(s == space_check.sectors_to_write);
+    if (cur_sector == -1)
+        PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
     PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
     PRIV(op)->pending_ops = s;
     PRIV(op)->op_state = 1;
-    inflight_writes++;
     return 1;
 }
 
@@ -150,11 +150,8 @@ resume_2:
 resume_3:
     if (!disable_journal_fsync)
     {
-        io_uring_sqe *sqe = get_sqe();
-        if (!sqe)
-        {
-            return 0;
-        }
+        io_uring_sqe *sqe;
+        BS_SUBMIT_GET_SQE_DECL(sqe);
         ring_data_t *data = ((ring_data_t*)sqe->user_data);
         my_uring_prep_fsync(sqe, journal.fd, IORING_FSYNC_DATASYNC);
         data->iov = { 0 };
@@ -173,11 +170,10 @@ resume_5:
         // Mark all dirty_db entries up to op->version as stable
         mark_stable(*v);
     }
-    inflight_writes--;
     // Acknowledge op
     op->retval = 0;
     FINISH_OP(op);
-    return 1;
+    return 2;
 }
 
 void blockstore_impl_t::mark_stable(const obj_ver_id & v)
@@ -205,9 +201,6 @@ void blockstore_impl_t::mark_stable(const obj_ver_id & v)
                 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);
@@ -223,7 +216,6 @@ void blockstore_impl_t::handle_stable_event(ring_data_t *data, blockstore_op_t *
     live = true;
     if (data->res != data->iov.iov_len)
     {
-        inflight_writes--;
         throw std::runtime_error(
             "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
             "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
@@ -233,9 +225,6 @@ void blockstore_impl_t::handle_stable_event(ring_data_t *data, blockstore_op_t *
     if (PRIV(op)->pending_ops == 0)
     {
         PRIV(op)->op_state++;
-        if (!continue_stable(op))
-        {
-            submit_queue.push_front(op);
-        }
+        ringloop->wakeup();
     }
 }

src/blockstore_sync.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "blockstore_impl.h"
 
@@ -12,8 +12,15 @@
 #define SYNC_JOURNAL_SYNC_SENT 7
 #define SYNC_DONE 8
 
-int blockstore_impl_t::dequeue_sync(blockstore_op_t *op)
+int blockstore_impl_t::continue_sync(blockstore_op_t *op, bool queue_has_in_progress_sync)
 {
+    if (immediate_commit == IMMEDIATE_ALL)
+    {
+        // We can return immediately because sync is only dequeued after all previous writes
+        op->retval = 0;
+        FINISH_OP(op);
+        return 2;
+    }
     if (PRIV(op)->op_state == 0)
     {
         stop_sync_submitted = false;
@@ -29,34 +36,15 @@ int blockstore_impl_t::dequeue_sync(blockstore_op_t *op)
             PRIV(op)->op_state = SYNC_HAS_SMALL;
         else
             PRIV(op)->op_state = SYNC_DONE;
-        // Always add sync to in_progress_syncs because we clear unsynced_big_writes and unsynced_small_writes
-        PRIV(op)->prev_sync_count = in_progress_syncs.size();
-        PRIV(op)->in_progress_ptr = in_progress_syncs.insert(in_progress_syncs.end(), op);
     }
-    continue_sync(op);
-    // Always dequeue because we always add syncs to in_progress_syncs
-    return 1;
-}
-
-int blockstore_impl_t::continue_sync(blockstore_op_t *op)
-{
-    auto cb = [this, op](ring_data_t *data) { handle_sync_event(data, op); };
     if (PRIV(op)->op_state == SYNC_HAS_SMALL)
     {
         // No big writes, just fsync the journal
-        for (; PRIV(op)->sync_small_checked < PRIV(op)->sync_small_writes.size(); PRIV(op)->sync_small_checked++)
-        {
-            if (IS_IN_FLIGHT(dirty_db[PRIV(op)->sync_small_writes[PRIV(op)->sync_small_checked]].state))
-            {
-                // Wait for small inflight writes to complete
-                return 0;
-            }
-        }
         if (journal.sector_info[journal.cur_sector].dirty)
         {
             // Write out the last journal sector if it happens to be dirty
             BS_SUBMIT_GET_ONLY_SQE(sqe);
-            prepare_journal_sector_write(journal, journal.cur_sector, sqe, cb);
+            prepare_journal_sector_write(journal, journal.cur_sector, sqe, [this, op](ring_data_t *data) { handle_sync_event(data, op); });
             PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
             PRIV(op)->pending_ops = 1;
             PRIV(op)->op_state = SYNC_JOURNAL_WRITE_SENT;
@@ -69,21 +57,13 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
     }
     if (PRIV(op)->op_state == SYNC_HAS_BIG)
     {
-        for (; PRIV(op)->sync_big_checked < PRIV(op)->sync_big_writes.size(); PRIV(op)->sync_big_checked++)
-        {
-            if (IS_IN_FLIGHT(dirty_db[PRIV(op)->sync_big_writes[PRIV(op)->sync_big_checked]].state))
-            {
-                // Wait for big inflight writes to complete
-                return 0;
-            }
-        }
         // 1st step: fsync data
         if (!disable_data_fsync)
         {
             BS_SUBMIT_GET_SQE(sqe, data);
             my_uring_prep_fsync(sqe, data_fd, IORING_FSYNC_DATASYNC);
             data->iov = { 0 };
-            data->callback = cb;
+            data->callback = [this, op](ring_data_t *data) { handle_sync_event(data, op); };
             PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
             PRIV(op)->pending_ops = 1;
             PRIV(op)->op_state = SYNC_DATA_SYNC_SENT;
@@ -96,46 +76,37 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
     }
     if (PRIV(op)->op_state == SYNC_DATA_SYNC_DONE)
     {
-        for (; PRIV(op)->sync_small_checked < PRIV(op)->sync_small_writes.size(); PRIV(op)->sync_small_checked++)
-        {
-            if (IS_IN_FLIGHT(dirty_db[PRIV(op)->sync_small_writes[PRIV(op)->sync_small_checked]].state))
-            {
-                // Wait for small inflight writes to complete
-                return 0;
-            }
-        }
         // 2nd step: Data device is synced, prepare & write journal entries
         // Check space in the journal and journal memory buffers
         blockstore_journal_check_t space_check(this);
-        if (!space_check.check_available(op, PRIV(op)->sync_big_writes.size(), sizeof(journal_entry_big_write), 0))
+        if (!space_check.check_available(op, PRIV(op)->sync_big_writes.size(), sizeof(journal_entry_big_write), JOURNAL_STABILIZE_RESERVATION))
         {
             return 0;
         }
         // Get SQEs. Don't bother about merging, submit each journal sector as a separate request
-        struct io_uring_sqe *sqe[space_check.sectors_required];
-        for (int i = 0; i < space_check.sectors_required; i++)
+        struct io_uring_sqe *sqe[space_check.sectors_to_write];
+        for (int i = 0; i < space_check.sectors_to_write; i++)
         {
             BS_SUBMIT_GET_SQE_DECL(sqe[i]);
         }
         // Prepare and submit journal entries
         auto it = PRIV(op)->sync_big_writes.begin();
         int s = 0, cur_sector = -1;
-        if ((journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_big_write) &&
-            journal.sector_info[journal.cur_sector].dirty)
-        {
-            if (cur_sector == -1)
-                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-            cur_sector = journal.cur_sector;
-            prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-        }
         while (it != PRIV(op)->sync_big_writes.end())
         {
+            if (!journal.entry_fits(sizeof(journal_entry_big_write)) &&
+                journal.sector_info[journal.cur_sector].dirty)
+            {
+                if (cur_sector == -1)
+                    PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+                prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], [this, op](ring_data_t *data) { handle_sync_event(data, op); });
+                cur_sector = journal.cur_sector;
+            }
             journal_entry_big_write *je = (journal_entry_big_write*)prefill_single_journal_entry(
                 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(
@@ -152,14 +123,11 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
             je->crc32 = je_crc32((journal_entry*)je);
             journal.crc32_last = je->crc32;
             it++;
-            if (cur_sector != journal.cur_sector)
-            {
-                if (cur_sector == -1)
-                    PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-                cur_sector = journal.cur_sector;
-                prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-            }
         }
+        prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], [this, op](ring_data_t *data) { handle_sync_event(data, op); });
+        assert(s == space_check.sectors_to_write);
+        if (cur_sector == -1)
+            PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
         PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
         PRIV(op)->pending_ops = s;
         PRIV(op)->op_state = SYNC_JOURNAL_WRITE_SENT;
@@ -172,7 +140,7 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
             BS_SUBMIT_GET_SQE(sqe, data);
             my_uring_prep_fsync(sqe, journal.fd, IORING_FSYNC_DATASYNC);
             data->iov = { 0 };
-            data->callback = cb;
+            data->callback = [this, op](ring_data_t *data) { handle_sync_event(data, op); };
             PRIV(op)->pending_ops = 1;
             PRIV(op)->op_state = SYNC_JOURNAL_SYNC_SENT;
             return 1;
@@ -182,9 +150,10 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
             PRIV(op)->op_state = SYNC_DONE;
         }
     }
-    if (PRIV(op)->op_state == SYNC_DONE)
+    if (PRIV(op)->op_state == SYNC_DONE && !queue_has_in_progress_sync)
     {
-        return ack_sync(op);
+        ack_sync(op);
+        return 2;
     }
     return 1;
 }
@@ -216,42 +185,16 @@ void blockstore_impl_t::handle_sync_event(ring_data_t *data, blockstore_op_t *op
         else if (PRIV(op)->op_state == SYNC_JOURNAL_SYNC_SENT)
         {
             PRIV(op)->op_state = SYNC_DONE;
-            ack_sync(op);
         }
         else
         {
             throw std::runtime_error("BUG: unexpected sync op state");
         }
+        ringloop->wakeup();
     }
 }
 
-int blockstore_impl_t::ack_sync(blockstore_op_t *op)
-{
-    if (PRIV(op)->op_state == SYNC_DONE && PRIV(op)->prev_sync_count == 0)
-    {
-        // Remove dependency of subsequent syncs
-        auto it = PRIV(op)->in_progress_ptr;
-        int done_syncs = 1;
-        ++it;
-        // Acknowledge sync
-        ack_one_sync(op);
-        while (it != in_progress_syncs.end())
-        {
-            auto & next_sync = *it++;
-            PRIV(next_sync)->prev_sync_count -= done_syncs;
-            if (PRIV(next_sync)->prev_sync_count == 0 && PRIV(next_sync)->op_state == SYNC_DONE)
-            {
-                done_syncs++;
-                // Acknowledge next_sync
-                ack_one_sync(next_sync);
-            }
-        }
-        return 2;
-    }
-    return 0;
-}
-
-void blockstore_impl_t::ack_one_sync(blockstore_op_t *op)
+void blockstore_impl_t::ack_sync(blockstore_op_t *op)
 {
     // Handle states
     for (auto it = PRIV(op)->sync_big_writes.begin(); it != PRIV(op)->sync_big_writes.end(); it++)
@@ -299,7 +242,6 @@ void blockstore_impl_t::ack_one_sync(blockstore_op_t *op)
             }
         }
     }
-    in_progress_syncs.erase(PRIV(op)->in_progress_ptr);
     op->retval = 0;
     FINISH_OP(op);
 }

src/blockstore_write.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "blockstore_impl.h"
 
@@ -7,7 +7,7 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
 {
     // Check or assign version number
     bool found = false, deleted = false, is_del = (op->opcode == BS_OP_DELETE);
-    bool is_inflight_big = false;
+    bool wait_big = false, wait_del = false;
     uint64_t version = 1;
     if (dirty_db.size() > 0)
     {
@@ -21,7 +21,8 @@ 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 & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE
+            wait_del = ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_DEL);
+            wait_big = (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE
                 ? !IS_SYNCED(dirty_it->second.state)
                 : ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG);
         }
@@ -38,23 +39,43 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
             deleted = true;
         }
     }
-    if (op->version == 0)
-    {
-        op->version = version;
-    }
-    else if (op->version < version)
-    {
-        // Invalid version requested
-        op->retval = -EEXIST;
-        return false;
-    }
     if (deleted && is_del)
     {
         // Already deleted
         op->retval = 0;
         return false;
     }
-    if (is_inflight_big && !is_del && !deleted && op->len < block_size &&
+    PRIV(op)->real_version = 0;
+    if (op->version == 0)
+    {
+        op->version = version;
+    }
+    else if (op->version < version)
+    {
+        // Implicit operations must be added like that: DEL [FLUSH] BIG [SYNC] SMALL SMALL
+        if (deleted || wait_del)
+        {
+            // It's allowed to write versions with low numbers over deletes
+            // However, we have to flush those deletes first as we use version number for ordering
+#ifdef BLOCKSTORE_DEBUG
+            printf("Write %lx:%lx v%lu over delete (real v%lu) offset=%u len=%u\n", op->oid.inode, op->oid.stripe, version, op->version, op->offset, op->len);
+#endif
+            wait_del = true;
+            PRIV(op)->real_version = op->version;
+            op->version = version;
+            flusher->unshift_flush((obj_ver_id){
+                .oid = op->oid,
+                .version = version-1,
+            }, true);
+        }
+        else
+        {
+            // Invalid version requested
+            op->retval = -EEXIST;
+            return false;
+        }
+    }
+    if (wait_big && !is_del && !deleted && op->len < block_size &&
         immediate_commit != IMMEDIATE_ALL)
     {
         // Issue an additional sync so that the previous big write can reach the journal
@@ -69,22 +90,31 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
 #ifdef BLOCKSTORE_DEBUG
     if (is_del)
         printf("Delete %lx:%lx v%lu\n", op->oid.inode, op->oid.stripe, op->version);
-    else
+    else if (!wait_del)
         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
+    // FIXME No strict need to add it into dirty_db here, it's just left
     // from the previous implementation where reads waited for writes
+    uint32_t state;
+    if (is_del)
+        state = BS_ST_DELETE | BS_ST_IN_FLIGHT;
+    else
+    {
+        state = (op->len == block_size || deleted ? BS_ST_BIG_WRITE : BS_ST_SMALL_WRITE);
+        if (wait_del)
+            state |= BS_ST_WAIT_DEL;
+        else if (state == BS_ST_SMALL_WRITE && wait_big)
+            state |= BS_ST_WAIT_BIG;
+        else
+            state |= BS_ST_IN_FLIGHT;
+        if (op->opcode == BS_OP_WRITE_STABLE)
+            state |= BS_ST_INSTANT;
+    }
     dirty_db.emplace((obj_ver_id){
         .oid = op->oid,
         .version = op->version,
     }, (dirty_entry){
-        .state = (uint32_t)(
-            is_del
-                ? (BS_ST_DELETE | BS_ST_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)))
-        ),
+        .state = state,
         .flags = 0,
         .location = 0,
         .offset = is_del ? 0 : op->offset,
@@ -94,6 +124,29 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
     return true;
 }
 
+void blockstore_impl_t::cancel_all_writes(blockstore_op_t *op, blockstore_dirty_db_t::iterator dirty_it, int retval)
+{
+    while (dirty_it != dirty_db.end() && dirty_it->first.oid == op->oid)
+    {
+        dirty_db.erase(dirty_it++);
+    }
+    bool found = false;
+    for (auto other_op: submit_queue)
+    {
+        if (!found && other_op == op)
+            found = true;
+        else if (found && other_op->oid == op->oid &&
+            (other_op->opcode == BS_OP_WRITE || other_op->opcode == BS_OP_WRITE_STABLE))
+        {
+            // Mark operations to cancel them
+            PRIV(other_op)->real_version = UINT64_MAX;
+            other_op->retval = retval;
+        }
+    }
+    op->retval = retval;
+    FINISH_OP(op);
+}
+
 // First step of the write algorithm: dequeue operation and submit initial write(s)
 int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
 {
@@ -106,12 +159,46 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         .version = op->version,
     });
     assert(dirty_it != dirty_db.end());
-    if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG)
+    if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) < BS_ST_IN_FLIGHT)
     {
         // Don't dequeue
         return 0;
     }
-    else if ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE)
+    if (PRIV(op)->real_version != 0)
+    {
+        if (PRIV(op)->real_version == UINT64_MAX)
+        {
+            // This is the flag value used to cancel operations
+            FINISH_OP(op);
+            return 2;
+        }
+        // Restore original low version number for unblocked operations
+#ifdef BLOCKSTORE_DEBUG
+        printf("Restoring %lx:%lx version: v%lu -> v%lu\n", op->oid.inode, op->oid.stripe, op->version, PRIV(op)->real_version);
+#endif
+        auto prev_it = dirty_it;
+        prev_it--;
+        if (prev_it->first.oid == op->oid && prev_it->first.version >= PRIV(op)->real_version)
+        {
+            // Original version is still invalid
+            // All subsequent writes to the same object must be canceled too
+            cancel_all_writes(op, dirty_it, -EEXIST);
+            return 2;
+        }
+        op->version = PRIV(op)->real_version;
+        PRIV(op)->real_version = 0;
+        dirty_entry e = dirty_it->second;
+        dirty_db.erase(dirty_it);
+        dirty_it = dirty_db.emplace((obj_ver_id){
+            .oid = op->oid,
+            .version = op->version,
+        }, e).first;
+    }
+    if (write_iodepth >= max_write_iodepth)
+    {
+        return 0;
+    }
+    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))
@@ -129,10 +216,10 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
                 PRIV(op)->wait_for = WAIT_FREE;
                 return 0;
             }
-            op->retval = -ENOSPC;
-            FINISH_OP(op);
-            return 1;
+            cancel_all_writes(op, dirty_it, -ENOSPC);
+            return 2;
         }
+        write_iodepth++;
         BS_SUBMIT_GET_SQE(sqe, data);
         dirty_it->second.location = loc << block_order;
         dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SUBMITTED;
@@ -185,6 +272,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         {
             return 0;
         }
+        write_iodepth++;
         // There is sufficient space. Get SQE(s)
         struct io_uring_sqe *sqe1 = NULL;
         if (immediate_commit != IMMEDIATE_NONE ||
@@ -282,14 +370,13 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         if (!PRIV(op)->pending_ops)
         {
             PRIV(op)->op_state = 4;
-            continue_write(op);
+            return continue_write(op);
         }
         else
         {
             PRIV(op)->op_state = 3;
         }
     }
-    inflight_writes++;
     return 1;
 }
 
@@ -297,30 +384,29 @@ int blockstore_impl_t::continue_write(blockstore_op_t *op)
 {
     io_uring_sqe *sqe = NULL;
     journal_entry_big_write *je;
+    int op_state = PRIV(op)->op_state;
+    if (op_state != 2 && op_state != 4)
+    {
+        // In progress
+        return 1;
+    }
     auto dirty_it = dirty_db.find((obj_ver_id){
         .oid = op->oid,
         .version = op->version,
     });
     assert(dirty_it != dirty_db.end());
-    if (PRIV(op)->op_state == 2)
+    if (op_state == 2)
         goto resume_2;
-    else if (PRIV(op)->op_state == 4)
+    else if (op_state == 4)
         goto resume_4;
-    else
-        return 1;
 resume_2:
     // Only for the immediate_commit mode: prepare and submit big_write journal entry
-    sqe = get_sqe();
-    if (!sqe)
-    {
-        return 0;
-    }
+    BS_SUBMIT_GET_SQE_DECL(sqe);
     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(
@@ -345,7 +431,7 @@ resume_2:
 resume_4:
     // Switch object state
 #ifdef BLOCKSTORE_DEBUG
-    printf("Ack write %lx:%lx v%lu = %d\n", op->oid.inode, op->oid.stripe, op->version, dirty_it->second.state);
+    printf("Ack write %lx:%lx v%lu = state %x\n", op->oid.inode, op->oid.stripe, op->version, dirty_it->second.state);
 #endif
     bool imm = (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE
         ? (immediate_commit == IMMEDIATE_ALL)
@@ -374,11 +460,11 @@ resume_4:
             dirty_it++;
         }
     }
-    inflight_writes--;
     // Acknowledge write
     op->retval = op->len;
+    write_iodepth--;
     FINISH_OP(op);
-    return 1;
+    return 2;
 }
 
 void blockstore_impl_t::handle_write_event(ring_data_t *data, blockstore_op_t *op)
@@ -386,7 +472,6 @@ void blockstore_impl_t::handle_write_event(ring_data_t *data, blockstore_op_t *o
     live = true;
     if (data->res != data->iov.iov_len)
     {
-        inflight_writes--;
         // FIXME: our state becomes corrupted after a write error. maybe do something better than just die
         throw std::runtime_error(
             "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
@@ -398,10 +483,7 @@ void blockstore_impl_t::handle_write_event(ring_data_t *data, blockstore_op_t *o
     {
         release_journal_sectors(op);
         PRIV(op)->op_state++;
-        if (!continue_write(op))
-        {
-            submit_queue.push_front(op);
-        }
+        ringloop->wakeup();
     }
 }
 
@@ -414,8 +496,8 @@ void blockstore_impl_t::release_journal_sectors(blockstore_op_t *op)
         uint64_t s = PRIV(op)->min_flushed_journal_sector;
         while (1)
         {
-            journal.sector_info[s-1].usage_count--;
-            if (s != (1+journal.cur_sector) && journal.sector_info[s-1].usage_count == 0)
+            journal.sector_info[s-1].flush_count--;
+            if (s != (1+journal.cur_sector) && journal.sector_info[s-1].flush_count == 0)
             {
                 // 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;
@@ -439,16 +521,21 @@ void blockstore_impl_t::release_journal_sectors(blockstore_op_t *op)
 
 int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
 {
+    if (PRIV(op)->op_state)
+    {
+        return continue_write(op);
+    }
     auto dirty_it = dirty_db.find((obj_ver_id){
         .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))
+    if (!space_check.check_available(op, 1, sizeof(journal_entry_del), JOURNAL_STABILIZE_RESERVATION))
     {
         return 0;
     }
+    write_iodepth++;
     io_uring_sqe *sqe = NULL;
     if (immediate_commit != IMMEDIATE_NONE ||
         (journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_del) &&
@@ -495,7 +582,10 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
         prepare_journal_sector_write(journal, journal.cur_sector, sqe, cb);
         PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
         PRIV(op)->pending_ops++;
-        // Remember small write as unsynced
+    }
+    else
+    {
+        // Remember delete as unsynced
         unsynced_small_writes.push_back((obj_ver_id){
             .oid = op->oid,
             .version = op->version,
@@ -504,7 +594,7 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
     if (!PRIV(op)->pending_ops)
     {
         PRIV(op)->op_state = 4;
-        continue_write(op);
+        return continue_write(op);
     }
     else
     {

src/cluster_client.cpp

@@ -1,12 +1,14 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
+#include <stdexcept>
 #include "cluster_client.h"
 
 cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json & config)
 {
     this->ringloop = ringloop;
     this->tfd = tfd;
+    this->config = config;
 
     msgr.osd_num = 0;
     msgr.tfd = tfd;
@@ -63,8 +65,7 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
         msgr.stop_client(op->peer_fd);
         delete op;
     };
-    msgr.use_sync_send_recv = config["use_sync_send_recv"].bool_value() ||
-        config["use_sync_send_recv"].uint64_value();
+    msgr.init();
 
     st_cli.tfd = tfd;
     st_cli.on_load_config_hook = [this](json11::Json::object & cfg) { on_load_config_hook(cfg); };
@@ -72,7 +73,6 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
     st_cli.on_change_hook = [this](json11::Json::object & changes) { on_change_hook(changes); };
     st_cli.on_load_pgs_hook = [this](bool success) { on_load_pgs_hook(success); };
 
-    log_level = config["log_level"].int64_value();
     st_cli.parse_config(config);
     st_cli.load_global_config();
 
@@ -182,17 +182,8 @@ void cluster_client_t::on_load_config_hook(json11::Json::object & config)
     {
         up_wait_retry_interval = 50;
     }
-    msgr.peer_connect_interval = config["peer_connect_interval"].uint64_value();
-    if (!msgr.peer_connect_interval)
-    {
-        msgr.peer_connect_interval = DEFAULT_PEER_CONNECT_INTERVAL;
-    }
-    msgr.peer_connect_timeout = config["peer_connect_timeout"].uint64_value();
-    if (!msgr.peer_connect_timeout)
-    {
-        msgr.peer_connect_timeout = DEFAULT_PEER_CONNECT_TIMEOUT;
-    }
-    st_cli.start_etcd_watcher();
+    msgr.parse_config(config);
+    msgr.parse_config(this->config);
     st_cli.load_pgs();
 }
 
@@ -202,6 +193,12 @@ void cluster_client_t::on_load_pgs_hook(bool success)
     {
         pg_counts[pool_item.first] = pool_item.second.real_pg_count;
     }
+    pgs_loaded = true;
+    for (auto fn: on_ready_hooks)
+    {
+        fn();
+    }
+    on_ready_hooks.clear();
     for (auto op: offline_ops)
     {
         execute(op);
@@ -253,6 +250,18 @@ void cluster_client_t::on_change_osd_state_hook(uint64_t peer_osd)
     }
 }
 
+void cluster_client_t::on_ready(std::function<void(void)> fn)
+{
+    if (pgs_loaded)
+    {
+        fn();
+    }
+    else
+    {
+        on_ready_hooks.push_back(fn);
+    }
+}
+
 /**
  * How writes are synced when immediate_commit is false
  *
@@ -283,7 +292,7 @@ void cluster_client_t::on_change_osd_state_hook(uint64_t peer_osd)
 
 void cluster_client_t::execute(cluster_op_t *op)
 {
-    if (!bs_disk_alignment)
+    if (!pgs_loaded)
     {
         // We're offline
         offline_ops.push_back(op);
@@ -453,7 +462,7 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
     // Primary OSDs still operate individual stripes, but their size is multiplied by PG minsize in case of EC
     auto & pool_cfg = st_cli.pool_config[INODE_POOL(op->inode)];
     uint64_t pg_block_size = bs_block_size * (
-        pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_minsize
+        pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks
     );
     uint64_t first_stripe = (op->offset / pg_block_size) * pg_block_size;
     uint64_t last_stripe = ((op->offset + op->len + pg_block_size - 1) / pg_block_size - 1) * pg_block_size;
@@ -468,7 +477,7 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
         uint64_t begin = (op->offset < stripe ? stripe : op->offset);
         uint64_t end = (op->offset + op->len) > (stripe + pg_block_size)
             ? (stripe + pg_block_size) : (op->offset + op->len);
-        op->parts[i] = {
+        op->parts[i] = (cluster_op_part_t){
             .parent = op,
             .offset = begin,
             .len = (uint32_t)(end - begin),
@@ -513,7 +522,7 @@ bool cluster_client_t::try_send(cluster_op_t *op, cluster_op_part_t *part)
             part->osd_num = primary_osd;
             part->sent = true;
             op->sent_count++;
-            part->op = {
+            part->op = (osd_op_t){
                 .op_type = OSD_OP_OUT,
                 .peer_fd = peer_fd,
                 .req = { .rw = {
@@ -674,7 +683,7 @@ void cluster_client_t::send_sync(cluster_op_t *op, cluster_op_part_t *part)
     assert(peer_it != msgr.osd_peer_fds.end());
     part->sent = true;
     op->sent_count++;
-    part->op = {
+    part->op = (osd_op_t){
         .op_type = OSD_OP_OUT,
         .peer_fd = peer_it->second,
         .req = {
@@ -737,7 +746,7 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
             assert(op == cur_sync);
             finish_sync();
         }
-        else
+        else if (!op->up_wait)
         {
             continue_rw(op);
         }

src/cluster_client.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #pragma once
 
@@ -8,7 +8,6 @@
 
 #define MIN_BLOCK_SIZE 4*1024
 #define MAX_BLOCK_SIZE 128*1024*1024
-#define DEFAULT_BLOCK_SIZE 128*1024
 #define DEFAULT_DISK_ALIGNMENT 4096
 #define DEFAULT_BITMAP_GRANULARITY 4096
 #define DEFAULT_CLIENT_DIRTY_LIMIT 32*1024*1024
@@ -64,8 +63,6 @@ class cluster_client_t
     int up_wait_retry_interval = 500; // ms
 
     uint64_t op_id = 1;
-    etcd_state_client_t st_cli;
-    osd_messenger_t msgr;
     ring_consumer_t consumer;
     // operations currently in progress
     std::set<cluster_op_t*> cur_ops;
@@ -79,10 +76,18 @@ class cluster_client_t
     std::vector<cluster_op_t*> offline_ops;
     uint64_t queued_bytes = 0;
 
+    bool pgs_loaded = false;
+    std::vector<std::function<void(void)>> on_ready_hooks;
+
 public:
+    etcd_state_client_t st_cli;
+    osd_messenger_t msgr;
+    json11::Json config;
+
     cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json & config);
     ~cluster_client_t();
     void execute(cluster_op_t *op);
+    void on_ready(std::function<void(void)> fn);
     void stop();
 
 protected:

src/crc32c.h

@@ -8,4 +8,10 @@
 // unsigned __int64 _mm_crc32_u64 (unsigned __int64 crc, unsigned __int64 v)
 // unsigned int _mm_crc32_u8 (unsigned int crc, unsigned char v)
 
+#ifdef __cplusplus
+extern "C" {
+#endif
 uint32_t crc32c(uint32_t crc, const void *buf, size_t len);
+#ifdef __cplusplus
+};
+#endif

src/dump_journal.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #define _LARGEFILE64_SOURCE
 #include <sys/types.h>
@@ -26,23 +26,32 @@ struct journal_dump_t
     uint64_t journal_offset;
     uint64_t journal_len;
     uint64_t journal_pos;
+    bool all;
+    bool started;
     int fd;
+    uint32_t crc32_last;
 
-    void dump_block(void *buf);
+    int dump_block(void *buf);
 };
 
 int main(int argc, char *argv[])
 {
-    if (argc < 5)
+    journal_dump_t self = { 0 };
+    int b = 1;
+    if (argc >= 2 && !strcmp(argv[1], "--all"))
     {
-        printf("USAGE: %s <journal_file> <journal_block_size> <offset> <size>\n", argv[0]);
+        self.all = true;
+        b = 2;
+    }
+    if (argc < b+4)
+    {
+        printf("USAGE: %s [--all] <journal_file> <journal_block_size> <offset> <size>\n", argv[0]);
         return 1;
     }
-    journal_dump_t self;
-    self.journal_device = argv[1];
-    self.journal_block = strtoul(argv[2], NULL, 10);
-    self.journal_offset = strtoull(argv[3], NULL, 10);
-    self.journal_len = strtoull(argv[4], NULL, 10);
+    self.journal_device = argv[b];
+    self.journal_block = strtoul(argv[b+1], NULL, 10);
+    self.journal_offset = strtoull(argv[b+2], NULL, 10);
+    self.journal_len = strtoull(argv[b+3], NULL, 10);
     if (self.journal_block < MEM_ALIGNMENT || (self.journal_block % MEM_ALIGNMENT) ||
         self.journal_block > 128*1024)
     {
@@ -57,30 +66,64 @@ int main(int argc, char *argv[])
     }
     void *data = memalign(MEM_ALIGNMENT, self.journal_block);
     self.journal_pos = 0;
-    while (self.journal_pos < self.journal_len)
+    if (self.all)
+    {
+        while (self.journal_pos < self.journal_len)
+        {
+            int r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
+            assert(r == self.journal_block);
+            uint64_t s;
+            for (s = 0; s < self.journal_block; s += 8)
+            {
+                if (*((uint64_t*)(data+s)) != 0)
+                    break;
+            }
+            if (s == self.journal_block)
+            {
+                printf("offset %08lx: zeroes\n", self.journal_pos);
+                self.journal_pos += self.journal_block;
+            }
+            else if (((journal_entry*)data)->magic == JOURNAL_MAGIC)
+            {
+                printf("offset %08lx:\n", self.journal_pos);
+                self.dump_block(data);
+            }
+            else
+            {
+                printf("offset %08lx: no magic in the beginning, looks like random data (pattern=%lx)\n", self.journal_pos, *((uint64_t*)data));
+                self.journal_pos += self.journal_block;
+            }
+        }
+    }
+    else
     {
         int r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
         assert(r == self.journal_block);
-        uint64_t s;
-        for (s = 0; s < self.journal_block; s += 8)
+        journal_entry *je = (journal_entry*)(data);
+        if (je->magic != JOURNAL_MAGIC || je->type != JE_START || je_crc32(je) != je->crc32)
         {
-            if (*((uint64_t*)(data+s)) != 0)
-                break;
-        }
-        if (s == self.journal_block)
-        {
-            printf("offset %08lx: zeroes\n", self.journal_pos);
-            self.journal_pos += self.journal_block;
-        }
-        else if (((journal_entry*)data)->magic == JOURNAL_MAGIC)
-        {
-            printf("offset %08lx:\n", self.journal_pos);
-            self.dump_block(data);
+            printf("offset %08lx: journal superblock is invalid\n", self.journal_pos);
         }
         else
         {
-            printf("offset %08lx: no magic in the beginning, looks like random data (pattern=%lx)\n", self.journal_pos, *((uint64_t*)data));
-            self.journal_pos += self.journal_block;
+            printf("offset %08lx:\n", self.journal_pos);
+            self.dump_block(data);
+            self.started = false;
+            self.journal_pos = je->start.journal_start;
+            while (1)
+            {
+                if (self.journal_pos >= self.journal_len)
+                    self.journal_pos = self.journal_block;
+                r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
+                assert(r == self.journal_block);
+                printf("offset %08lx:\n", self.journal_pos);
+                r = self.dump_block(data);
+                if (r <= 0)
+                {
+                    printf("end of the journal\n");
+                    break;
+                }
+            }
         }
     }
     free(data);
@@ -88,7 +131,7 @@ int main(int argc, char *argv[])
     return 0;
 }
 
-void journal_dump_t::dump_block(void *buf)
+int journal_dump_t::dump_block(void *buf)
 {
     uint32_t pos = 0;
     journal_pos += journal_block;
@@ -97,12 +140,19 @@ 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_MIN || je->type > JE_MAX)
+        if (je->magic != JOURNAL_MAGIC || je->type < JE_MIN || je->type > JE_MAX ||
+            !all && started && je->crc32_prev != crc32_last)
         {
             break;
         }
-        const char *crc32_valid = je_crc32(je) == je->crc32 ? "(valid)" : "(invalid)";
-        printf("entry % 3d: crc32=%08x %s prev=%08x ", entry, je->crc32, crc32_valid, je->crc32_prev);
+        bool crc32_valid = je_crc32(je) == je->crc32;
+        if (!all && !crc32_valid)
+        {
+            break;
+        }
+        started = true;
+        crc32_last = je->crc32;
+        printf("entry % 3d: crc32=%08x %s prev=%08x ", entry, je->crc32, (crc32_valid ? "(valid)" : "(invalid)"), je->crc32_prev);
         if (je->type == JE_START)
         {
             printf("je_start start=%08lx\n", je->start.journal_start);
@@ -170,4 +220,5 @@ void journal_dump_t::dump_block(void *buf)
     {
         journal_pos = journal_len;
     }
+    return entry;
 }

src/epoll_manager.cpp

@@ -1,9 +1,10 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #include <sys/epoll.h>
 #include <sys/poll.h>
 #include <unistd.h>
+#include <stdexcept>
 
 #include "epoll_manager.h"
 
@@ -83,8 +84,12 @@ void epoll_manager_t::handle_epoll_events()
         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);
+            auto cb_it = epoll_handlers.find(events[i].data.fd);
+            if (cb_it != epoll_handlers.end())
+            {
+                auto & cb = cb_it->second;
+                cb(events[i].data.fd, events[i].events);
+            }
         }
     } while (nfds == MAX_EPOLL_EVENTS);
 }

src/epoll_manager.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #pragma once
 

src/etcd_state_client.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #include "osd_ops.h"
 #include "pg_states.h"
@@ -7,6 +7,16 @@
 #include "http_client.h"
 #include "base64.h"
 
+etcd_state_client_t::~etcd_state_client_t()
+{
+    etcd_watches_initialised = -1;
+    if (etcd_watch_ws)
+    {
+        etcd_watch_ws->close();
+        etcd_watch_ws = NULL;
+    }
+}
+
 json_kv_t etcd_state_client_t::parse_etcd_kv(const json11::Json & kv_json)
 {
     json_kv_t kv;
@@ -46,6 +56,23 @@ 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::add_etcd_url(std::string addr)
+{
+    if (addr.length() > 0)
+    {
+        if (strtolower(addr.substr(0, 7)) == "http://")
+            addr = addr.substr(7);
+        else if (strtolower(addr.substr(0, 8)) == "https://")
+        {
+            printf("HTTPS is unsupported for etcd. Either use plain HTTP or setup a local proxy for etcd interaction\n");
+            exit(1);
+        }
+        if (addr.find('/') < 0)
+            addr += "/v3";
+        this->etcd_addresses.push_back(addr);
+    }
+}
+
 void etcd_state_client_t::parse_config(json11::Json & config)
 {
     this->etcd_addresses.clear();
@@ -55,13 +82,7 @@ void etcd_state_client_t::parse_config(json11::Json & config)
         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);
-            }
+            add_etcd_url(pos >= 0 ? ea.substr(0, pos) : ea);
             if (pos >= 0)
                 ea = ea.substr(pos+1);
             else
@@ -72,13 +93,7 @@ void etcd_state_client_t::parse_config(json11::Json & config)
     {
         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);
-            }
+            add_etcd_url(ea.string_value());
         }
     }
     this->etcd_prefix = config["etcd_prefix"].string_value();
@@ -136,7 +151,7 @@ void etcd_state_client_t::start_etcd_watcher()
                 }
                 for (auto & kv: changes)
                 {
-                    if (this->log_level > 0)
+                    if (this->log_level > 3)
                     {
                         printf("Incoming event: %s -> %s\n", kv.first.c_str(), kv.second.dump().c_str());
                     }
@@ -160,7 +175,7 @@ void etcd_state_client_t::start_etcd_watcher()
                     start_etcd_watcher();
                 });
             }
-            else
+            else if (etcd_watches_initialised > 0)
             {
                 // Connection was live, retry immediately
                 start_etcd_watcher();
@@ -173,6 +188,7 @@ void etcd_state_client_t::start_etcd_watcher()
             { "range_end", base64_encode(etcd_prefix+"/config0") },
             { "start_revision", etcd_watch_revision+1 },
             { "watch_id", ETCD_CONFIG_WATCH_ID },
+            { "progress_notify", true },
         } }
     }).dump());
     etcd_watch_ws->post_message(WS_TEXT, json11::Json(json11::Json::object {
@@ -181,6 +197,7 @@ void etcd_state_client_t::start_etcd_watcher()
             { "range_end", base64_encode(etcd_prefix+"/osd/state0") },
             { "start_revision", etcd_watch_revision+1 },
             { "watch_id", ETCD_OSD_STATE_WATCH_ID },
+            { "progress_notify", true },
         } }
     }).dump());
     etcd_watch_ws->post_message(WS_TEXT, json11::Json(json11::Json::object {
@@ -189,6 +206,7 @@ void etcd_state_client_t::start_etcd_watcher()
             { "range_end", base64_encode(etcd_prefix+"/pg/state0") },
             { "start_revision", etcd_watch_revision+1 },
             { "watch_id", ETCD_PG_STATE_WATCH_ID },
+            { "progress_notify", true },
         } }
     }).dump());
     etcd_watch_ws->post_message(WS_TEXT, json11::Json(json11::Json::object {
@@ -197,6 +215,7 @@ void etcd_state_client_t::start_etcd_watcher()
             { "range_end", base64_encode(etcd_prefix+"/pg/history0") },
             { "start_revision", etcd_watch_revision+1 },
             { "watch_id", ETCD_PG_HISTORY_WATCH_ID },
+            { "progress_notify", true },
         } }
     }).dump());
 }
@@ -216,10 +235,6 @@ void etcd_state_client_t::load_global_config()
             });
             return;
         }
-        if (!etcd_watch_revision)
-        {
-            etcd_watch_revision = data["header"]["revision"].uint64_value();
-        }
         json11::Json::object global_config;
         if (data["kvs"].array_items().size() > 0)
         {
@@ -229,6 +244,11 @@ void etcd_state_client_t::load_global_config()
                 global_config = kv.value.object_items();
             }
         }
+        bs_block_size = global_config["block_size"].uint64_value();
+        if (!bs_block_size)
+        {
+            bs_block_size = DEFAULT_BLOCK_SIZE;
+        }
         on_load_config_hook(global_config);
     });
 }
@@ -287,6 +307,10 @@ void etcd_state_client_t::load_pgs()
             on_load_pgs_hook(false);
             return;
         }
+        if (!etcd_watch_revision)
+        {
+            etcd_watch_revision = data["header"]["revision"].uint64_value();
+        }
         for (auto & res: data["responses"].array_items())
         {
             for (auto & kv_json: res["response_range"]["kvs"].array_items())
@@ -296,6 +320,7 @@ void etcd_state_client_t::load_pgs()
             }
         }
         on_load_pgs_hook(true);
+        start_etcd_watcher();
     });
 }
 
@@ -309,65 +334,99 @@ void etcd_state_client_t::parse_state(const std::string & key, const json11::Jso
         }
         for (auto & pool_item: value.object_items())
         {
+            pool_config_t pc;
+            // ID
             pool_id_t pool_id = stoull_full(pool_item.first);
             if (!pool_id || pool_id >= POOL_ID_MAX)
             {
                 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;
             }
-            if (pool_item.second["pg_size"].uint64_value() < 1 ||
-                pool_item.second["scheme"] == "xor" && pool_item.second["pg_size"].uint64_value() < 3)
-            {
-                printf("Pool %u has invalid pg_size, skipping pool\n", pool_id);
-                continue;
-            }
-            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("Pool %u has invalid pg_minsize, skipping pool\n", pool_id);
-                continue;
-            }
-            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() == "")
+            pc.id = pool_id;
+            // Pool Name
+            pc.name = pool_item.second["name"].string_value();
+            if (pc.name == "")
             {
                 printf("Pool %u has empty name, skipping pool\n", pool_id);
                 continue;
             }
-            if (pool_item.second["scheme"] != "replicated" && pool_item.second["scheme"] != "xor")
+            // Failure Domain
+            pc.failure_domain = pool_item.second["failure_domain"].string_value();
+            // Coding Scheme
+            if (pool_item.second["scheme"] == "replicated")
+                pc.scheme = POOL_SCHEME_REPLICATED;
+            else if (pool_item.second["scheme"] == "xor")
+                pc.scheme = POOL_SCHEME_XOR;
+            else if (pool_item.second["scheme"] == "jerasure")
+                pc.scheme = POOL_SCHEME_JERASURE;
+            else
             {
-                printf("Pool %u has invalid coding scheme (only \"xor\" and \"replicated\" are allowed), skipping pool\n", pool_id);
+                printf("Pool %u has invalid coding scheme (one of \"xor\", \"replicated\" or \"jerasure\" required), 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)
+            // PG Size
+            pc.pg_size = pool_item.second["pg_size"].uint64_value();
+            if (pc.pg_size < 1 ||
+                pool_item.second["pg_size"].uint64_value() < 3 &&
+                (pc.scheme == POOL_SCHEME_XOR || pc.scheme == POOL_SCHEME_JERASURE) ||
+                pool_item.second["pg_size"].uint64_value() > 256)
+            {
+                printf("Pool %u has invalid pg_size, skipping pool\n", pool_id);
+                continue;
+            }
+            // Parity Chunks
+            pc.parity_chunks = pool_item.second["parity_chunks"].uint64_value();
+            if (pc.scheme == POOL_SCHEME_XOR)
+            {
+                if (pc.parity_chunks > 1)
+                {
+                    printf("Pool %u has invalid parity_chunks (must be 1), skipping pool\n", pool_id);
+                    continue;
+                }
+                pc.parity_chunks = 1;
+            }
+            if (pc.scheme == POOL_SCHEME_JERASURE &&
+                (pc.parity_chunks < 1 || pc.parity_chunks > pc.pg_size-2))
+            {
+                printf("Pool %u has invalid parity_chunks (must be between 1 and pg_size-2), skipping pool\n", pool_id);
+                continue;
+            }
+            // PG MinSize
+            pc.pg_minsize = pool_item.second["pg_minsize"].uint64_value();
+            if (pc.pg_minsize < 1 || pc.pg_minsize > pc.pg_size ||
+                (pc.scheme == POOL_SCHEME_XOR || pc.scheme == POOL_SCHEME_JERASURE) &&
+                pc.pg_minsize < (pc.pg_size-pc.parity_chunks))
+            {
+                printf("Pool %u has invalid pg_minsize, skipping pool\n", pool_id);
+                continue;
+            }
+            // PG Count
+            pc.pg_count = pool_item.second["pg_count"].uint64_value();
+            if (pc.pg_count < 1)
+            {
+                printf("Pool %u has invalid pg_count, skipping pool\n", pool_id);
+                continue;
+            }
+            // Max OSD Combinations
+            pc.max_osd_combinations = pool_item.second["max_osd_combinations"].uint64_value();
+            if (!pc.max_osd_combinations)
+                pc.max_osd_combinations = 10000;
+            if (pc.max_osd_combinations > 0 && pc.max_osd_combinations < 100)
             {
                 printf("Pool %u has invalid max_osd_combinations (must be at least 100), skipping pool\n", pool_id);
                 continue;
             }
+            // PG Stripe Size
+            pc.pg_stripe_size = pool_item.second["pg_stripe_size"].uint64_value();
+            uint64_t min_stripe_size = bs_block_size * (pc.scheme == POOL_SCHEME_REPLICATED ? 1 : (pc.pg_size-pc.parity_chunks));
+            if (pc.pg_stripe_size < min_stripe_size)
+                pc.pg_stripe_size = min_stripe_size;
+            // Save
+            pc.real_pg_count = this->pool_config[pool_id].real_pg_count;
+            std::swap(pc.pg_config, this->pool_config[pool_id].pg_config);
+            std::swap(this->pool_config[pool_id], pc);
             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)

src/etcd_state_client.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #pragma once
 
@@ -16,7 +16,7 @@
 #define ETCD_SLOW_TIMEOUT 5000
 #define ETCD_QUICK_TIMEOUT 1000
 
-#define DEFAULT_PG_STRIPE_SIZE 4*1024*1024
+#define DEFAULT_BLOCK_SIZE 128*1024
 
 struct json_kv_t
 {
@@ -43,7 +43,7 @@ struct pool_config_t
     pool_id_t id;
     std::string name;
     uint64_t scheme;
-    uint64_t pg_size, pg_minsize;
+    uint64_t pg_size, pg_minsize, parity_chunks;
     uint64_t pg_count;
     uint64_t real_pg_count;
     std::string failure_domain;
@@ -54,6 +54,9 @@ struct pool_config_t
 
 struct etcd_state_client_t
 {
+protected:
+    void add_etcd_url(std::string);
+public:
     std::vector<std::string> etcd_addresses;
     std::string etcd_prefix;
     int log_level = 0;
@@ -62,6 +65,7 @@ struct etcd_state_client_t
     int etcd_watches_initialised = 0;
     uint64_t etcd_watch_revision = 0;
     websocket_t *etcd_watch_ws = NULL;
+    uint64_t bs_block_size = 0;
     std::map<pool_id_t, pool_config_t> pool_config;
     std::map<osd_num_t, json11::Json> peer_states;
 
@@ -80,4 +84,5 @@ struct etcd_state_client_t
     void load_pgs();
     void parse_state(const std::string & key, const json11::Json & value);
     void parse_config(json11::Json & config);
+    ~etcd_state_client_t();
 };

src/fio_cluster.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 // FIO engine to test cluster I/O
 //
@@ -28,12 +28,7 @@
 
 #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"
-}
+#include "fio_headers.h"
 
 struct sec_data
 {
@@ -98,7 +93,7 @@ static struct fio_option options[] = {
     {
         .name   = "cluster_log_level",
         .lname  = "cluster log level",
-        .type   = FIO_OPT_BOOL,
+        .type   = FIO_OPT_INT,
         .off1   = offsetof(struct sec_options, cluster_log),
         .help   = "Set log level for the Vitastor client",
         .def    = "0",
@@ -122,8 +117,15 @@ static struct fio_option options[] = {
 
 static int sec_setup(struct thread_data *td)
 {
+    sec_options *o = (sec_options*)td->eo;
     sec_data *bsd;
 
+    if (!o->etcd_host)
+    {
+        td_verror(td, EINVAL, "etcd address is missing");
+        return 1;
+    }
+
     bsd = new sec_data;
     if (!bsd)
     {
@@ -150,9 +152,7 @@ static void sec_cleanup(struct thread_data *td)
         delete bsd->cli;
         delete bsd->epmgr;
         delete bsd->ringloop;
-        bsd->cli = NULL;
-        bsd->epmgr = NULL;
-        bsd->ringloop = NULL;
+        delete bsd;
     }
 }
 

src/fio_engine.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 // FIO engine to test Blockstore
 //
@@ -25,12 +25,7 @@
 //     -bs_config='{"data_device":"./test_data.bin"}' -size=1000M
 
 #include "blockstore.h"
-extern "C" {
-#define CONFIG_HAVE_GETTID
-#define CONFIG_PWRITEV2
-#include "fio/fio.h"
-#include "fio/optgroup.h"
-}
+#include "fio_headers.h"
 
 #include "json11/json11.hpp"
 

src/fio_headers.h

@@ -0,0 +1,16 @@
+extern "C" {
+// Kill atomics in fio headers
+#define _STDATOMIC_H
+#include "fio/arch/arch.h"
+
+#undef atomic_load_acquire
+#undef atomic_store_release
+#define atomic_load_acquire(p) *(p)
+#define atomic_store_release(p, v) (*(p)) = (v)
+
+#define CONFIG_HAVE_GETTID
+#define CONFIG_SYNC_FILE_RANGE
+#define CONFIG_PWRITEV2
+#include "fio/fio.h"
+#include "fio/optgroup.h"
+}

src/fio_sec_osd.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 // FIO engine to test Blockstore through Secondary OSD interface
 //
@@ -30,12 +30,7 @@
 
 #include "rw_blocking.h"
 #include "osd_ops.h"
-extern "C" {
-#define CONFIG_HAVE_GETTID
-#define CONFIG_PWRITEV2
-#include "fio/fio.h"
-#include "fio/optgroup.h"
-}
+#include "fio_headers.h"
 
 struct sec_data
 {
@@ -145,6 +140,7 @@ static void sec_cleanup(struct thread_data *td)
     if (bsd)
     {
         close(bsd->connect_fd);
+        delete bsd;
     }
 }
 
@@ -317,6 +313,7 @@ static int sec_getevents(struct thread_data *td, unsigned int min, unsigned int
             exit(1);
         }
         io_u* io = it->second;
+        bsd->queue.erase(it);
         if (io->ddir == DDIR_READ)
         {
             if (reply.hdr.retval != io->xfer_buflen)

src/http_client.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #include <netinet/tcp.h>
 #include <sys/epoll.h>
@@ -13,6 +13,8 @@
 #include <fcntl.h>
 #include <string.h>
 
+#include <stdexcept>
+
 #include "json11/json11.hpp"
 #include "http_client.h"
 #include "timerfd_manager.h"
@@ -20,7 +22,6 @@
 #define READ_BUFFER_SIZE 9000
 
 static int extract_port(std::string & host);
-static std::string strtolower(const std::string & in);
 static std::string trim(const std::string & in);
 static std::string ws_format_frame(int type, uint64_t size);
 static bool ws_parse_frame(std::string & buf, int & type, std::string & res);
@@ -671,7 +672,7 @@ static int extract_port(std::string & host)
     return port;
 }
 
-static std::string strtolower(const std::string & in)
+std::string strtolower(const std::string & in)
 {
     std::string s = in;
     for (int i = 0; i < s.length(); i++)

src/http_client.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #pragma once
 #include <string>
@@ -49,6 +49,8 @@ std::vector<std::string> getifaddr_list(bool include_v6 = false);
 
 uint64_t stoull_full(const std::string & str, int base = 10);
 
+std::string strtolower(const std::string & in);
+
 void http_request(timerfd_manager_t *tfd, const std::string & host, const std::string & request,
     const http_options_t & options, std::function<void(const http_response_t *response)> callback);
 

src/malloc_or_die.h

@@ -1,9 +1,10 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #pragma once
 
 #include <malloc.h>
+#include <stdlib.h>
 
 inline void* memalign_or_die(size_t alignment, size_t size)
 {

src/messenger.cpp

@@ -1,11 +1,12 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 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 <stdexcept>
 
 #include "messenger.h"
 
@@ -25,14 +26,106 @@ osd_op_t::~osd_op_t()
     }
 }
 
+void osd_messenger_t::init()
+{
+    keepalive_timer_id = tfd->set_timer(1000, true, [this](int)
+    {
+        for (auto cl_it = clients.begin(); cl_it != clients.end();)
+        {
+            auto cl = (cl_it++)->second;
+            if (!cl->osd_num)
+            {
+                // Do not run keepalive on regular clients
+                continue;
+            }
+            if (cl->ping_time_remaining > 0)
+            {
+                cl->ping_time_remaining--;
+                if (!cl->ping_time_remaining)
+                {
+                    // Ping timed out, stop the client
+                    stop_client(cl->peer_fd, true);
+                }
+            }
+            else if (cl->idle_time_remaining > 0)
+            {
+                cl->idle_time_remaining--;
+                if (!cl->idle_time_remaining)
+                {
+                    // Connection is idle for <osd_idle_time>, send ping
+                    osd_op_t *op = new osd_op_t();
+                    op->op_type = OSD_OP_OUT;
+                    op->peer_fd = cl->peer_fd;
+                    op->req = (osd_any_op_t){
+                        .hdr = {
+                            .magic = SECONDARY_OSD_OP_MAGIC,
+                            .id = this->next_subop_id++,
+                            .opcode = OSD_OP_PING,
+                        },
+                    };
+                    op->callback = [this, cl](osd_op_t *op)
+                    {
+                        int fail_fd = (op->reply.hdr.retval != 0 ? op->peer_fd : -1);
+                        cl->ping_time_remaining = 0;
+                        delete op;
+                        if (fail_fd >= 0)
+                        {
+                            stop_client(fail_fd, true);
+                        }
+                    };
+                    outbox_push(op);
+                    cl->ping_time_remaining = osd_ping_timeout;
+                    cl->idle_time_remaining = osd_idle_timeout;
+                }
+            }
+            else
+            {
+                cl->idle_time_remaining = osd_idle_timeout;
+            }
+        }
+    });
+}
+
 osd_messenger_t::~osd_messenger_t()
 {
+    if (keepalive_timer_id >= 0)
+    {
+        tfd->clear_timer(keepalive_timer_id);
+        keepalive_timer_id = -1;
+    }
     while (clients.size() > 0)
     {
-        stop_client(clients.begin()->first);
+        stop_client(clients.begin()->first, true);
     }
 }
 
+void osd_messenger_t::parse_config(const json11::Json & config)
+{
+    this->use_sync_send_recv = config["use_sync_send_recv"].bool_value() ||
+        config["use_sync_send_recv"].uint64_value();
+    this->peer_connect_interval = config["peer_connect_interval"].uint64_value();
+    if (!this->peer_connect_interval)
+    {
+        this->peer_connect_interval = DEFAULT_PEER_CONNECT_INTERVAL;
+    }
+    this->peer_connect_timeout = config["peer_connect_timeout"].uint64_value();
+    if (!this->peer_connect_timeout)
+    {
+        this->peer_connect_timeout = DEFAULT_PEER_CONNECT_TIMEOUT;
+    }
+    this->osd_idle_timeout = config["osd_idle_timeout"].uint64_value();
+    if (!this->osd_idle_timeout)
+    {
+        this->osd_idle_timeout = DEFAULT_OSD_PING_TIMEOUT;
+    }
+    this->osd_ping_timeout = config["osd_ping_timeout"].uint64_value();
+    if (!this->osd_ping_timeout)
+    {
+        this->osd_ping_timeout = DEFAULT_OSD_PING_TIMEOUT;
+    }
+    this->log_level = config["log_level"].uint64_value();
+}
+
 void osd_messenger_t::connect_peer(uint64_t peer_osd, json11::Json peer_state)
 {
     if (wanted_peers.find(peer_osd) == wanted_peers.end())
@@ -80,6 +173,7 @@ void osd_messenger_t::try_connect_peer(uint64_t peer_osd)
 
 void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer_host, int peer_port)
 {
+    assert(peer_osd != this->osd_num);
     struct sockaddr_in addr;
     int r;
     if ((r = inet_pton(AF_INET, peer_host, &addr.sin_addr)) != 1)
@@ -96,17 +190,6 @@ void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer
         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)
     {
@@ -114,8 +197,18 @@ void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer
         on_connect_peer(peer_osd, -errno);
         return;
     }
-    assert(peer_osd != this->osd_num);
-    clients[peer_fd] = (osd_client_t){
+    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.at(peer_fd)->osd_num;
+            stop_client(peer_fd, true);
+            on_connect_peer(peer_osd, -EIO);
+            return;
+        });
+    }
+    clients[peer_fd] = new osd_client_t((osd_client_t){
         .peer_addr = addr,
         .peer_port = peer_port,
         .peer_fd = peer_fd,
@@ -123,7 +216,7 @@ void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer
         .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
@@ -133,13 +226,13 @@ void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer
 
 void osd_messenger_t::handle_connect_epoll(int peer_fd)
 {
-    auto & cl = clients[peer_fd];
-    if (cl.connect_timeout_id >= 0)
+    auto cl = clients[peer_fd];
+    if (cl->connect_timeout_id >= 0)
     {
-        tfd->clear_timer(cl.connect_timeout_id);
-        cl.connect_timeout_id = -1;
+        tfd->clear_timer(cl->connect_timeout_id);
+        cl->connect_timeout_id = -1;
     }
-    osd_num_t peer_osd = cl.osd_num;
+    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)
@@ -148,13 +241,13 @@ void osd_messenger_t::handle_connect_epoll(int peer_fd)
     }
     if (result != 0)
     {
-        stop_client(peer_fd);
+        stop_client(peer_fd, true);
         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;
+    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);
@@ -170,16 +263,16 @@ void osd_messenger_t::handle_peer_epoll(int peer_fd, int epoll_events)
     {
         // Stop client
         printf("[OSD %lu] client %d disconnected\n", this->osd_num, peer_fd);
-        stop_client(peer_fd);
+        stop_client(peer_fd, true);
     }
     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)
+        auto cl = clients[peer_fd];
+        cl->read_ready++;
+        if (cl->read_ready == 1)
         {
-            read_ready_clients.push_back(cl.peer_fd);
+            read_ready_clients.push_back(cl->peer_fd);
             if (ringloop)
                 ringloop->wakeup();
             else
@@ -219,17 +312,20 @@ void osd_messenger_t::on_connect_peer(osd_num_t peer_osd, int peer_fd)
         }
         return;
     }
-    printf("Connected with peer OSD %lu (fd %d)\n", peer_osd, peer_fd);
+    if (log_level > 0)
+    {
+        printf("[OSD %lu] Connected with peer OSD %lu (client %d)\n", osd_num, peer_osd, peer_fd);
+    }
     wanted_peers.erase(peer_osd);
     repeer_pgs(peer_osd);
 }
 
-void osd_messenger_t::check_peer_config(osd_client_t & cl)
+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 = {
+    op->peer_fd = cl->peer_fd;
+    op->req = (osd_any_op_t){
         .show_conf = {
             .header = {
                 .magic = SECONDARY_OSD_OP_MAGIC,
@@ -238,16 +334,15 @@ void osd_messenger_t::check_peer_config(osd_client_t & cl)
             },
         },
     };
-    op->callback = [this](osd_op_t *op)
+    op->callback = [this, cl](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);
+            printf("Failed to get config from OSD %lu (retval=%ld), disconnecting peer\n", cl->osd_num, op->reply.hdr.retval);
         }
         else
         {
@@ -255,46 +350,37 @@ void osd_messenger_t::check_peer_config(osd_client_t & cl)
             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());
+                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)
+            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);
+                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;
+            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);
+        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)
+void osd_messenger_t::cancel_osd_ops(osd_client_t *cl)
 {
-    for (auto p: cl.sent_ops)
+    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;
-    }
+    cl->sent_ops.clear();
+    cl->outbox.clear();
 }
 
 void osd_messenger_t::cancel_op(osd_op_t *op)
@@ -315,7 +401,7 @@ void osd_messenger_t::cancel_op(osd_op_t *op)
     }
 }
 
-void osd_messenger_t::stop_client(int peer_fd)
+void osd_messenger_t::stop_client(int peer_fd, bool force)
 {
     assert(peer_fd != 0);
     auto it = clients.find(peer_fd);
@@ -324,32 +410,49 @@ void osd_messenger_t::stop_client(int peer_fd)
         return;
     }
     uint64_t repeer_osd = 0;
-    osd_client_t cl = it->second;
-    if (cl.peer_state == PEER_CONNECTED)
+    osd_client_t *cl = it->second;
+    if (cl->peer_state == PEER_CONNECTED)
     {
-        if (cl.osd_num)
+        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;
+            if (log_level > 0)
+                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);
+            if (log_level > 0)
+                printf("[OSD %lu] Stopping client %d (regular client)\n", osd_num, peer_fd);
         }
     }
+    else if (!force)
+    {
+        return;
+    }
+    cl->peer_state = PEER_STOPPED;
     clients.erase(it);
     tfd->set_fd_handler(peer_fd, false, NULL);
-    if (cl.osd_num)
+    if (cl->connect_timeout_id >= 0)
     {
-        osd_peer_fds.erase(cl.osd_num);
-        // Cancel outbound operations
-        cancel_osd_ops(cl);
+        tfd->clear_timer(cl->connect_timeout_id);
+        cl->connect_timeout_id = -1;
     }
-    if (cl.read_op)
+    if (cl->osd_num)
     {
-        delete cl.read_op;
-        cl.read_op = NULL;
+        osd_peer_fds.erase(cl->osd_num);
+    }
+    if (cl->read_op)
+    {
+        if (cl->read_op->callback)
+        {
+            cancel_op(cl->read_op);
+        }
+        else
+        {
+            delete cl->read_op;
+        }
+        cl->read_op = NULL;
     }
     for (auto rit = read_ready_clients.begin(); rit != read_ready_clients.end(); rit++)
     {
@@ -367,12 +470,24 @@ void osd_messenger_t::stop_client(int peer_fd)
             break;
         }
     }
-    free(cl.in_buf);
+    free(cl->in_buf);
+    cl->in_buf = NULL;
     close(peer_fd);
     if (repeer_osd)
     {
+        // First repeer PGs as canceling OSD ops may push new operations
+        // and we need correct PG states when we do that
         repeer_pgs(repeer_osd);
     }
+    if (cl->osd_num)
+    {
+        // Cancel outbound operations
+        cancel_osd_ops(cl);
+    }
+    if (cl->refs <= 0)
+    {
+        delete cl;
+    }
 }
 
 void osd_messenger_t::accept_connections(int listen_fd)
@@ -390,13 +505,13 @@ void osd_messenger_t::accept_connections(int listen_fd)
         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] = {
+        clients[peer_fd] = new osd_client_t((osd_client_t){
             .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)
         {

src/messenger.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #pragma once
 
@@ -30,9 +30,11 @@
 
 #define PEER_CONNECTING 1
 #define PEER_CONNECTED 2
+#define PEER_STOPPED 3
 
 #define DEFAULT_PEER_CONNECT_INTERVAL 5
 #define DEFAULT_PEER_CONNECT_TIMEOUT 5
+#define DEFAULT_OSD_PING_TIMEOUT 5
 
 // Kind of a vector with small-list-optimisation
 struct osd_op_buf_list_t
@@ -190,11 +192,15 @@ struct osd_op_t
 
 struct osd_client_t
 {
+    int refs = 0;
+
     sockaddr_in peer_addr;
     int peer_port;
     int peer_fd;
     int peer_state;
     int connect_timeout_id = -1;
+    int ping_time_remaining = 0;
+    int idle_time_remaining = 0;
     osd_num_t osd_num = 0;
 
     void *in_buf = NULL;
@@ -202,8 +208,8 @@ struct osd_client_t
     // Read state
     int read_ready = 0;
     osd_op_t *read_op = NULL;
-    iovec read_iov;
-    msghdr read_msg;
+    iovec read_iov = { 0 };
+    msghdr read_msg = { 0 };
     int read_remaining = 0;
     int read_state = 0;
     osd_op_buf_list_t recv_list;
@@ -212,17 +218,16 @@ struct osd_client_t
     std::vector<osd_op_t*> received_ops;
 
     // Outbound operations
-    std::deque<osd_op_t*> outbox;
-    std::map<int, osd_op_t*> sent_ops;
+    std::map<uint64_t, 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;
+    msghdr write_msg = { 0 };
     int write_state = 0;
-    osd_op_buf_list_t send_list;
+    std::vector<iovec> send_list, next_send_list;
+    std::vector<osd_op_t*> outbox, next_outbox;
 };
 
 struct osd_wanted_peer_t
@@ -249,6 +254,7 @@ struct osd_messenger_t
 {
     timerfd_manager_t *tfd;
     ring_loop_t *ringloop;
+    int keepalive_timer_id = -1;
 
     // osd_num_t is only for logging and asserts
     osd_num_t osd_num;
@@ -256,6 +262,8 @@ struct osd_messenger_t
     int receive_buffer_size = 64*1024;
     int peer_connect_interval = DEFAULT_PEER_CONNECT_INTERVAL;
     int peer_connect_timeout = DEFAULT_PEER_CONNECT_TIMEOUT;
+    int osd_idle_timeout = DEFAULT_OSD_PING_TIMEOUT;
+    int osd_ping_timeout = DEFAULT_OSD_PING_TIMEOUT;
     int log_level = 0;
     bool use_sync_send_recv = false;
 
@@ -263,7 +271,7 @@ struct osd_messenger_t
     std::map<uint64_t, int> osd_peer_fds;
     uint64_t next_subop_id = 1;
 
-    std::map<int, osd_client_t> clients;
+    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;
@@ -272,8 +280,10 @@ struct osd_messenger_t
     osd_op_stats_t stats;
 
 public:
+    void init();
+    void parse_config(const json11::Json & config);
     void connect_peer(uint64_t osd_num, json11::Json peer_state);
-    void stop_client(int peer_fd);
+    void stop_client(int peer_fd, bool force = false);
     void outbox_push(osd_op_t *cur_op);
     std::function<void(osd_op_t*)> exec_op;
     std::function<void(osd_num_t)> repeer_pgs;
@@ -288,15 +298,16 @@ protected:
     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 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 try_send(osd_client_t *cl);
+    void measure_exec(osd_op_t *cur_op);
+    void handle_send(int result, osd_client_t *cl);
 
-    bool handle_read(int result, int peer_fd);
-    bool handle_finished_read(osd_client_t & cl);
+    bool handle_read(int result, osd_client_t *cl);
+    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);

src/msgr_receive.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #include "messenger.h"
 
@@ -8,135 +8,146 @@ 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)
+        osd_client_t *cl = clients[peer_fd];
+        if (cl->read_msg.msg_iovlen)
         {
-            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;
+            continue;
+        }
+        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();
+            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();
         }
+        cl->refs++;
         if (ringloop && !use_sync_send_recv)
         {
             io_uring_sqe* sqe = ringloop->get_sqe();
             if (!sqe)
             {
+                cl->read_msg.msg_iovlen = 0;
                 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);
+            data->callback = [this, cl](ring_data_t *data) { handle_read(data->res, cl); };
+            my_uring_prep_recvmsg(sqe, peer_fd, &cl->read_msg, 0);
         }
         else
         {
-            int result = recvmsg(peer_fd, &cl.read_msg, 0);
+            int result = recvmsg(peer_fd, &cl->read_msg, 0);
             if (result < 0)
             {
                 result = -errno;
             }
-            handle_read(result, peer_fd);
+            handle_read(result, cl);
         }
     }
     read_ready_clients.clear();
 }
 
-bool osd_messenger_t::handle_read(int result, int peer_fd)
+bool osd_messenger_t::handle_read(int result, osd_client_t *cl)
 {
     bool ret = false;
-    auto cl_it = clients.find(peer_fd);
-    if (cl_it != clients.end())
+    cl->read_msg.msg_iovlen = 0;
+    cl->refs--;
+    if (cl->peer_state == PEER_STOPPED)
     {
-        auto & cl = cl_it->second;
-        if (result <= 0 && result != -EAGAIN)
+        if (cl->refs <= 0)
         {
-            // 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;
+            delete cl;
         }
-        if (result == -EAGAIN || result < cl.read_iov.iov_len)
+        return false;
+    }
+    if (result <= 0 && result != -EAGAIN)
+    {
+        // this is a client socket, so don't panic on error. just disconnect it
+        if (result != 0)
         {
-            cl.read_ready--;
-            if (cl.read_ready > 0)
-                read_ready_clients.push_back(peer_fd);
+            printf("Client %d socket read error: %d (%s). Disconnecting client\n", cl->peer_fd, -result, strerror(-result));
+        }
+        stop_client(cl->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(cl->peer_fd);
+    }
+    else
+    {
+        read_ready_clients.push_back(cl->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 = cl->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
         {
-            read_ready_clients.push_back(peer_fd);
+            // 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 > 0)
+        if (result >= cl->read_iov.iov_len)
         {
-            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;
-            }
+            ret = true;
         }
     }
 fin:
@@ -148,30 +159,36 @@ fin:
     return ret;
 }
 
-bool osd_messenger_t::handle_finished_read(osd_client_t & cl)
+bool osd_messenger_t::handle_finished_read(osd_client_t *cl)
 {
-    cl.recv_list.reset();
-    if (cl.read_state == CL_READ_HDR)
+    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);
+        if (cl->read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
+            return handle_reply_hdr(cl);
+        else if (cl->read_op->req.hdr.magic == SECONDARY_OSD_OP_MAGIC)
+            handle_op_hdr(cl);
         else
-            handle_op_hdr(&cl);
+        {
+            printf("Received garbage: magic=%lx id=%lu opcode=%lx from %d\n", cl->read_op->req.hdr.magic, cl->read_op->req.hdr.id, cl->read_op->req.hdr.opcode, cl->peer_fd);
+            stop_client(cl->peer_fd);
+            return false;
+        }
     }
-    else if (cl.read_state == CL_READ_DATA)
+    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;
+        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)
+    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;
+        handle_reply_ready(cl->read_op);
+        cl->read_op = NULL;
+        cl->read_state = 0;
     }
     else
     {
@@ -247,6 +264,14 @@ bool osd_messenger_t::handle_reply_hdr(osd_client_t *cl)
     {
         // Read data. In this case we assume that the buffer is preallocated by the caller (!)
         assert(op->iov.count > 0);
+        if (op->reply.hdr.retval != (op->reply.hdr.opcode == OSD_OP_SEC_READ ? op->req.sec_rw.len : op->req.rw.len))
+        {
+            // Check reply length to not overflow the buffer
+            printf("Client %d read reply of different length\n", cl->peer_fd);
+            cl->sent_ops[op->req.hdr.id] = op;
+            stop_client(cl->peer_fd);
+            return false;
+        }
         cl->recv_list.append(op->iov);
         delete cl->read_op;
         cl->read_op = op;

src/msgr_send.cpp

@@ -0,0 +1,237 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#define _XOPEN_SOURCE
+#include <limits.h>
+
+#include "messenger.h"
+
+void osd_messenger_t::outbox_push(osd_op_t *cur_op)
+{
+    assert(cur_op->peer_fd);
+    osd_client_t *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
+        // FIXME: Review if this is still needed
+        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;
+        }
+    }
+    auto & to_send_list = cl->write_msg.msg_iovlen ? cl->next_send_list : cl->send_list;
+    auto & to_outbox = cl->write_msg.msg_iovlen ? cl->next_outbox : cl->outbox;
+    if (cur_op->op_type == OSD_OP_IN)
+    {
+        measure_exec(cur_op);
+        to_send_list.push_back((iovec){ .iov_base = cur_op->reply.buf, .iov_len = OSD_PACKET_SIZE });
+    }
+    else
+    {
+        to_send_list.push_back((iovec){ .iov_base = cur_op->req.buf, .iov_len = OSD_PACKET_SIZE });
+        cl->sent_ops[cur_op->req.hdr.id] = cur_op;
+    }
+    to_outbox.push_back(NULL);
+    // Operation data
+    if ((cur_op->op_type == OSD_OP_IN
+        ? (cur_op->req.hdr.opcode == OSD_OP_READ ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_READ ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_LIST ||
+        cur_op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)
+        : (cur_op->req.hdr.opcode == OSD_OP_WRITE ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK)) && cur_op->iov.count > 0)
+    {
+        for (int i = 0; i < cur_op->iov.count; i++)
+        {
+            assert(cur_op->iov.buf[i].iov_base);
+            to_send_list.push_back(cur_op->iov.buf[i]);
+            to_outbox.push_back(NULL);
+        }
+    }
+    if (cur_op->op_type == OSD_OP_IN)
+    {
+        // To free it later
+        to_outbox[to_outbox.size()-1] = cur_op;
+    }
+    if (!ringloop)
+    {
+        // FIXME: It's worse because it doesn't allow batching
+        while (cl->outbox.size())
+        {
+            try_send(cl);
+        }
+    }
+    else if (cl->write_msg.msg_iovlen > 0 || !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();
+    }
+}
+
+void osd_messenger_t::measure_exec(osd_op_t *cur_op)
+{
+    // Measure execution latency
+    if (cur_op->req.hdr.opcode > OSD_OP_MAX)
+    {
+        return;
+    }
+    timespec tv_end;
+    clock_gettime(CLOCK_REALTIME, &tv_end);
+    stats.op_stat_count[cur_op->req.hdr.opcode]++;
+    if (!stats.op_stat_count[cur_op->req.hdr.opcode])
+    {
+        stats.op_stat_count[cur_op->req.hdr.opcode]++;
+        stats.op_stat_sum[cur_op->req.hdr.opcode] = 0;
+        stats.op_stat_bytes[cur_op->req.hdr.opcode] = 0;
+    }
+    stats.op_stat_sum[cur_op->req.hdr.opcode] += (
+        (tv_end.tv_sec - cur_op->tv_begin.tv_sec)*1000000 +
+        (tv_end.tv_nsec - cur_op->tv_begin.tv_nsec)/1000
+    );
+    if (cur_op->req.hdr.opcode == OSD_OP_READ ||
+        cur_op->req.hdr.opcode == OSD_OP_WRITE)
+    {
+        stats.op_stat_bytes[cur_op->req.hdr.opcode] += cur_op->req.rw.len;
+    }
+    else 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)
+    {
+        stats.op_stat_bytes[cur_op->req.hdr.opcode] += cur_op->req.sec_rw.len;
+    }
+}
+
+bool osd_messenger_t::try_send(osd_client_t *cl)
+{
+    int peer_fd = cl->peer_fd;
+    if (!cl->send_list.size() || cl->write_msg.msg_iovlen > 0)
+    {
+        return true;
+    }
+    if (ringloop && !use_sync_send_recv)
+    {
+        io_uring_sqe* sqe = ringloop->get_sqe();
+        if (!sqe)
+        {
+            return false;
+        }
+        cl->write_msg.msg_iov = cl->send_list.data();
+        cl->write_msg.msg_iovlen = cl->send_list.size() < IOV_MAX ? cl->send_list.size() : IOV_MAX;
+        cl->refs++;
+        ring_data_t* data = ((ring_data_t*)sqe->user_data);
+        data->callback = [this, cl](ring_data_t *data) { handle_send(data->res, cl); };
+        my_uring_prep_sendmsg(sqe, peer_fd, &cl->write_msg, 0);
+    }
+    else
+    {
+        cl->write_msg.msg_iov = cl->send_list.data();
+        cl->write_msg.msg_iovlen = cl->send_list.size() < IOV_MAX ? cl->send_list.size() : IOV_MAX;
+        cl->refs++;
+        int result = sendmsg(peer_fd, &cl->write_msg, MSG_NOSIGNAL);
+        if (result < 0)
+        {
+            result = -errno;
+        }
+        handle_send(result, cl);
+    }
+    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];
+        auto cl_it = clients.find(peer_fd);
+        if (cl_it != clients.end() && !try_send(cl_it->second))
+        {
+            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, osd_client_t *cl)
+{
+    cl->write_msg.msg_iovlen = 0;
+    cl->refs--;
+    if (cl->peer_state == PEER_STOPPED)
+    {
+        if (!cl->refs)
+        {
+            delete cl;
+        }
+        return;
+    }
+    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", cl->peer_fd, -result, strerror(-result));
+        stop_client(cl->peer_fd);
+        return;
+    }
+    if (result >= 0)
+    {
+        int done = 0;
+        while (result > 0 && done < cl->send_list.size())
+        {
+            iovec & iov = cl->send_list[done];
+            if (iov.iov_len <= result)
+            {
+                if (cl->outbox[done])
+                {
+                    // Reply fully sent
+                    delete cl->outbox[done];
+                }
+                result -= iov.iov_len;
+                done++;
+            }
+            else
+            {
+                iov.iov_len -= result;
+                iov.iov_base += result;
+                break;
+            }
+        }
+        if (done > 0)
+        {
+            cl->send_list.erase(cl->send_list.begin(), cl->send_list.begin()+done);
+            cl->outbox.erase(cl->outbox.begin(), cl->outbox.begin()+done);
+        }
+        if (cl->next_send_list.size())
+        {
+            cl->send_list.insert(cl->send_list.end(), cl->next_send_list.begin(), cl->next_send_list.end());
+            cl->outbox.insert(cl->outbox.end(), cl->next_outbox.begin(), cl->next_outbox.end());
+            cl->next_send_list.clear();
+            cl->next_outbox.clear();
+        }
+        cl->write_state = cl->outbox.size() > 0 ? CL_WRITE_READY : 0;
+    }
+    if (cl->write_state != 0)
+    {
+        write_ready_clients.push_back(cl->peer_fd);
+    }
+}

src/nbd_proxy.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 // Similar to qemu-nbd, but sets timeout and uses io_uring
 
 #include <linux/nbd.h>
@@ -17,6 +17,10 @@
 #include "epoll_manager.h"
 #include "cluster_client.h"
 
+#ifndef MSG_ZEROCOPY
+#define MSG_ZEROCOPY 0
+#endif
+
 const char *exe_name = NULL;
 
 class nbd_proxy
@@ -107,7 +111,7 @@ public:
     {
         printf(
             "Vitastor NBD proxy\n"
-            "(c) Vitaliy Filippov, 2020 (VNPL-1.0 or GNU GPL 2.0+)\n\n"
+            "(c) Vitaliy Filippov, 2020 (VNPL-1.1)\n\n"
             "USAGE:\n"
             "  %s map --etcd_address <etcd_address> --pool <pool> --inode <inode> --size <size in bytes>\n"
             "  %s unmap /dev/nbd0\n"

src/object_id.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #pragma once
 

src/osd.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include <sys/socket.h>
 #include <sys/poll.h>
@@ -28,11 +28,16 @@ osd_t::osd_t(blockstore_config_t & config, blockstore_t *bs, ring_loop_t *ringlo
     {
         print_stats();
     });
+    this->tfd->set_timer(slow_log_interval*1000, true, [this](int timer_id)
+    {
+        print_slow();
+    });
 
     c_cli.tfd = this->tfd;
     c_cli.ringloop = this->ringloop;
     c_cli.exec_op = [this](osd_op_t *op) { exec_op(op); };
     c_cli.repeer_pgs = [this](osd_num_t peer_osd) { repeer_pgs(peer_osd); };
+    c_cli.init();
 
     init_cluster();
 
@@ -49,6 +54,9 @@ osd_t::~osd_t()
 
 void osd_t::parse_config(blockstore_config_t & config)
 {
+    if (config.find("log_level") == config.end())
+        config["log_level"] = "1";
+    log_level = strtoull(config["log_level"].c_str(), NULL, 10);
     // Initial startup configuration
     json11::Json json_config = json11::Json(config);
     st_cli.parse_config(json_config);
@@ -60,6 +68,8 @@ void osd_t::parse_config(blockstore_config_t & config)
         throw std::runtime_error("osd_num is required in the configuration");
     c_cli.osd_num = osd_num;
     run_primary = config["run_primary"] != "false" && config["run_primary"] != "0" && config["run_primary"] != "no";
+    no_rebalance = config["no_rebalance"] == "true" || config["no_rebalance"] == "1" || config["no_rebalance"] == "yes";
+    no_recovery = config["no_recovery"] == "true" || config["no_recovery"] == "1" || config["no_recovery"] == "yes";
     // Cluster configuration
     bind_address = config["bind_address"];
     if (bind_address == "")
@@ -86,19 +96,18 @@ void osd_t::parse_config(blockstore_config_t & config)
     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;
+    recovery_sync_batch = strtoull(config["recovery_sync_batch"].c_str(), NULL, 10);
+    if (recovery_sync_batch < 1 || recovery_sync_batch > MAX_RECOVERY_QUEUE)
+        recovery_sync_batch = DEFAULT_RECOVERY_BATCH;
     if (config["readonly"] == "true" || config["readonly"] == "1" || config["readonly"] == "yes")
         readonly = true;
     print_stats_interval = strtoull(config["print_stats_interval"].c_str(), NULL, 10);
     if (!print_stats_interval)
         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 = 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 = DEFAULT_PEER_CONNECT_TIMEOUT;
-    log_level = strtoull(config["log_level"].c_str(), NULL, 10);
-    c_cli.log_level = log_level;
+    slow_log_interval = strtoull(config["slow_log_interval"].c_str(), NULL, 10);
+    if (!slow_log_interval)
+        slow_log_interval = 10;
+    c_cli.parse_config(json_config);
 }
 
 void osd_t::bind_socket()
@@ -202,6 +211,12 @@ void osd_t::exec_op(osd_op_t *cur_op)
         finish_op(cur_op, -EINVAL);
         return;
     }
+    if (cur_op->req.hdr.opcode == OSD_OP_PING)
+    {
+        // Pong
+        finish_op(cur_op, 0);
+        return;
+    }
     if (readonly &&
         cur_op->req.hdr.opcode != OSD_OP_SEC_READ &&
         cur_op->req.hdr.opcode != OSD_OP_SEC_LIST &&
@@ -252,9 +267,9 @@ void osd_t::reset_stats()
 
 void osd_t::print_stats()
 {
-    for (int i = 0; i <= OSD_OP_MAX; i++)
+    for (int i = OSD_OP_MIN; i <= OSD_OP_MAX; i++)
     {
-        if (c_cli.stats.op_stat_count[i] != prev_stats.op_stat_count[i])
+        if (c_cli.stats.op_stat_count[i] != prev_stats.op_stat_count[i] && i != OSD_OP_PING)
         {
             uint64_t avg = (c_cli.stats.op_stat_sum[i] - prev_stats.op_stat_sum[i])/(c_cli.stats.op_stat_count[i] - prev_stats.op_stat_count[i]);
             uint64_t bw = (c_cli.stats.op_stat_bytes[i] - prev_stats.op_stat_bytes[i]) / print_stats_interval;
@@ -275,7 +290,7 @@ void osd_t::print_stats()
             prev_stats.op_stat_bytes[i] = c_cli.stats.op_stat_bytes[i];
         }
     }
-    for (int i = 0; i <= OSD_OP_MAX; i++)
+    for (int i = OSD_OP_MIN; i <= OSD_OP_MAX; i++)
     {
         if (c_cli.stats.subop_stat_count[i] != prev_stats.subop_stat_count[i])
         {
@@ -313,3 +328,73 @@ void osd_t::print_stats()
         printf("[OSD %lu] %lu object(s) misplaced\n", osd_num, misplaced_objects);
     }
 }
+
+void osd_t::print_slow()
+{
+    char alloc[1024];
+    timespec now;
+    clock_gettime(CLOCK_REALTIME, &now);
+    for (auto & kv: c_cli.clients)
+    {
+        for (auto op: kv.second->received_ops)
+        {
+            if ((now.tv_sec - op->tv_begin.tv_sec) >= slow_log_interval)
+            {
+                int l = sizeof(alloc), n;
+                char *buf = alloc;
+#define bufprintf(s, ...) { n = snprintf(buf, l, s, __VA_ARGS__); n = n < 0 ? 0 : n; buf += n; l -= n; }
+                bufprintf("[OSD %lu] Slow op", osd_num);
+                if (kv.second->osd_num)
+                {
+                    bufprintf(" from peer OSD %lu (client %d)", kv.second->osd_num, kv.second->peer_fd);
+                }
+                else
+                {
+                    bufprintf(" from client %d", kv.second->peer_fd);
+                }
+                bufprintf(": %s id=%lu", osd_op_names[op->req.hdr.opcode], op->req.hdr.id);
+                if (op->req.hdr.opcode == OSD_OP_SEC_READ || op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+                    op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE || op->req.hdr.opcode == OSD_OP_SEC_DELETE)
+                {
+                    bufprintf(" %lx:%lx v", op->req.sec_rw.oid.inode, op->req.sec_rw.oid.stripe);
+                    if (op->req.sec_rw.version == UINT64_MAX)
+                    {
+                        bufprintf("%s", "max");
+                    }
+                    else
+                    {
+                        bufprintf("%lu", op->req.sec_rw.version);
+                    }
+                    if (op->req.hdr.opcode != OSD_OP_SEC_DELETE)
+                    {
+                        bufprintf(" offset=%x len=%x", op->req.sec_rw.offset, op->req.sec_rw.len);
+                    }
+                }
+                else if (op->req.hdr.opcode == OSD_OP_SEC_STABILIZE || op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK)
+                {
+                    for (uint64_t i = 0; i < op->req.sec_stab.len; i += sizeof(obj_ver_id))
+                    {
+                        obj_ver_id *ov = (obj_ver_id*)(op->buf + i);
+                        bufprintf(i == 0 ? " %lx:%lx v%lu" : ", %lx:%lx v%lu", ov->oid.inode, ov->oid.stripe, ov->version);
+                    }
+                }
+                else if (op->req.hdr.opcode == OSD_OP_SEC_LIST)
+                {
+                    bufprintf(
+                        " inode=%lx-%lx pg=%u/%u, stripe=%lu",
+                        op->req.sec_list.min_inode, op->req.sec_list.max_inode,
+                        op->req.sec_list.list_pg, op->req.sec_list.pg_count,
+                        op->req.sec_list.pg_stripe_size
+                    );
+                }
+                else if (op->req.hdr.opcode == OSD_OP_READ || op->req.hdr.opcode == OSD_OP_WRITE ||
+                    op->req.hdr.opcode == OSD_OP_DELETE)
+                {
+                    bufprintf(" inode=%lx offset=%lx len=%x", op->req.rw.inode, op->req.rw.offset, op->req.rw.len);
+                }
+#undef bufprintf
+                printf("%s\n", alloc);
+            }
+        }
+    }
+}

src/osd.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #pragma once
 
@@ -37,6 +37,7 @@
 #define DEFAULT_AUTOSYNC_INTERVAL 5
 #define MAX_RECOVERY_QUEUE 2048
 #define DEFAULT_RECOVERY_QUEUE 4
+#define DEFAULT_RECOVERY_BATCH 16
 
 //#define OSD_STUB
 
@@ -64,15 +65,19 @@ class osd_t
     bool readonly = false;
     osd_num_t osd_num = 1; // OSD numbers start with 1
     bool run_primary = false;
+    bool no_rebalance = false;
+    bool no_recovery = false;
     std::string bind_address;
     int bind_port, listen_backlog;
     // FIXME: Implement client queue depth limit
     int client_queue_depth = 128;
     bool allow_test_ops = true;
     int print_stats_interval = 3;
+    int slow_log_interval = 10;
     int immediate_commit = IMMEDIATE_NONE;
     int autosync_interval = DEFAULT_AUTOSYNC_INTERVAL; // sync every 5 seconds
     int recovery_queue_depth = DEFAULT_RECOVERY_QUEUE;
+    int recovery_sync_batch = DEFAULT_RECOVERY_BATCH;
     int log_level = 0;
 
     // cluster state
@@ -94,9 +99,11 @@ class osd_t
     std::map<pool_pg_num_t, pg_t> pgs;
     std::set<pool_pg_num_t> dirty_pgs;
     std::set<osd_num_t> dirty_osds;
+    int copies_to_delete_after_sync_count = 0;
     uint64_t misplaced_objects = 0, degraded_objects = 0, incomplete_objects = 0;
     int peering_state = 0;
     std::map<object_id, osd_recovery_op_t> recovery_ops;
+    int recovery_done = 0;
     osd_op_t *autosync_op = NULL;
 
     // Unstable writes
@@ -138,6 +145,7 @@ class osd_t
     void create_osd_state();
     void renew_lease();
     void print_stats();
+    void print_slow();
     void reset_stats();
     json11::Json get_statistics();
     void report_statistics();
@@ -158,6 +166,7 @@ class osd_t
     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_t & pg);
+    void reset_pg(pg_t & pg);
     void finish_stop_pg(pg_t & pg);
 
     // flushing, recovery and backfill
@@ -196,6 +205,7 @@ class osd_t
     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, 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, uint64_t set_size, pg_osd_set_t & loc_set);
+    void submit_primary_del_batch(osd_op_t *cur_op, obj_ver_osd_t *chunks_to_delete, int chunks_to_delete_count);
     void submit_primary_sync_subops(osd_op_t *cur_op);
     void submit_primary_stab_subops(osd_op_t *cur_op);
 

src/osd_cluster.cpp

@@ -1,9 +1,10 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "osd.h"
 #include "base64.h"
 #include "etcd_state_client.h"
+#include "osd_rmw.h"
 
 // Startup sequence:
 //   Start etcd watcher -> Load global OSD configuration -> Bind socket -> Acquire lease -> Report&lock OSD state
@@ -32,12 +33,26 @@ void osd_t::init_cluster()
             }
             pgs[{ 1, 1 }] = (pg_t){
                 .state = PG_PEERING,
+                .scheme = POOL_SCHEME_XOR,
                 .pg_cursize = 0,
+                .pg_size = 3,
+                .pg_minsize = 2,
+                .pg_data_size = 2,
                 .pool_id = 1,
                 .pg_num = 1,
                 .target_set = { 1, 2, 3 },
                 .cur_set = { 0, 0, 0 },
             };
+            st_cli.pool_config[1] = (pool_config_t){
+                .exists = true,
+                .id = 1,
+                .name = "testpool",
+                .scheme = POOL_SCHEME_XOR,
+                .pg_size = 3,
+                .pg_minsize = 2,
+                .pg_count = 1,
+                .real_pg_count = 1,
+            };
             report_pg_state(pgs[{ 1, 1 }]);
             pg_counts[1] = 1;
         }
@@ -127,7 +142,7 @@ json11::Json osd_t::get_statistics()
     }
     st["host"] = self_state["host"];
     json11::Json::object op_stats, subop_stats;
-    for (int i = 0; i <= OSD_OP_MAX; i++)
+    for (int i = OSD_OP_MIN; i <= OSD_OP_MAX; i++)
     {
         op_stats[osd_op_names[i]] = json11::Json::object {
             { "count", c_cli.stats.op_stat_count[i] },
@@ -135,7 +150,7 @@ json11::Json osd_t::get_statistics()
             { "bytes", c_cli.stats.op_stat_bytes[i] },
         };
     }
-    for (int i = 0; i <= OSD_OP_MAX; i++)
+    for (int i = OSD_OP_MIN; i <= OSD_OP_MAX; i++)
     {
         subop_stats[osd_op_names[i]] = json11::Json::object {
             { "count", c_cli.stats.subop_stat_count[i] },
@@ -223,8 +238,11 @@ void osd_t::on_change_osd_state_hook(osd_num_t peer_osd)
 void osd_t::on_change_etcd_state_hook(json11::Json::object & changes)
 {
     // FIXME apply config changes in runtime (maybe, some)
-    apply_pg_count();
-    apply_pg_config();
+    if (run_primary)
+    {
+        apply_pg_count();
+        apply_pg_config();
+    }
 }
 
 void osd_t::on_change_pg_history_hook(pool_id_t pool_id, pg_num_t pg_num)
@@ -270,7 +288,6 @@ void osd_t::on_load_config_hook(json11::Json::object & global_config)
     }
     parse_config(osd_config);
     bind_socket();
-    st_cli.start_etcd_watcher();
     acquire_lease();
 }
 
@@ -367,6 +384,7 @@ void osd_t::create_osd_state()
         {
             st_cli.load_pgs();
         }
+        report_statistics();
     });
 }
 
@@ -477,7 +495,11 @@ void osd_t::apply_pg_count()
             }
             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);
+                printf(
+                    "[OSD %lu] PG count change detected for pool %u (new is %lu, old is %u),"
+                    " but %u PG(s) are still active. This is not allowed. Exiting\n",
+                    this->osd_num, pool_item.first, pool_item.second.real_pg_count, pg_counts[pool_item.first], still_active
+                );
                 force_stop(1);
                 return;
             }
@@ -571,7 +593,10 @@ void osd_t::apply_pg_config()
                     }
                     else
                     {
-                        throw std::runtime_error("Unexpected PG "+std::to_string(pg_num)+" state: "+std::to_string(pg_it->second.state));
+                        throw std::runtime_error(
+                            "Unexpected PG "+std::to_string(pool_id)+"/"+std::to_string(pg_num)+
+                            " state: "+std::to_string(pg_it->second.state)
+                        );
                     }
                 }
                 auto & pg = this->pgs[{ .pool_id = pool_id, .pg_num = pg_num }];
@@ -581,6 +606,8 @@ void osd_t::apply_pg_config()
                     .pg_cursize = 0,
                     .pg_size = pool_item.second.pg_size,
                     .pg_minsize = pool_item.second.pg_minsize,
+                    .pg_data_size = pg.scheme == POOL_SCHEME_REPLICATED
+                         ? 1 : pool_item.second.pg_size - pool_item.second.parity_chunks,
                     .pool_id = pool_id,
                     .pg_num = pg_num,
                     .reported_epoch = pg_cfg.epoch,
@@ -588,6 +615,10 @@ void osd_t::apply_pg_config()
                     .all_peers = std::vector<osd_num_t>(all_peers.begin(), all_peers.end()),
                     .target_set = pg_cfg.target_set,
                 };
+                if (pg.scheme == POOL_SCHEME_JERASURE)
+                {
+                    use_jerasure(pg.pg_size, pg.pg_data_size, true);
+                }
                 this->pg_state_dirty.insert({ .pool_id = pool_id, .pg_num = pg_num });
                 pg.print_state();
                 if (pg_cfg.cur_primary == this->osd_num)
@@ -634,7 +665,21 @@ void osd_t::report_pg_states()
         auto & pg = pg_it->second;
         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.pool_id)+"/"+std::to_string(pg.pg_num));
-        if (pg.state == PG_STARTING)
+        bool pg_state_exists = false;
+        if (pg.state != PG_STARTING)
+        {
+            auto pool_it = st_cli.pool_config.find(pg.pool_id);
+            if (pool_it != st_cli.pool_config.end())
+            {
+                auto pg_it = pool_it->second.pg_config.find(pg.pg_num);
+                if (pg_it != pool_it->second.pg_config.end() &&
+                    pg_it->second.cur_state != 0)
+                {
+                    pg_state_exists = true;
+                }
+            }
+        }
+        if (!pg_state_exists)
         {
             // Check that the PG key does not exist
             // Failed check indicates an unsuccessful PG lock attempt in this case
@@ -646,9 +691,7 @@ void osd_t::report_pg_states()
         }
         else
         {
-            // Check that the key is ours
-            // Failed check indicates success for OFFLINE pgs (PG lock is already deleted)
-            // and an unexpected race condition for started pgs (PG lock is held by someone else)
+            // Check that the key is ours if it already exists
             checks.push_back(json11::Json::object {
                 { "target", "LEASE" },
                 { "lease", etcd_lease_id },
@@ -770,13 +813,16 @@ void osd_t::report_pg_states()
             for (auto pp: reporting_pgs)
             {
                 auto pg_it = this->pgs.find(pp.first);
-                if (pg_it != this->pgs.end())
+                if (pg_it != this->pgs.end() &&
+                    pg_it->second.state == PG_OFFLINE &&
+                    pg_state_dirty.find(pp.first) == pg_state_dirty.end())
                 {
-                    if (pg_it->second.state == PG_OFFLINE)
+                    // Forget offline PGs after reporting their state
+                    if (pg_it->second.scheme == POOL_SCHEME_JERASURE)
                     {
-                        // Remove offline PGs after reporting their state
-                        this->pgs.erase(pg_it);
+                        use_jerasure(pg_it->second.pg_size, pg_it->second.pg_data_size, false);
                     }
+                    this->pgs.erase(pg_it);
                 }
             }
             // Push other PG state updates, if any

src/osd_flush.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "osd.h"
 
@@ -95,7 +95,7 @@ void osd_t::handle_flush_op(bool rollback, pool_id_t pool_id, pg_num_t pg_num, p
     {
         // This flush batch is done
         std::vector<osd_op_t*> continue_ops;
-        auto & pg = pgs[pg_id];
+        auto & pg = pgs.at(pg_id);
         auto it = pg.flush_actions.begin(), prev_it = it;
         auto erase_start = it;
         while (1)
@@ -166,7 +166,7 @@ void osd_t::submit_flush_op(pool_id_t pool_id, pg_num_t pg_num, pg_flush_batch_t
     {
         // local
         clock_gettime(CLOCK_REALTIME, &op->tv_begin);
-        op->bs_op = new blockstore_op_t({
+        op->bs_op = new blockstore_op_t((blockstore_op_t){
             .opcode = (uint64_t)(rollback ? BS_OP_ROLLBACK : BS_OP_STABLE),
             .callback = [this, op, pool_id, pg_num, fb](blockstore_op_t *bs_op)
             {
@@ -188,7 +188,7 @@ void osd_t::submit_flush_op(pool_id_t pool_id, pg_num_t pg_num, pg_flush_batch_t
         op->op_type = OSD_OP_OUT;
         op->iov.push_back(op->buf, count * sizeof(obj_ver_id));
         op->peer_fd = peer_fd;
-        op->req = {
+        op->req = (osd_any_op_t){
             .sec_stab = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,
@@ -209,32 +209,38 @@ void osd_t::submit_flush_op(pool_id_t pool_id, pg_num_t pg_num, pg_flush_batch_t
 
 bool osd_t::pick_next_recovery(osd_recovery_op_t &op)
 {
-    for (auto pg_it = pgs.begin(); pg_it != pgs.end(); pg_it++)
+    if (!no_recovery)
     {
-        if ((pg_it->second.state & (PG_ACTIVE | PG_HAS_DEGRADED)) == (PG_ACTIVE | PG_HAS_DEGRADED))
+        for (auto pg_it = pgs.begin(); pg_it != pgs.end(); pg_it++)
         {
-            for (auto obj_it = pg_it->second.degraded_objects.begin(); obj_it != pg_it->second.degraded_objects.end(); obj_it++)
+            if ((pg_it->second.state & (PG_ACTIVE | PG_HAS_DEGRADED)) == (PG_ACTIVE | PG_HAS_DEGRADED))
             {
-                if (recovery_ops.find(obj_it->first) == recovery_ops.end())
+                for (auto obj_it = pg_it->second.degraded_objects.begin(); obj_it != pg_it->second.degraded_objects.end(); obj_it++)
                 {
-                    op.degraded = true;
-                    op.oid = obj_it->first;
-                    return true;
+                    if (recovery_ops.find(obj_it->first) == recovery_ops.end())
+                    {
+                        op.degraded = true;
+                        op.oid = obj_it->first;
+                        return true;
+                    }
                 }
             }
         }
     }
-    for (auto pg_it = pgs.begin(); pg_it != pgs.end(); pg_it++)
+    if (!no_rebalance)
     {
-        if ((pg_it->second.state & (PG_ACTIVE | PG_HAS_MISPLACED)) == (PG_ACTIVE | PG_HAS_MISPLACED))
+        for (auto pg_it = pgs.begin(); pg_it != pgs.end(); pg_it++)
         {
-            for (auto obj_it = pg_it->second.misplaced_objects.begin(); obj_it != pg_it->second.misplaced_objects.end(); obj_it++)
+            if ((pg_it->second.state & (PG_ACTIVE | PG_HAS_MISPLACED)) == (PG_ACTIVE | PG_HAS_MISPLACED))
             {
-                if (recovery_ops.find(obj_it->first) == recovery_ops.end())
+                for (auto obj_it = pg_it->second.misplaced_objects.begin(); obj_it != pg_it->second.misplaced_objects.end(); obj_it++)
                 {
-                    op.degraded = false;
-                    op.oid = obj_it->first;
-                    return true;
+                    if (recovery_ops.find(obj_it->first) == recovery_ops.end())
+                    {
+                        op.degraded = false;
+                        op.oid = obj_it->first;
+                        return true;
+                    }
                 }
             }
         }
@@ -246,7 +252,7 @@ void osd_t::submit_recovery_op(osd_recovery_op_t *op)
 {
     op->osd_op = new osd_op_t();
     op->osd_op->op_type = OSD_OP_OUT;
-    op->osd_op->req = {
+    op->osd_op->req = (osd_any_op_t){
         .rw = {
             .header = {
                 .magic = SECONDARY_OSD_OP_MAGIC,
@@ -258,15 +264,23 @@ void osd_t::submit_recovery_op(osd_recovery_op_t *op)
             .len = 0,
         },
     };
+    if (log_level > 2)
+    {
+        printf("Submitting recovery operation for %lx:%lx\n", op->oid.inode, op->oid.stripe);
+    }
     op->osd_op->callback = [this, op](osd_op_t *osd_op)
     {
-        // Don't sync the write, it will be synced by our regular sync coroutine
         if (osd_op->reply.hdr.retval < 0)
         {
             // Error recovering object
             if (osd_op->reply.hdr.retval == -EPIPE)
             {
                 // PG is stopped or one of the OSDs is gone, error is harmless
+                printf(
+                    "Recovery operation failed with object %lx:%lx (PG %u/%u)\n",
+                    op->oid.inode, op->oid.stripe, INODE_POOL(op->oid.inode),
+                    map_to_pg(op->oid, st_cli.pool_config.at(INODE_POOL(op->oid.inode)).pg_stripe_size)
+                );
             }
             else
             {
@@ -277,6 +291,17 @@ void osd_t::submit_recovery_op(osd_recovery_op_t *op)
         op->osd_op = NULL;
         recovery_ops.erase(op->oid);
         delete osd_op;
+        if (immediate_commit != IMMEDIATE_ALL)
+        {
+            recovery_done++;
+            if (recovery_done >= recovery_sync_batch)
+            {
+                // Force sync every <recovery_sync_batch> operations
+                // This is required not to pile up an excessive amount of delete operations
+                autosync();
+                recovery_done = 0;
+            }
+        }
         continue_recovery();
     };
     exec_op(op->osd_op);

src/osd_id.h

@@ -1,10 +1,11 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #pragma once
 
 #define POOL_SCHEME_REPLICATED 1
 #define POOL_SCHEME_XOR 2
+#define POOL_SCHEME_JERASURE 3
 #define POOL_ID_MAX 0x10000
 #define POOL_ID_BITS 16
 #define INODE_POOL(inode) (pool_id_t)((inode) >> (64 - POOL_ID_BITS))

src/osd_main.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 (see README.md for details)
+// License: VNPL-1.1 (see README.md for details)
 
 #include "osd.h"
 

src/osd_ops.cpp

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #include "osd_ops.h"
 
@@ -19,4 +19,5 @@ const char* osd_op_names[] = {
     "primary_write",
     "primary_sync",
     "primary_delete",
+    "ping",
 };

src/osd_ops.h

@@ -1,5 +1,5 @@
 // Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.0 or GNU GPL-2.0+ (see README.md for details)
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #pragma once
 
@@ -27,7 +27,8 @@
 #define OSD_OP_WRITE                12
 #define OSD_OP_SYNC                 13
 #define OSD_OP_DELETE               14
-#define OSD_OP_MAX                  14
+#define OSD_OP_PING                 15
+#define OSD_OP_MAX                  15
 // Alignment & limit for read/write operations
 #ifndef MEM_ALIGNMENT
 #define MEM_ALIGNMENT               512