A couple of fixes for EC pools
- Fix a segfault possible on partial EC overwrite in 1234 -> 5030 rebalance scenario
- Fix two problems leading to EC pools stalling on rebalance & parallel sudden stops
of OSDs, for example during a sudden poweroff of a host:
- Recovery auto-tuning (1.4.0 feature) could apply too large delays and stall
the EC journal - fixed by limiting delays with a new recovery_tune_sleep_cutoff_us
parameter (10 seconds by default) and applying recovery pauses before write
operations, not after them, to not occupy space in the journal for long time
- Dynamic journal space reservation (1.3.0 feature) wasn't accounting new writes
when checking the limit so OSDs could still fill the journal fully and stall -
fixed by including new writes into the limit
- Print etcd dbSize instead of dbSizeInUse in status
Hotfix for hotfix O:-)
- "Write stall fix" was incomplete and EC write stalls could
continue even on 1.4.2. Now they're finally fixed O:-)
- Make monitor ignore statistics of stopped OSDs. Previously if you stopped all
OSDs the last total I/O numbers would remain the same indefinitely
- Log to systemd by default
- Fix excessive autosyncs after every operation with disabled immediate_commit (introduced in 1.1.0)
- Fix a possible write stall with EC due to the lack of OSD wakeup after stabilizing previous writes
- Change sync operation semantics as a final fix to possible write stalls with EC and disabled immediate_commit
- Sync after deleting data in CLI rm / rm-data if immediate_commit is disabled
- Fix OSDs ignoring syncs & autosyncs for delete operations
- Fix OSD space reporting sometimes adding garbage zeros for deleted inodes (causing extra pool/stats etcd keys for deleted pools)
- Speed up monitor failover - change default etcd_mon_ttl from 30 to 5 seconds
- Speed up operation retries - change default up_wait_retry_interval to 50 ms
- Add patch for libvirt 9.10
Should be a final remaining fix to EC + non-capacitor (non-immediate-commit) write hangs :).
First it was breaking non-EC ("instantly stable") writes because they sometimes
complete out of order which was leading to the following error:
terminate called after throwing an instance of 'std::runtime_error'
what(): BUG: Unexpected dirty_entry 1000000000001:29480000 v65540 unstable state during flush: 0x151
But it is easily fixed by scanning previous and next dirty_entries in mark_stable.
- Fix a monitor crash on primary OSD switching introduced in 1.4.0
- Fix "partly outside array bounds" warnings for GCC 12 in cpp-btree
- Fix a realloc memory leak in theory possible with too large listings (OSD_OP_LIST)
New features:
- Intelligent recovery/rebalance speed auto-tuning to reduce its impact on clients (see README -> Features)
- Auto-restoration of dead VDUSE daemons in CSI plugin
- Add vitastor-disk update-sb command
- Update QEMU for Debian Bookworm to 8.1 and use it for CSI plugin
Bug fixes:
- Fix pools SOMETIMES staying inactive after stopping a node due to OSDs not reacting
to PG state changes caused by incorrect full reload of state from etcd on reconnection
- Make monitors retry pool configuration changes quickier which fixes them being unable
to apply changes when an ongoing rebalance is quickly making a lot of PGs clean
- Fix CSI plugin not accepting array of strings as etcd address in /etc/vitastor/vitastor.conf
- Allow multiple interfaces with the same IP address, for "simple routed" full mesh network
- Do not ignore loopback addresses for OSD network (to make ECMP setups with frr possible)
- Fix a rare client crash during OSD reconnections
- Only treat data partitions as existing OSDs in vitastor-disk prepare
- Remove etcd parameter from default command examples
- Fix reported free space sometimes changing non-immediately after deletion of data from OSDs
- Fix a possible OSD crash on print_slow when bs_op is NULL
- Use the same etcd_ws_keepalive_interval in mon as in OSD
- Fix mon not using values from config when /config/global is not present
- Remove pve-storage-portal-dns-list format for vitastor_etcd_address
- Parse log_level in cluster_client
- Fix vitastor-nbd image existence check not working because of non-zeroed inode_watch fields
- Do not warn on EPIPE in client unless log_level is raised explicitly
- Fix incorrect error in CSI when searching for the device in /sys
- Remove 2 last prints to stdout in etcd_state_client
- Fix a possible OSD crash when checking corrupted journal entries
Fixes a possible use-after-free in case of continue_ops() calling try_send(),
then connect_peer() -> set_timer() -> trigger_nearest() -> handle_op_part() -> continue_ops() again
Also add protection from etcd watcher messages being split into multiple websocket
messages - I'm not sure if etcd actually does that, but it's better to have extra
protection anyway.
Also check that all etcd watchers are started in the keepalive routine, otherwise
it sometimes tries to revive etcd watchers starting with revision=1 which obviously
always fails because this revision is nearly always compacted.
All these changes should fix an old rarely reproduced bug where SOMETIMES OSDs
didn't react to PG config changes which was leading to offline pools on node reboot.
It happened on the full reload of state from etcd.
New features:
- RDMA without ODP - much faster and all cards are now supported, not just Mellanox
- VDUSE in CSI - faster, more stable and can even recover after CSI pod restart!
- Reserve journal space for stabilize requests dynamically to prevent stalls under load with EC
- Raise default NBD timeout from 30 to 300 seconds and allow to take it from /etc/vitastor/vitastor.conf
- Remove explicit etcdUrl/etcdPrefix K8S storage class parameter support to prevent
etcd migration issues for volumes created with these parameters
- Support QEMU 8.1 and pve-qemu 8.1
Bug fixes:
- Fix RDMA connection (and thus memory) leak
- Fix rare crashes under load due to incorrect io_uring queue size tracking
- Fix monitor statistics aggregation in case of empty /osd/stats keys
- Fix crash on unknown long argument to vitastor-disk
- Allow trailing comma in JSONs again
- Fix crash on attempts to dump a long listing of objects "to stabilize" or "to rollback" in a slow op
Vitaliy Filippov