Vitaliy Filippov
278852b4d5
How many (bugs!) I've knifed, how many I've slit! General note: most bug fixes now include regression tests to verify that they don't repeat in the future. Most bugs fixed in this release were detected by using LLM analysis (Claude Opus/Fable, GPT 5.5). OSD: - Fix OSD hanging with an infinite loop when setting autosync_interval to 0 at runtime - (IMPORTANT) Fix EC PGs hanging in REPEERING when the last final commit/rollback in a batch completes with an error - Limit pg_size by 64 because peering doesn't handle larger values with EC — they just lead to 'incomplete' objects - Fix OSD crash with an "assertion failed" error on EIO retry in snapshot chain read (i.e. when some chunks belong to a corrupted replica with checksum mismatch) - (IMPORTANT) Disable chunked PG count resharding due to possible interference with compaction changes (will be re-enabled after fixes) - (IMPORTANT) Fix incorrect snapshot allocation bitmap recovery during EC chained read - Add on-wire request size validation to prevent possible OOM/DoS/heap corruption on receiving invalid data from the network - (IMPORTANT) Fix parity-less EC writes destroying snapshot allocation bitmaps (i.e. when all parity OSDs in a PG are missing) - (IMPORTANT) Fix EC N+K, K>=2 recovery destroying snapshot allocation bitmaps of live parity chunks - (IMPORTANT) Fix a possible OSD crash during EC misplaced object scrubbing - (IMPORTANT) Verify object bitmap consistency during scrub (only data was checked previously) - (IMPORTANT) Fix corrupted object chunks incorrectly marked as non-corrupted on the second scrub - (IMPORTANT) Fix cached EC decoding of multiple stripes with ISA-L (ISA-L is the default) New store: - Fix a theoretically possible OSD crash on startup when using the previously added workaround for the "double-claim" problem - Remove theoretically possible incorrect metadata block writes during batch EC COMMITs restarted due to a full metadata area - Fix incorrect compaction counter tracking after OSD restart (could probably lead to compaction not restarted correctly after a restart) - (IMPORTANT) Fix some of parallel big_writes possibly not waiting for data fsync, thus not providing durability - Fix possible OSD crash on sync retry when io_uring is full - Fix a possible crash during startup on corrupted on-disk data with too small entry sizes Old store: - Prevent loading extra garbage metadata entries from the last 4 MB of metadata area - Fix read operations possibly crashing if a metadata read (with inmemory_metadata=false) was restarted due to a full io_uring - Fix a possible memory leak of temporary buffers and bitmaps/checksums when a read was restarted due to a full io_uring (reproducible with either inmemory_metadata=false or block_size>256k) - Fix a possible OSD crash during padded checksum reads if buffer count exceeded 1024 (IOV_MAX) (reproducible only with csum_block_size > 4k and block_size >= 4M) - (IMPORTANT) Fix partial padded read journal checksum verification with csum_block_size > 4k - Fix incorrect marking of corrupted objects as non-corrupted after flushing data from journal (with inmemory_journal=false) - (IMPORTANT) Fix deferred freeing of a different block when a block was used by a parallel read - Fix per-inode statistics not being disabled for FS and S3 pools correctly, leading to etcd overload with unneeded per-inode statistics, slower etcd operation, increased memory usage, and too many Prometheus statistics exported by the monitor Both stores: - Fix possibly left garbage in the metadata area if the first OSD startup was interrupted — metadata header is now written only after initializating metadata - Check for short reads during initialization (just in case, doesn't happen in real life) Clients: - Fix write-back queue item split in case when write-back is enabled at runtime - Implement bdrv_detach_aio_context & bdrv_attach_aio_context in the QEMU driver (should fix migration with iothread) - Do not crash on full io_uring in ublk server - Fix missing --readonly option handling in NBD server - Stop gracefully on NBD_CMD_DISC instead of just exit(0) in NBD server - Fix writeback detection in ublk server for --image mode - Limit the amount of incoming data for NFS clients to prevent choking on memory in async mount mode Tools (vitastor-disk/vitastor-cli): - Prevent vitastor-cli merge possibly exiting before completing the last sync/delete operations - Fix vitastor-disk incorrectly validating too large small_write entry length - Fix vitastor-cli merge ignoring input option validation errors - Fix vitastor-cli rm-data always skipping the final fsync - Fix vitastor-disk resize not moving the last used data block - Fix vitastor-disk write-meta incorrectly importing new store small_write entries - Fix vitastor-disk write-journal and write-meta importing old store data incorrectly when csum_block_size is > 4k - Support --io option for vitastor-disk dump-journal/write-journal - Fix vitastor-disk resize crash when converting from very old (0.5.x) metadata - Fix vitastor-disk trim incorrectly rounding block ranges with --discard_granularity option explicitly set to a value > 4k, possibly leading to discarding live data - Fix vitastor-disk write-meta importing new store metadata incorrectly with > 4 GB metadata area size - (IMPORTANT) Fix vitastor-cli modify --resize to a smaller size clearing all image data O_o Other: - Do not crash with an uncaught exception when an invalid /osd/state/ with a non-numeric suffix is present in etcd (in OSD and all client services) - Fix possible crash in vitastor-kv when handling a corrupted DB due to a uint32 overflow - Fix NFS-RDMA memory allocator crashing in some situations - Fix small shared file extend-write potentially reading unallocated memory (NFS) - Add bounds checks to prevent uint32 overflows in NFS/XDR - Re-enable accidentally disabled safety checks (asserts) in files with included cpp-btree - Fix too small memory allocation in NFS portmap
Vitastor
The Idea
Make Clustered Block Storage Fast Again.
Vitastor is a distributed block, file and object SDS, direct replacement of Ceph RBD, CephFS and RGW, and also internal SDS's of public clouds. However, in contrast to them, Vitastor is fast and simple at the same time. The only thing is it's slightly young :-).
Vitastor is 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).
Vitastor targets primarily SSD and SSD+HDD clusters with at least 10 Gbit/s network, supports TCP and RDMA and may achieve 4 KB read and write latency as low as ~0.1 ms with proper hardware which is ~10 times faster than other popular SDS's like Ceph or internal systems of public clouds.
Vitastor supports QEMU, UBLK, NBD, NFS protocols, OpenStack, OpenNebula, Proxmox, Kubernetes drivers. More drivers may be created easily.
Read more details in the documentation. You can start from here: Quick Start.
Talks and presentations
- KuberConf'2025: video
- Highload'2025: video, youtube, presentation (in Russian, in English)
- Highload'2022: presentation (in Russian), video
- DevOpsConf'2021: presentation (in Russian, in English), video
Documentation
- Introduction
- Installation
- Configuration
- Usage
- vitastor-cli (command-line interface)
- vitastor-disk (disk management tool)
- fio for benchmarks
- UBLK for kernel mounts
- NBD - old interface for kernel mounts
- QEMU, qemu-img and VDUSE
- NFS clustered file system and pseudo-FS proxy
- Administration
- Performance
Author and License
Copyright (c) Vitaliy Filippov (vitalif [at] yourcmc.ru), 2019+
Join Vitastor Telegram Chat: https://t.me/vitastor
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.
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 1.1 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-1.1 in the file VNPL-1.1.txt. GPL 2.0 is also included in this repository as GPL-2.0.txt.