Compare commits

..
Author SHA1 Message Date
Vitaliy Filippov b30635b932 Release 3.0.8
- Surrogate increment-only peer IDs are now used instead of file descriptor numbers,
  and client cleanup sequence is reworked in another attempt to fix a possible PG hang
  in 'peering' states. The history of this bug is still unclear and it still hasn't been
  reproduced in tests, but several users have reported it since 3.0.4 (only with enabled RDMA).
- Fix a minor bug where an OSD could skip reporting misplaced PG state when moving
  an object in non-immediate_commit mode.
- Add Ubuntu 26.04 (Resolute Raccoon) support and patches for qemu 10.2 and libvirt 12.0
2026-03-30 02:09:02 +03:00
Vitaliy Filippov 0c1154833c Add Ubuntu 26.04 support 2026-03-30 02:06:10 +03:00
Vitaliy Filippov c227bb05b6 Fix incorrect patch contents for qemu 10.1, add patch for qemu 10.2 2026-03-30 02:06:10 +03:00
Vitaliy Filippov dd85315f22 Update CMake supported versions 2026-03-30 02:06:10 +03:00
Vitaliy Filippov 47d2f4e0be Add patch for libvirt 12.0 2026-03-30 02:06:10 +03:00
Vitaliy Filippov 2a5028d17f Use robin_hood_map in messenger 2026-03-30 02:06:10 +03:00
Vitaliy Filippov 07915c2881 Use surrogate peer IDs instead of FDs 2026-03-30 02:06:10 +03:00
Vitaliy Filippov 79141eb383 Fix commented out //this->pg_state_dirty.insert 2026-03-28 20:08:52 +03:00
Vitaliy Filippov f7cbb6ed56 Remove FD from the event loop in destroy_client, not in stop_client 2026-03-28 14:18:20 +03:00
Vitaliy Filippov 8f8172db99 clear_immediate_ops is now done by stop_client 2026-03-26 01:08:23 +03:00
Vitaliy Filippov 94be147e80 Clear received_ops - they are already freed 2026-03-26 01:08:02 +03:00
Vitaliy Filippov 538620b400 Release 3.0.6
- Fix one more bug in old store checksums O_o - checksum verification wasn't
  working correctly on reads quickly after small initial offsetted writes
  (when a BIG_WRITE was read from the journal).
- Fix clients possibly not destroyed after being switched to RDMA. Not 100%
  verified, but in theory it could rarely lead to PGs hanging in peering/stopping
  states with RDMA enabled.
- Use vitastor-cli in the Cinder driver for modifications instead of direct etcd
  interaction.
- Fix mon_https_ca parsing.
- Fix Docker install.sh, add instructions for Podman.
2026-03-22 22:49:51 +03:00
Vitaliy Filippov 4c34a47179 Fix str_replace 2026-03-19 01:19:29 +03:00
Vitaliy Filippov df16ab627a Fix another bug in old store checksums O_o
Checksum verification wasn't working correctly on reads of small initial offsetted writes
from the journal (until the write was flushed to "clean" metadata).
2026-03-17 22:26:38 +00:00
Vitaliy Filippov 44c895dc30 Fix clients possibly not destroyed after being switched to RDMA
It could in theory lead to hung operations on RDMA client disconnections.

Workflow (probable):
* OSD switches an outbound connection to PEER_RDMA and removes the FD from epoll
* But a receive request in io_uring is still active
* Receive response is never handled, refs remains > 0
* The connection is dropped, but osd_client_t is not destroyed because refs > 0
* Outbound ops are not canceled, suboperations hang, primary operations hang too
2026-03-17 01:04:17 +03:00
Vitaliy Filippov fdea595913 Use vitastor-cli in the Cinder driver for modifications instead of direct etcd interaction 2026-03-15 01:44:30 +03:00
Vitaliy Filippov ef4c91ecc8 Fix mon_https_ca parsing 2026-03-15 01:01:41 +03:00
Vitaliy Filippov 5192c6cf50 Fix Docker install.sh, add instructions for Podman 2026-03-13 20:08:18 +03:00
Vitaliy Filippov cb639a130d Release 3.0.5
A single urgent fix for the new store: metadata was written incorrectly when
filled metadata area exceeded 4 GB due to a simple & stupid 32-bit integer overflow.
This was leading to OSDs corrupting metadata and being unable to start on the next
run.

The bug affected only the new store (meta_format=3) and OSDs with metadata area size
larger than 4 GB. You can check your OSD metadata area size by checking `data_offset`
in `vitastor-disk read-sb /dev/vitastor/osdXX-data` for single-disk OSDs or
`blockdev --getsize64 /dev/vitastor/osdXX-meta` for hybrid OSDs.

If you're affected, it's highly recommended to:
- Install updated packages without restarting OSDs
- Restart OSDs **one by one**
- If a restart of an OSD triggers "has_degraded" rebalance - wait for it to finish
  before restarting another OSD
- If an OSD fails to start with "double-claimed" errors in the log - purge and recreate it,
  and again wait for "has_degraded" rebalance to finish before proceeding to the next OSD
2026-03-07 13:54:44 +03:00
Vitaliy Filippov 892e3a8b6d Add Ubuntu 26.04 build 2026-03-07 13:54:44 +03:00
Vitaliy Filippov 0bd9c26620 Make block_sizes uint64_t to prevent other integer overflow errors in the future 2026-03-07 12:02:57 +03:00
Vitaliy Filippov ae2b1f7802 Some useless code changes for beauty 2026-03-07 00:15:16 +03:00
Vitaliy Filippov 883b2e45da Fix uint32 overflow in the new store :-X 2026-03-06 17:21:22 +03:00
Vitaliy Filippov 09df74cfac Release 3.0.4
New store bug fixes:

- Fix very slow OSD startup (hours) possible in unlucky cases
- Fix incorrect garbage entry validation on start in some cases, preventing OSD startup
- Fix small_write_throttling not functioning and either hanging or crashing OSDs
- Fix "different fact_versions returned from subops" error possible with EC in some cases
- Add a recovery option for the new store: skip_corrupted_meta_entries
- Disable intent writes (even 4k) for hybrid OSDs by default for better write buffering

Other changes:

- Fix chmod/chown of a new empty VitastorFS root directory not working
- Fix vitastor-dd treating all < 32k reads as always empty
- Fix vitastor-disk always setting 32k csum_block_size for HDDs when checksums are enabled
- Fix OSD missing PG state updates and PGs hanging in "starting" in rare cases
- Fix a possible OSD crash with "map::at" message in rare cases when stopping a client
- Fix a possible OSD crash on start when RDMA is available but fails to initialize
- Fix removal and overwriting of totaled objects (objects with all corrupted copies)
- Do not allow data_block_size < 8*bitmap_granularity
- Remove RDMA ODP (On-Demand Paging) support
- Update antietcd to 1.2.4
- Update Docker build to Debian Trixie
- Add packages for AlmaLinux 10
2026-03-03 20:12:31 +03:00
Vitaliy Filippov 2ef3f012c9 Fix Docker builds 2026-03-03 20:09:23 +03:00
Vitaliy Filippov 7175d99c64 Add packages for Alma 10 to docs 2026-03-03 11:17:20 +03:00
Vitaliy Filippov 230a26772e Do not allow data_block_size < 8*bitmap_granularity 2026-03-03 02:09:16 +03:00
Vitaliy Filippov 637684c579 Fix chmod/chown of empty NFS root directory 2026-03-03 02:05:37 +03:00
Vitaliy Filippov c541dd422f Add skip_corrupted_meta_entries docs 2026-03-03 01:52:22 +03:00
Vitaliy Filippov 22c39561cf Add a recovery option for the new store: skip_corrupted_meta_entries 2026-03-02 02:51:09 +03:00
Vitaliy Filippov d4c67b4879 Ignore array-bounds warning in bs_heap 2026-03-02 02:49:56 +03:00
Vitaliy Filippov d1b7167861 Fix vitastor-disk always overwriting csum_block_size with 32k, add default atomic_write_size 0 for hybrid OSDs 2026-03-01 23:47:35 +00:00
Vitaliy Filippov 33b561b73a Fix throttling with the new store 2026-03-01 23:46:30 +00:00
Vitaliy Filippov a95a60c600 Add sudo -E to preserve asan suppressions 2026-03-02 01:48:50 +03:00
Vitaliy Filippov b3bc815354 Sorry, udevadm trigger --settle sometimes just fails in docker :-( 2026-03-02 01:48:50 +03:00
Vitaliy Filippov 42fc45d6da Use iterate_with_stable in list_objects too
Fixes "different fact_versions returned from subops" with EC in some cases
2026-03-02 01:48:33 +03:00
Vitaliy Filippov 3fbe2e7f8a Add more details for "different fact_versions" 2026-03-02 01:48:33 +03:00
Vitaliy Filippov 310c512b43 Correctly delete handle_immediate_ops when stopping a client
Fixes map::at crash in some cases
2026-03-02 01:48:19 +03:00
Vitaliy Filippov fd3e3b4ef0 Update buildenv to bookworm 2026-03-02 01:47:05 +03:00
Vitaliy Filippov 93cf69f89f Merge osd_id.h into object_id.h 2026-03-02 01:46:17 +03:00
Vitaliy Filippov a8dd8bf06c Update antietcd to 1.2.4 2026-03-02 01:45:56 +03:00
Vitaliy Filippov 7e23b57014 Do not save watch revision from initial created: true messages 2026-02-27 01:11:18 +03:00
Vitaliy Filippov b2427836a1 Fix dd treating all < 32k reads as always empty :-X 2026-02-22 14:19:13 +03:00
Vitaliy Filippov 2ed6760447 Remove ODP support 2026-02-17 02:35:46 +03:00
Vitaliy Filippov a1d215ea2f Fix garbage validation, skip garbage entries on load 2026-02-17 01:12:26 +03:00
Vitaliy Filippov 4fa442a5de Sort items on load to fix O(N^2) startup 2026-02-17 01:12:26 +03:00
Vitaliy Filippov 3a057ed5af Update docker to trixie 2026-02-13 01:09:34 +03:00
Vitaliy Filippov 0ad042b28b Add packages for Alma 10 2026-02-12 02:11:06 +03:00
Vitaliy Filippov 4d75c6c8f9 Fix removal and overwriting of totaled objects (objects with all corrupted copies) 2026-02-11 01:24:58 +03:00
Vitaliy Filippov 9ec18f0aa1 Fix double ibv_close_device on RDMA init failure 2026-02-11 00:32:12 +03:00
Vitaliy Filippov facaa2cc6a Release 3.0.3
- Fix csum_block_size > 0 unusable with atomic writes in the new store
  (almost all atomic write requests generated invalid checksums with csum_block_size > 0)
- Fix monitor sometimes randomly failing to optimize PGs with the "problem is
  infeasible or unbounded" message due to not waiting to read full stdout of lp_solve
- Remove reshard_abort optimisation to fix chunked resharding introduced in 3.0.2 possibly
  corrupting in-memory OSD state when handling multiple rapid PG count change requests
- Fix removed inodes not disappearing from OSD statistics in the new store, leading to
  bloating of the statistics with old inodes
- Increase test coverage for the new store and fix several minor bugs:
  - Enabling/disabling of used_for_app was recalculating inode space statistics incorrectly
  - Fix object metadata validation on OSD startup rejecting some correct sequences of entries,
    possibly leading to OSD being unable to start
  - Fix PG activation with EC failing in rare cases with EBUSY when requesting to commit an
    already committed write
  - Fix a theoretically possible metadata writeback issue on ENOSPC during commit
- Fix monitor failing to optimize PGs in presence of a host with name convertible to
  js Number (like 04e278988710) :D
- Fix vitastor-cli dd sometimes (rarely) truncating the image when writing to stdout
- Fix a theoretically possible client connection object leak when io_uring is full
- Fix a leak of RDMA-CM connection objects
- Fix crashes with data_block_size < 32KB (useless setup, but anyway) (#113)
2026-02-08 01:39:57 +03:00
Vitaliy Filippov d2ac8e3827 Fix block writeback on enospc during commit 2026-02-07 16:27:26 +03:00
Vitaliy Filippov d6dacc67db Do not return EBUSY when already stable in the new store 2026-02-07 16:10:39 +03:00
Vitaliy Filippov cbe51595cb Remove reshard_abort 2026-02-07 16:05:40 +03:00
Vitaliy Filippov 6bd0830ab8 Add a test for chunked resharding 2026-02-07 16:02:17 +03:00
Vitaliy Filippov 796e82f34d Check get_compacted_count and get_compact_queue_size 2026-02-07 15:44:54 +03:00
Vitaliy Filippov db39283970 Add a test for complete_lsn_write 2026-02-07 15:36:46 +03:00
Vitaliy Filippov 9f8c686321 Add a test for redirect_intent 2026-02-07 15:09:58 +03:00
Vitaliy Filippov 2618e559d1 Add rollback tests, simplify rollback logic 2026-02-07 15:09:58 +03:00
Vitaliy Filippov 037d2bc162 Add a test for recalc_inode_stats in the new store and fix a bug in it 2026-02-07 14:00:10 +03:00
Vitaliy Filippov f0b64adb32 Fix removed inodes not disappearing from statistics in the new store 2026-02-07 02:48:49 +03:00
Vitaliy Filippov f06d64d879 Fix validate_object commit validation bug 2026-02-06 18:52:37 +03:00
Vitaliy Filippov d164499a1c Add more iterate_compaction test cases 2026-02-06 18:52:19 +03:00
Vitaliy Filippov 725e9aa8ae Remove unused read_locked_entry 2026-02-06 17:38:59 +03:00
Vitaliy Filippov 0715feffa1 Wait for finished read, not for process exit in lp_optimizer 2026-02-04 02:33:08 +03:00
Vitaliy Filippov e9f37e8dc3 Fix NaN PG optimisation error when a hostname has a name convertible to js Number... some shit like 04e278988710 2026-02-04 15:08:03 +03:00
Vitaliy Filippov 4fbe4b5654 Fix BIG_INTENT rechecks with enabled checksums 2026-02-04 02:33:50 +03:00
Vitaliy Filippov 9fb645693b Fix checksums with BIG_INTENT writes in the new store 2026-02-04 02:33:50 +03:00
Vitaliy Filippov 9e47828383 Add a test for big_intent with checksums 2026-02-04 02:33:50 +03:00
Vitaliy Filippov ac2ce48cb2 Fix missing wait for short writes in vitastor-cli dd 2026-02-03 01:28:13 +03:00
Vitaliy Filippov 9cc2beed95 Fix new compile warnings from merged PR 2026-01-30 00:47:58 +03:00
MoonBoi9001andVitaliy Filippov fb1c870f5c Fix: integer division bug with block_size < 32KB (#113)
Test scenarios:
- 128KB aligned I/O (may work despite bug)
- 4KB random I/O (triggers assertion failure)
- 4KB mixed read/write (triggers crash)
- 4KB sequential write (triggers crash)

I accept Vitastor CLA agreement: https://git.yourcmc.ru/vitalif/vitastor/src/branch/master/CLA-en.md
2026-01-29 02:04:30 +03:00
ston3luandVitaliy Filippov 2d616d8058 Fix VLA init in test_heap
I accept Vitastor CLA agreement: https://git.yourcmc.ru/vitalif/vitastor/src/branch/master/CLA-en.md
2026-01-29 02:01:21 +03:00
Vitaliy Filippov 3f7f6f442b Fix some compilation warnings 2026-01-29 02:00:25 +03:00
ston3luandVitaliy Filippov 7e7b95eeb4 Fix void* arithmetic in memset/memcpy
I accept Vitastor CLA agreement: https://git.yourcmc.ru/vitalif/vitastor/src/branch/master/CLA-en.md
2026-01-29 01:59:35 +03:00
ston3luandVitaliy Filippov dd588a0783 Fix ref leak on read path when SQE allocation fails
I accept Vitastor CLA agreement: https://git.yourcmc.ru/vitalif/vitastor/src/branch/master/CLA-en.md
2026-01-29 01:52:07 +03:00
ston3luandVitaliy Filippov 028a6cab68 Clean RDMA entry
I accept Vitastor CLA agreement: https://git.yourcmc.ru/vitalif/vitastor/src/branch/master/CLA-en.md
2026-01-29 01:51:31 +03:00
Vitaliy Filippov d75334ddf0 Release 3.0.2
- Antietcd is now officially safe to use: the release includes a fixed version of antietcd
  which passes Jepsen transaction serializability tests.
- Fix a huge checksum bug in the old store: incorrect checksums for small initial writes.
  The bug affected writes of exactly csum_block_size (4k by default) into new (unallocated)
  objects and generated invalid checksums in the store for the written 4k block. Moreover,
  generation of these invalid checksums was very slow because it was calculating CRC32
  for 4 GB of zeroes. If the block wasn't then overwritten as a part of a larger write
  request it became unreadable even though the stored data was correct. The bug affected
  all versions since 1.0.0, or since 2.3.0 because vitastor-disk didn't allow to enable
  checksums prior to 2.3.0 because of another bug. O:-)
- Prevent OSD disconnections due to long blocking of event-loop caused by PG resharding
  (moving objects between old and new PGs in memory) when changing pool PG count or
  just on restart of an OSD with a large database (for example, with a filled 8 TB SSD).
  The issue should now be fixed because OSD now performs resharding in chunks with pauses
  between chunks.
- Prevent pools stuck in paused state on an aborted PG count change.
- Fix a possible OSD crash with "division by zero" when trying to handle an operation
  before pool PG count is applied to the in-memory store.
2026-01-25 16:08:02 +03:00
Vitaliy Filippov bf0875128e Prevent secondary ops execution on resharded pools 2026-01-25 16:07:51 +03:00
Vitaliy Filippov 9a6a7b7f75 Remove separate storage of PG counts in pg_counts 2026-01-25 12:26:49 +03:00
Vitaliy Filippov c4c17ee6fb Support chunked resharding in OSD 2026-01-25 10:42:48 +03:00
Vitaliy Filippov 2b801a7ffa Support chunked resharding in blockstores 2026-01-25 01:56:29 +03:00
Vitaliy Filippov 233d2b2a09 Resume pool on monitor restart if nothing changed 2026-01-25 01:56:29 +03:00
Vitaliy Filippov 8380d4c6a6 Fix possible FPE in osd_primary 2026-01-23 02:11:16 +03:00
Vitaliy Filippov 73f9c7293f Initialize blockstore after loading pool configuration to pre-shard the DB correctly on start 2026-01-23 01:45:13 +03:00
Vitaliy Filippov 1c66c3e5ba Update antietcd to 1.2.2 2026-01-23 01:45:00 +03:00
Vitaliy Filippov eddfa93c18 Refactor vitastor-cli & fio & qemu-img options in tests 2026-01-16 20:08:13 +03:00
Vitaliy Filippov ddd755a0e6 Fix incorrect checksums for small initial writes in the old store
Details:
- Write size should be exactly csum_block_size (4k by default)
- Write should be made into a new object (unallocated space)
- In this case, the checksum of the block was calculated as if the block was
  padded with extra (2^32 - size) zero bytes due to a simple integer overflow
- As a cherry on the cake, such calculation was 'slightly' slow because it
  was processing almost 4 GB of zeroes for a small write
2026-01-16 01:07:39 +03:00
Vitaliy Filippov 819f5b7ec9 Add a test for an idiotic checksum bug 2026-01-16 01:07:39 +03:00
Vitaliy Filippov 34d0a6d9b1 Allow -1 args 2026-01-16 01:07:34 +03:00
Vitaliy Filippov fe83825ead Release 3.0.1
Important fixes:

- Disable RWF_ATOMIC by default because Linux incorrectly requires all atomic
  writes to be power-of-2-sized and length-aligned. Details: [use_atomic_flag](https://vitastor.io/en/docs/config/osd.html#use_atomic_flag)
- Fix cross-pool snapshots not working at all - always reading old data after taking the snapshot
- Fix level_placement (broken in 2.2.0)
- Fix CAS write return values in the client library (broken in 2.4.4), also breaking unaligned writes in VitastorFS
- Fix VitastorFS possibly losing some of intersecting parallel unaligned writes
- Prevent possible reads of the old data during unfinished intent writes in the new store
- Tests added for all of above problems to prevent future regressions

Other fixes:

- Allow to specify OSD tags and weights during prepare
- Only clear the first block instead of whole OSD metadata and journal areas during vitastor-disk
  prepare - OSD anyway clears them on the first run
- Fix vitastor-cli dd non-seekable detection and error status
- Do not stop OSD on zero-copy tcp short send errors - these aren't a bug, but just caused
  by TCP client disconnections
- Fix modify-osd not working after deleting the OSD configuration key from etcd
- Fix regular antietcd websocket disconnections in OSD due to lack of WS_PONG responses to WS_PING
- Reduce the number of allocated RDMA memory regions
- Allow routed RoCE
2025-12-22 02:17:43 +03:00
Vitaliy Filippov 8ec7faa675 Disable use_atomic_flag by default 2025-12-22 01:52:07 +03:00
Vitaliy Filippov 21cf5c8815 Fix optimized RMW at non-zero shared inode offset too 2025-12-21 12:05:06 +03:00
Vitaliy Filippov 44eeb1ed13 Do not read incomplete intent writes 2025-12-21 11:49:56 +03:00
Vitaliy Filippov cc6c445cf0 Add unaligned_write test for small file 2025-12-21 02:39:06 +03:00
Vitaliy Filippov bd6af0db09 Remove unused end_pad 2025-12-21 02:39:06 +03:00
Vitaliy Filippov 8860101e99 Fix optimized RMW in shared inodes at zero offset after disabling skip 2025-12-21 02:22:34 +03:00
Vitaliy Filippov c92661b364 Fix vitastor-cli dd non-seekable detection and error status 2025-12-21 01:33:57 +03:00
Vitaliy Filippov 9a02a592e3 Do not skip RMW at the end of the file - skipping it may lead to data corruption (fix #116) 2025-12-20 18:34:20 +03:00
Vitaliy Filippov 8b7fa3d3bc Add a test for parallel unaligned appends to a file 2025-12-20 18:34:08 +03:00
Vitaliy Filippov db5eaa2eee Fix CAS write return value (broken in 2.4.4) 2025-12-20 18:33:45 +03:00
Vitaliy Filippov d35727dbb7 Add test_level_placement and fix it with the 2.2.0+ PG "folding" system 2025-12-15 14:19:30 +03:00
Vitaliy Filippov b78f526696 Clear only the first OSD metadata and journal blocks - it anyway clears the rest 2025-12-14 14:44:27 +00:00
Vitaliy Filippov a23df12260 Fix modify-osd for the case when OSD config is deleted from etcd 2025-12-14 14:42:50 +00:00
Vitaliy Filippov 166e16102e Allow to specify tags and weights during prepare 2025-12-14 17:30:15 +03:00
Vitaliy Filippov 06c602110c Change "BUG" short send message to a client disconnection 2025-12-14 17:15:16 +03:00
Vitaliy Filippov 1de68c30af Fix cross-pool snapshots and fix the test for it 2025-12-13 18:20:16 +03:00
Vitaliy Filippov 2a0aca6e94 Fix Proxmox supported version note 2025-12-13 17:52:47 +03:00
Vitaliy Filippov e808332e12 Reply to WS_PING with WS_PONG to fix OSD websocket disconnections 2025-12-12 18:11:21 +00:00
Vitaliy Filippov fc5a183959 Allow routed RoCE 2025-12-12 15:06:49 +00:00
Vitaliy Filippov 55de37e58a Use a single memory region per client instead of separate mrs per each malloc 2025-12-12 15:05:33 +00:00
Vitaliy Filippov aacfdf0dec Rename try_recv_rdma to init_recv_rdma 2025-12-07 21:06:36 +03:00
Vitaliy Filippov a67c415e0f Release 3.0.0
A single new feature: the new log-structured metadata store implementation, described
in the presentation from Moscow Highload'2025 (check it out [here](https://vitastor.io/en/docs/#talks-and-presentations)).

It's now the default store for new OSDs. The support for the old store is also left in place,
you can still choose it for new OSDs with `vitastor-disk prepare --meta_format 2`.

OSDs from previous versions with the old store format will also continue to operate just like before.
2025-12-06 02:12:00 +03:00
Vitaliy Filippov 495f3fb4cd Remove libvitastor_blk from rpms 2025-12-06 01:48:56 +03:00
Vitaliy Filippov 934752d617 Add define RWF_DSYNC 2025-12-06 01:24:19 +03:00
Vitaliy Filippov 1e6e233426 More documentation 2025-12-05 02:20:31 +03:00
Vitaliy Filippov a5768a8ef6 Parse empty meta_format in superblock as v2 (fix updating from 2.x) 2025-12-02 01:52:42 +03:00
Vitaliy Filippov 216707f101 Add new parameter docs: meta_format, atomic_write_size, use_atomic_flag 2025-12-02 01:52:42 +03:00
Vitaliy Filippov 8ffdb93ed3 Implement test_resize_auto for the new (lsm/heap) store 2025-12-02 01:52:42 +03:00
Vitaliy Filippov 2fb7022c78 Fix OLD=1 ./test_resize 2025-12-02 01:52:42 +03:00
Vitaliy Filippov 2633978fec Duplicate tests for old store 2025-12-02 01:52:41 +03:00
Vitaliy Filippov 615d4825c0 Remove FIXME 2025-12-02 01:52:12 +03:00
Vitaliy Filippov e55ca26ff6 Reject all-zero blocks, mark empty space with a special entry type during init 2025-12-02 01:52:12 +03:00
Vitaliy Filippov abe093b9a3 Reserve for max_entry_size, not for min_entry_size 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 07e6eb0b16 Return no_inode_stats feature back 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 0eacce1e1d Crazy shit: optimize inode maps for very small item counts (i.e. inodes with only 1-15 objects) 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 3df410acc7 Fix test_scrub with BIG_INTENTs 2025-12-02 01:52:12 +03:00
Vitaliy Filippov e55927076c Abort compaction if overwritten 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 60fcb168fe Fix assert 2025-12-02 01:52:12 +03:00
Vitaliy Filippov b17691ba02 Add a redirect_intent write mode for non-capacitor SSDs 2025-12-02 01:52:12 +03:00
Vitaliy Filippov f2cbe793e2 Fix intent_write_allowed check, add offset & len validation on start 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 0053546f8b Remove zerofill in blockstore (an old and unused artifact) 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 959f792f82 Auto-guess atomic_write_size from /sys 2025-12-02 01:52:12 +03:00
Vitaliy Filippov eaff4509ca Complete small/intent writes explicitly, not implicitly on metadata block write 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 04eefce30b Fix recheck logging 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 746844d301 Add double-claim block check 2025-12-02 01:52:12 +03:00
Vitaliy Filippov c8f5b6cb19 Fix submitting forced fsyncs in flusher 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 0f4837e9bb Fix test_resize 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 25ce82a729 Fix checksum flushing for the case of empty small writes, fix checksum calculation skipping during read 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 944499135f Reduce hashmap memory usage by 8 bytes per key :) 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 08f21edaf7 Fix enospc detection logic and test_enospc 2025-12-02 01:52:12 +03:00
Vitaliy Filippov e9f0639e62 Fix iterate_compaction for the case of unfinished commit/rollback 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 4e0b203552 Fix iterate_compaction for commit with non-latest version 2025-12-02 01:52:12 +03:00
Vitaliy Filippov b772a3cd04 Fix freeing repeated big_intents 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 856ad79a02 Fix mark_completed_lsns for enabled fsyncs 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 39a8772d7f Fix loading 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 993f40de37 Utility to test atomic writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 5607222921 Revert f9554a02e4b75e87c5b44085b68f84e69100c9d8
Do not "Limit the number of unstable versions per object" - LSMeta doesn't need it
2025-12-02 01:52:12 +03:00
Vitaliy Filippov 378fff6f67 Write recheck_modified_blocks to disk 2025-12-02 01:52:12 +03:00
Vitaliy Filippov daf2cc3fb1 Validate objects on start 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 17d61c5868 Fix disk_tool for lsmeta 2025-12-02 01:52:12 +03:00
Vitaliy Filippov e709657de4 Save completed_lsn in the superblock 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 3eecf9048c Simplify compaction 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 1852caaeec Implement even more simplified big_intent writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 7b40561141 Fix GC and loading 2025-12-02 01:52:12 +03:00
Vitaliy Filippov d80c12ced2 Prevent parallel modifications of the same object in fio_bs 2025-12-02 01:52:12 +03:00
Vitaliy Filippov b9713deecd Pass bitmap in fio_bs 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 852734270e wip 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 8f92979a18 Move prev/next/block_num to struct heap_list_item_t 2025-12-02 01:52:12 +03:00
Vitaliy Filippov b720af74c2 WIP experimental - double linked list 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 96df2966cc Reserve block entries 2025-12-02 01:52:12 +03:00
Vitaliy Filippov d3b171e047 FIXME 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 7287b7fc25 Remove std::set meta_nearfull 2025-12-02 01:52:12 +03:00
Vitaliy Filippov e1bb670491 Apply inflights in batches 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 1e5a01def8 Another experiment: store writes in mallocs 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 371e630f52 Recheck repeat using map 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 0a7ae616f3 Do not use pending_modified_blocks to check 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 82f5fb7edd Remap without intermediate hashmap 2025-12-02 01:52:12 +03:00
Vitaliy Filippov ec8527c89d Experimental: move prev_count to the map itself 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 5afef7ca6d WIP Integrate LSMeta :-) 2025-12-02 01:52:12 +03:00
Vitaliy Filippov e512e1eeb1 LSMeta 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 7ab60c00ab Use 32-bit big write location (OK for up to 512 TB OSDs) 2025-12-02 01:52:12 +03:00
Vitaliy Filippov ac8e0ef231 Do not store offset & len in big_writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 94f3634602 Fix object crc32c calculation 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 2b8a9e3f90 Add entry_type to heap_object_t too
This is required to:
1) later inline the last "big_write" entry into the object to slightly reduce memory usage
2) eliminate an ugly hack where entry type is determined by its size
3) make the storage scheme extensible i.e. when adding new entry types
2025-12-02 01:52:12 +03:00
Vitaliy Filippov ef792608b0 Rename flags to entry_type 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 04531bcfbb Use robin_hood::unordered_flat_map - it has 1 byte overhead instead of 8 byte 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 1b40fa1cee Revert try_get 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 3ea9230ed0 WIP dump/load heap 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 97dfbfad75 Add missing list_stable_limit support 2025-12-02 01:52:12 +03:00
Vitaliy Filippov c148f97ee4 Fix checksum validation in !inmemory_journal mode 2025-12-02 01:52:12 +03:00
Vitaliy Filippov b1f61eb5c8 Fix checksum padding during read 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 9db8748647 Remove unused bs->buffer_area + buffer calc 2025-12-02 01:52:12 +03:00
Vitaliy Filippov e747319c1e Add missing memset zero_object to 0 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 0cb0e31ceb Mark in-memory data reads as SKIP_CSUM 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 0703efd8b9 Add missing calc_crc32c after updating block checksums 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 4a0720a231 Fix checksum padding during flush 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 57d83ecf7c Fix free_read_buffers 2025-12-02 01:52:12 +03:00
Vitaliy Filippov d4f7bbb412 Fix assert(region_marker) at the end of the block in find_block_run() 2025-12-02 01:52:12 +03:00
Vitaliy Filippov e5e71fc21a Fix assert(is_buffer_area_free) with size=0 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 0ec4f1608a Add a copy of wyhash 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 232e416658 Fix test build with isa-l 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 23d5fec580 Remove assert !region_marker & FREE 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 16881a3d6b Use wyhash 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 7cde5c75b9 Change emhashes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 4c32244409 unordered_map mvcc 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 30c5a79772 Fix bad resharding due to the lack of iteration order in a hashmap 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 6e856c2719 Use emhash::try_get 2025-12-02 01:52:12 +03:00
Vitaliy Filippov af710d3c65 Use emhash hashmap (2x faster) 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 8189c3a4ef Use unordered_maps for object-block index 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 552ba5d885 B-tree is slower... 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 14bbf18ede Unordered_map for pool settings 2025-12-02 01:52:12 +03:00
Vitaliy Filippov c3eaaa4b94 Fix read fio bench 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 8b35c09e12 Fix vitastor-disk prepare and param validation 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 74b8ea1303 Fix skipping of corrupted objects, fix use_buffer_area with zero size 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 6a30e6653a Support heap format in simple-offsets 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 2d7da127ad Return all unstable versions in listing 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 4143d56db7 Actually fsync after stabilize 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 5067554e46 Move test & build_test to top-level cmakelists 2025-12-02 01:52:12 +03:00
Vitaliy Filippov bc48eb1ff8 Add missing request_trim 2025-12-02 01:52:12 +03:00
Vitaliy Filippov b0809b33aa Use multilist_index_t instead of multiple bitmap allocators 2025-12-02 01:52:12 +03:00
Vitaliy Filippov d61cf2303f Implement another multilist-style allocator for metadata blocks 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 712f22d6f8 Fix a bug with unstable_big over unstable_small 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 4340082315 Limit the number of unstable versions per object 2025-12-02 01:52:12 +03:00
Vitaliy Filippov a1a449686a Integrate moving objects 2025-12-02 01:52:12 +03:00
Vitaliy Filippov fb87870734 Support moving objects between blocks 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 99bddb976a Fix zero-length writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 4f997791b8 Fix op_stable slowdowns 2025-12-02 01:52:12 +03:00
Vitaliy Filippov c980168d10 Fix a bs_read bug 2025-12-02 01:52:12 +03:00
Vitaliy Filippov dce4a6c37a Fix test dependencies 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 38d5175f66 Add read_blocks() API 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 7569fb959b Fix block checksum calculation in write_journal for the old blockstore version 2025-12-02 01:52:12 +03:00
Vitaliy Filippov aa1e62a502 Remove block_order parameter 2025-12-02 01:52:12 +03:00
Vitaliy Filippov dcbdb0ae33 Extract (flags & BS_HEAP_TYPE) into a function 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 6e5f990801 Make dump-journal --format data default 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 845e76a0ec Add fio options 2025-12-02 01:52:12 +03:00
Vitaliy Filippov a98aee7906 Fix loading for out-of-order lsns 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 064a94166c Add include for older gcc 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 38112e9012 Fix vitastor-disk build (with old metadata and journal formats) 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 353460cc83 Collapse intent_writes on other write types too 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 92631bb6b3 Fix buffer overflow in test_heap 2025-12-02 01:52:12 +03:00
Vitaliy Filippov ce050e7eda Add a test for parallel reads with block checksums 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 556fc9a876 Remove cancel_all_writes
Not needed because a) parallel writes to the same object are forbidden
b) subsequent writes don't depend on previous ones anyway.
2025-12-02 01:52:12 +03:00
Vitaliy Filippov 274f9ecda5 Add a test for "perfect_csum_update" mode 2025-12-02 01:52:12 +03:00
Vitaliy Filippov e0d2705294 Add 2 tests for intent writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 57d2f30303 Add a test with fsync 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 7f4c541a6f Implement buffered disk_mock_t mode, extract ringloop_mock.cpp 2025-12-02 01:52:12 +03:00
Vitaliy Filippov b0b495a991 unaligned_intent does not need special handling anymore 2025-12-02 01:52:12 +03:00
Vitaliy Filippov da8bf2b73b Add an option for global coverage 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 615e9d1274 Fix some blockstore bugs discovered by the mocked test! 2025-12-02 01:52:12 +03:00
Vitaliy Filippov ec5e93307d Add a basic mocked blockstore test 2025-12-02 01:52:12 +03:00
Vitaliy Filippov b599334c4a Add mocks for blockstore integration tests: timerfd, ring_loop_mock_t and disk_mock_t 2025-12-02 01:52:12 +03:00
Vitaliy Filippov bcde273ca1 Test calc_checksums 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 87fe1bc00f Check 2/3 blocks in test_recheck 2025-12-02 01:52:12 +03:00
Vitaliy Filippov d4c465f786 Fix intent writes with padded checksums 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 62995243f3 Fix reads from intent writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 7ea4884ef6 Disable punching block checksums and allow to enable it with a parameter 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 8823ddf48e Process big_writes as intents to avoid fsyncs 2025-12-02 01:52:12 +03:00
Vitaliy Filippov db14037ac8 Fix collapsing intent-over-intent checksums 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 64db505357 Allow 1 intent_write over big_write in fsync mode 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 8992eb57df Block lists by previous writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov f3048d0858 Allow multiple writes with the same version 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 714b0783bf Add tests for intent writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov e8e2aa5dba Call finish_load after async recheck 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 9063bcaa41 More tests for incorrect data cases 2025-12-02 01:52:12 +03:00
Vitaliy Filippov c392e914e2 Do not block writes on previous writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 249e04ac0d Add atomic_write_size parameter 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 16dee7c136 Use ui32 for block sizes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov adddf9b3b1 Remove extra unneeded read_entry-s 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 6a5044ae36 Do not recheck data location on intent-write 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 49407afa17 Remove FIXMEs 2025-12-02 01:52:12 +03:00
Vitaliy Filippov a00fc1bc24 Fsync after stabilizing 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 3a96d41c93 Fsync & update metadata when block checksums are enabled 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 2e4d5ae5bb Return new_lsn from erase and rollback 2025-12-02 01:52:12 +03:00
Vitaliy Filippov db458fc999 Allow to cancel compaction for unfinished writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 574520be0f Fix space allocation & compaction on start 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 715e7df51f Use inflight_lsn iterators 2025-12-02 01:52:12 +03:00
Vitaliy Filippov ea5ee0c46f Free space on overwrites correctly 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 47bec8af47 Use fsynced_lsn in flusher 2025-12-02 01:52:12 +03:00
Vitaliy Filippov ec3bf4ae6c Correctly track fsynced_lsn when fsyncs are enabled 2025-12-02 01:52:12 +03:00
Vitaliy Filippov cb45b1865d Extract multilist_alloc_t 2025-12-02 01:52:12 +03:00
Vitaliy Filippov f656545f4a Check for overlaps during blockstore loading 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 0241a61412 Fix multilist_alloc_t bug, move verify and print to lib 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 0bc81f5320 Fix pending_ops 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 84961f6d0a Fix repeating cur_oid 2025-12-02 01:52:12 +03:00
Vitaliy Filippov cb085f9c8f Mark overwritten heap_writes as immediately compacted 2025-12-02 01:52:12 +03:00
Vitaliy Filippov a9773b1908 Move "ack write" debug message 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 11783a2d7a Fsync data on trim_lsn, not when writing compacted data 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 7915605609 Batch big_write data fsyncs 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 3ba3eed0cf Use the same "inflight" queue to track compaction 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 5dc0b42146 Use new LSNs on stabilize 2025-12-02 01:52:12 +03:00
Vitaliy Filippov b44c3a7971 Assign new LSN on stabilize 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 59b7b2e0c3 WIP Only save MVCC copy when overwriting an object 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 7f0c78113b Do not use wr_offset 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 20bbeb4095 Prevent compaction of incomplete object writes 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 3c687a2993 Experimental INTENT_WRITE write mode with WA=2 instead of 3 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 7530bdbec7 Add test_compact_block 2025-12-02 01:52:12 +03:00
Vitaliy Filippov c78b4d184d Remove alloc_buffer_area 2025-12-02 01:52:12 +03:00
Vitaliy Filippov c7a6b77c21 Two more unordered_maps 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 3cb7ec69bc Remove sync_to_repeat map and use simpler repeating 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 954e7b658d Use a sequence of bitmap_allocs for metadata instead of std::sets... 2025-12-02 01:52:12 +03:00
Vitaliy Filippov 2c6ea8a521 It seems tcmalloc is actually slower, disable it 2025-12-02 01:52:11 +03:00
Vitaliy Filippov bfd5575425 Use single add_used_space call instead of unmark+mark allocated_block 2025-12-02 01:52:11 +03:00
Vitaliy Filippov 85e61c9c31 Remove compact_queue_lsn map 2025-12-02 01:52:11 +03:00
Vitaliy Filippov f580cee936 Implement a really crazy "multi-linked-list" allocator for buffered data 2025-12-02 01:52:11 +03:00
Vitaliy Filippov 6d0460500a Use linked list heap in blockstore code
WIP, still slower than the old version :-E
2025-12-02 01:52:11 +03:00
Vitaliy Filippov 01ae800d34 Use linked lists in heap to avoid excessive memory copying 2025-12-02 01:52:11 +03:00
Vitaliy Filippov dbb885a6b1 v1 (old) store fixes for merged version 2025-12-02 01:52:11 +03:00
Vitaliy Filippov 8ff2c268f7 Integrate "heap" metadata storage into blockstore 2025-12-02 01:52:11 +03:00
Vitaliy Filippov 8430104c19 "Heap" metadata storage scheme 2025-12-02 01:52:11 +03:00
Vitaliy Filippov 4c11e3ad3d Release 2.4.4
- Implement `vitastor-kv rescue` command which dumps all readable blocks from the
  database regardless of the tree structure and allows to recover a corrupted database.
- Simplify primary OSD selection - use just PG number instead of the hash. This allows
  to distribute primary OSDs more evenly and reach better read speeds.
- Fix vitastor-kv-stress hang on parallel list & reopen.
- Fix CAS writes without immediate_commit. Fix is sufficient for vitastor-kv to
  operate correctly in non-immediate_commit pools.
- Fix NFS ACCESS calls for broken inodes (allow root access to them).
- Fix broken vitastor-kv-stress --runtime parameter.
- Fix bad vitastor-kv dump output buffering leading to incorrect JSON in dumps.
- Fix writing garbage from uninitialized memory to the empty space in VitastorKV DB.
- Refactor some code in preparation to the release of the new log-structured metadata store.
- Replace assert(done != expected) error message, probably caused by kernel issues with
  zero-copy TCP send, with a more concise error description.
2025-12-01 02:22:27 +03:00
Vitaliy Filippov d88b49872b Fix vitastor-kv-stress hang on parallel list & reopen 2025-11-30 16:54:23 +03:00
Vitaliy Filippov ca27b91919 Handle CAS writes with built-in SYNC and without retries in the client 2025-11-30 14:50:50 +03:00
Vitaliy Filippov 94f31b96b8 Cancel block changes on write failures/CAS retries (should fix #115) 2025-11-30 01:49:49 +03:00
Vitaliy Filippov 76c7c26d32 Fix broken inode removal again (ACCESS will now allow root access to such inodes) 2025-11-29 19:30:20 +03:00
Vitaliy Filippov 8aa2c49202 Use immediate_commit in test_nfs 2025-11-29 19:22:01 +03:00
Vitaliy Filippov 0f330b10f1 Add kv-stress test to CI 2025-11-29 16:48:38 +03:00
Vitaliy Filippov 477b54a0d8 Fix broken vitastor-kv-stress --runtime parameter 2025-11-29 16:47:13 +03:00
Vitaliy Filippov 5823a7de66 Disable K/V DB operation on non-immediate_commit pools (unsafe because of broken CAS) 2025-11-29 16:43:46 +03:00
Vitaliy Filippov eb0deaa3f5 Implement vitastor-kv rescue command 2025-11-29 16:34:34 +03:00
Vitaliy Filippov 59e6527303 Fix dump buffering, do not write uninitialized memory to the end K/V blocks on serialize 2025-11-29 16:34:34 +03:00
Vitaliy Filippov 67ba9f9b7c Simplify primary OSD selection - use just PG number instead of the hash 2025-11-28 16:38:55 +03:00
Vitaliy Filippov 3ad83e8d13 Use only space_left, not sqes_left 2025-11-23 15:17:57 +03:00
Vitaliy Filippov 5d3f3f47a7 Move all #defines to internal.h 2025-11-23 14:53:46 +03:00
Vitaliy Filippov 8ee7058ec8 Use an interface instead of explicit blockstore_t wrapper 2025-11-23 01:48:44 +03:00
Vitaliy Filippov 1badc6ad13 Remove include of blockstore_impl.h from disk_tool and osd 2025-11-23 01:48:44 +03:00
Vitaliy Filippov be1858848e Remove block_order and replace << >> by / * data_block_size 2025-11-22 18:10:32 +03:00
Vitaliy Filippov d75b1cb2d2 Refactor some defines 2025-11-22 18:07:44 +03:00
Vitaliy Filippov a1c17d90a3 Add done != expected send error message 2025-11-18 01:23:52 +03:00
Vitaliy Filippov f0112050ce Release 2.4.3
- Daemonize before forking in NFS proxy to fix OSD RDMA support (#107)
- Fix possible PG_INCOMPLETE on node outage when allow_net_split is false
- Fix build under Ubuntu 25.10
- Fix journal_no_same_sector_overwrites mode after optimizing WA (#109)
2025-11-09 02:18:35 +03:00
Vitaliy Filippov d6b8d921d6 Daemonize before forking in NFS proxy to fix OSD RDMA support (fix #107) 2025-11-09 01:07:11 +03:00
Vitaliy Filippov 65872f5d0e Remove RDMA context FD handlers during msgr destroy 2025-11-08 14:11:03 +03:00
Vitaliy Filippov 15eef27d44 Fix PG_INCOMPLETE on node outage when allow_net_split is false 2025-11-07 21:56:56 +03:00
Vitaliy Filippov aa1e51de5f Add include stdint (fix ubuntu 25.10 build) 2025-11-07 21:56:56 +03:00
Flynn049andGitHub c164adb43c Fix: journal sector written flag not set when submitted (#109) 2025-11-07 00:59:04 +03:00
Vitaliy Filippov 622631c146 Release 2.4.2
A micro-fix to the make-etcd script and a minor fix to vitastor-cli rm
leaving "deleted" flag after itself when using "inverse delete".
2025-11-04 17:51:09 +03:00
Vitaliy Filippov 4ab93d8481 Use tar.xz from bookworm for trixie O_o 2025-11-04 17:51:09 +03:00
Vitaliy Filippov bd64770317 Use -it in debian build scripts 2025-11-04 16:56:04 +03:00
Vitaliy Filippov 34cb48d553 Support QEMU 10.1 2025-11-04 15:39:10 +03:00
Vitaliy Filippov 724d2ffa04 Do not mark "inverse child" as deleted during rename-delete 2025-11-04 01:30:00 +03:00
Vitaliy Filippov b6bfe1435d Followup fix for make-etcd ipv6 support 2025-11-03 16:18:38 +03:00
Vitaliy Filippov 74a23dcb63 Release 2.4.1
- Optimize WA - reduce it from 4-5 to 3.x by batching journal writes
- Fix ls-pool showing wrong used % when osds are reweighted (#105)
- Fix vitastor-cli rm sometimes hanging when PGs are down (#99)
- Fix vitastor-cli ls displaying raw used space for all inodes when -p is specified (#101)
- Fix "Cannot set property of undefined" in monitor when deleting pool with live inode op stats (#103)
- Revert vitastor-kv change from 2.2.0 leading to crashes in vitastor-kv-stress (#100)
- Add pg_size change check for XOR pools
- Fix root access to VitastorFS files
- Add IPv6 support to make-etcd
- Fix IPv6 etcd address support in OSD & client library
2025-10-29 01:39:33 +03:00
Flynn049andVitaliy Filippov 93fd23b2bb Fix ls-pool showing wrong used % when osds are reweighted (#105) 2025-10-28 21:01:46 +03:00
Vitaliy Filippov eedc700b83 Do not try executing operations on inactive PGs at all 2025-10-28 01:38:54 +03:00
Vitaliy Filippov c8cc17dbe9 Fix tv_nsec == 1000000000 case in timerfd_manager (#99) 2025-10-27 20:57:29 +03:00
Vitaliy Filippov b55d406386 Fix vitastor-cli ls displaying raw used space for all inodes when -p is specified (fix #101) 2025-10-27 20:55:17 +03:00
Vitaliy Filippov 0c46dbd333 Fix "Cannot set property of undefined" in monitor when deleting pool with live inode op stats (should fix #103) 2025-10-27 20:48:07 +03:00
Vitaliy Filippov ed94aa52cf Ensure that the retry timer is set when checking PG retry (for #99) 2025-10-27 20:19:17 +03:00
Vitaliy Filippov 555ae613c2 Revert commit 2d07449e74 (fixes #100) 2025-10-27 20:12:11 +03:00
Vitaliy Filippov cad6ea0360 Add pg_size change check for XOR pools 2025-10-24 20:53:27 +03:00
Vitaliy Filippov d60709dce1 Init list_retry_time to 0 2025-10-20 20:32:49 +03:00
Vitaliy Filippov a03ffd0d73 Fix prefix in openstack doc 2025-10-14 01:47:05 +03:00
Vitaliy Filippov 89b76a87b6 Fix root access to VitastorFS files 2025-10-09 19:24:05 +03:00
Vitaliy Filippov ceba343ac0 Add [] for ipv6 etcd IPs in make-etcd 2025-10-08 16:39:48 +03:00
Vitaliy Filippov 3bc04d8250 Fix ipv6 parsing in make-etcd 2025-10-08 01:33:11 +03:00
Vitaliy Filippov d228fbfb68 Fix ipv6 port parsing in addr_util 2025-10-08 01:30:00 +03:00
Vitaliy Filippov e3c8fd28b4 Allow to test EC writes with fio_blockstore 2025-09-27 14:12:24 +03:00
Vitaliy Filippov d87e7d1a37 Optimize WA - do not move to the next journal sector until the actual write 2025-09-27 11:23:05 +03:00
Vitaliy Filippov 59f87c3e30 Release 2.4.0
New features:

- Support UBLK in CSI and make it the default
- Add image tree output: `vitastor-cli ls --tree`

Bug fixes:

- Fix OSDs crashing on localized reads from snapshotted images
- Several NFS fixes from github PR #95
- Fix snapshotted bitmap reads not working in EC pools (fix #92)
- Fix UBLK reporting incorrect device size (1/8 of actual size), report logical_block_size=4096
- Do not repeer when stopping PEER_CONNECTING osd_clients (increase PG peering stability on failover)
- Fix marking PGs as incomplete when peers are unavailable
- Fix FS formatter (mkfs and fsck) error handling in CSI (it didn't report mounting errors)
- Fix removal of block PVC devices not working in CSI
- Do not attempt to free outbound ops in clear_immediate_ops (fix #88)
- Add vitastor-disk prepare --dry-run option
- Add librdmacm-dev to build-deps
- Log has_invalid objects
- Do not warn on incomplete+has_invalid PG states as unexpected
- Fix OSD reweight values between 0 and 1 not working in monitor
- Fix OSD assertion failed: n_subops != sent when all object copies are corrupted
- Install ibverbs-providers in vitastor Docker builds
- Fix #86 - base64_decode on arm64 O_o
- Fix Proxmox 9.0 support (remove some kind of a whitelist added in upstream)
2025-09-24 15:40:41 +03:00
Vitaliy Filippov eba383f66f Fix warn_on_unused_result in kv_cli 2025-09-24 15:40:19 +03:00
Vitaliy Filippov 4e5e8822c0 Fix invalid object log 2025-09-24 15:33:21 +03:00
Vitaliy Filippov 60933c1d00 Fix local_reads with chain reads 2025-09-24 15:02:41 +03:00
ston3luandVitaliy Filippov 1ad6933953 Fix uninitialized variable in nfstime
https://github.com/vitalif/vitastor/pull/95

By submitting this pull request, I accept Vitastor CLA
2025-09-24 15:02:41 +03:00
ston3luandVitaliy Filippov 8a250f4fca add proper handling of '.' and '..' in LOOKUP procedure
https://github.com/vitalif/vitastor/pull/95

By submitting this pull request, I accept Vitastor CLA
2025-09-24 15:02:41 +03:00
ston3luandVitaliy Filippov 94ddf20667 fix: refresh mtime after remove file
https://github.com/vitalif/vitastor/pull/95

By submitting this pull request, I accept Vitastor CLA
2025-09-24 15:02:41 +03:00
ston3luandVitaliy Filippov 5f18496c04 change delete operation to set for restoring direntry in nfs_kv_remove
https://github.com/vitalif/vitastor/pull/95

By submitting this pull request, I accept Vitastor CLA
2025-09-24 15:02:41 +03:00
Vitaliy Filippov 08a3dcd587 Prevent vector out-of-bounds in kv alloc_block (from #95) 2025-09-24 15:02:41 +03:00
Vitaliy Filippov 3c5b9d2744 Support vitastor-cli ls --tree syntax 2025-09-24 13:09:43 +03:00
Vitaliy Filippov cff08d2c72 Fix snapshotted bitmap reads not working in EC pools (fix #92) 2025-09-24 02:33:04 +03:00
Vitaliy Filippov 1e1f395947 Do not repeer on stopping PEER_CONNECTING osd_clients 2025-09-24 02:17:34 +03:00
Vitaliy Filippov e6c2628960 Fix marking PGs as incomplete when peers are unavailable 2025-09-24 02:14:47 +03:00
Vitaliy Filippov 887f7c1530 Remove separate close_sync in nbd_proxy, too 2025-09-24 02:11:32 +03:00
Vitaliy Filippov 2c6bddd831 Support UBLK in CSI and make it default 2025-09-24 02:09:39 +03:00
Vitaliy Filippov e1715c33bb Fix warn_unused_result on write/read in nbd and ublk 2025-09-24 02:07:58 +03:00
Vitaliy Filippov 2ef80bf0b8 Remove separate close_sync in ublk_proxy, it's already inside cli->flush 2025-09-24 01:54:43 +03:00
Vitaliy Filippov 85ba710718 Add CSI Dockerfile for build with local packages 2025-09-24 01:50:04 +03:00
Vitaliy Filippov c16b0e7f92 Fix ublk device size & logical_block_size 2025-09-24 01:42:14 +03:00
Vitaliy Filippov b3d388228a Fix formatter error handling in CSI 2025-09-23 02:43:21 +03:00
Vitaliy Filippov bcde9de7da Fix the removal of block PVC devices not working 2025-09-22 10:29:03 +03:00
Vitaliy Filippov 52bc3261e9 Do not attempt to free outbound ops in clear_immediate_ops (fix #88) 2025-09-17 12:22:48 +03:00
Vitaliy Filippov 2d42f29385 Add vitastor-disk prepare --dry-run option 2025-09-05 23:34:11 +00:00
Vitaliy Filippov 17240c6144 Add librdmacm-dev to build-deps 2025-09-06 02:09:44 +03:00
Vitaliy Filippov 9e627a4414 Log has_invalid objects 2025-09-05 02:14:03 +03:00
Vitaliy Filippov 90b1019636 Do not warn on incomplete+has_invalid PG states as unexpected 2025-09-05 02:10:57 +03:00
Vitaliy Filippov df604afbd5 Fix OSD reweight values between 0 and 1 not working 2025-09-05 02:05:01 +03:00
Vitaliy Filippov 47c7aa62de Remove unused SUBMIT_SCRUB_READ 2025-08-30 16:30:52 +03:00
Vitaliy Filippov 9f2dc48d0f Fix OSD assertion failed: n_subops != sent when all object copies are corrupted 2025-08-30 16:30:47 +03:00
Vitaliy Filippov 6d951b21fb Install ibverbs-providers in vitastor Docker builds 2025-08-30 02:28:26 +03:00
Vitaliy Filippov 552f28cb3e Fix #86 - base64_decode on arm64 O_o 2025-08-30 02:27:05 +03:00
Vitaliy Filippov e87b6e26f7 Fix Proxmox 9.0 support - oops :) 2025-08-26 20:24:12 +03:00
Vitaliy Filippov 0c89886374 Release 2.3.0
New features:

- Add a new kernel device mounting method: [ublk](https://vitastor.io/docs/usage/ublk.html).
  It's the fastest method for random IOPS, and it's on par with VDUSE for linear MB/s.
- Disable io_uring waits being reported as iowait on kernels which support it (6.15+)
- Allow to enforce permissions at the VitastorFS NFS server side
- Add qemu_file_mirror_path option to the config to allow to trick Veeam and make it work
- Speed up CRC32C calculation in OSD by fixing a bug and enabling AVX512 version
- Support QEMU 10
- Support Debian 13 Trixie and Proxmox 9.0
- Remove the dependency on system liburing in package builds (build it statically)

Bug fixes:

- Fix checksums NEVER BEING ENABLED in vitastor-disk prepare, even when explicitly requested :-D
- Use default uid and gid from NFS AUTH_SYS when creating files
- Fix object bitmaps supposedly & possibly being corrupted in some rare cases with EC N+2+
- Avoid multiple inflight overwrites to meta blocks - fixes possible data corruption with
  one specific SSD model: Memblaze PBlaze5 910 (github #79)
- Fix a bug in antietcd which was leading to leases sometimes not expiring correctly
- Fix a bug in NFS where ".." entry had its `cookie` equal to 0 instead of 1 (github #78)
- Fix snapshots not being deleted during VM deletion in Proxmox plugin (github #85)
- Fix monitor not filtering OSDs by block size correctly
2025-08-25 21:39:19 +03:00
Vitaliy Filippov e79bef8751 Fix pve-qemu-10.0 patch 2025-08-25 21:38:18 +03:00
Vitaliy Filippov ad76f84e1c Fix compat.h for even older kernel headers 2025-08-25 21:38:04 +03:00
Vitaliy Filippov db827cb34c Add --enforce 1 to docs 2025-08-25 16:05:40 +03:00
Vitaliy Filippov e5c6d85ea1 Add Debian trixie to docs 2025-08-25 00:47:23 +03:00
Vitaliy Filippov 6cc44c1f54 Add qemu_file_mirror_path docs 2025-08-25 00:45:52 +03:00
Vitaliy Filippov c20450c1f1 Add build.sh for Debian Trixie 2025-08-24 22:15:55 +03:00
Vitaliy Filippov db63e58b3d Add UBLK docs 2025-08-24 16:55:58 +03:00
Vitaliy Filippov 31b7021330 Implement Vitastor ublk server 2025-08-24 16:55:58 +03:00
Vitaliy Filippov 2ebe3a468c Mark all symbols hidden by default, export only required ones 2025-08-24 16:55:58 +03:00
Vitaliy Filippov 9892fccfb0 Disable io_uring waits being reported as iowait 2025-08-24 16:55:58 +03:00
Vitaliy Filippov 0be86a306d Remove old liburing version support as it's now included 2025-08-24 16:55:58 +03:00
Vitaliy Filippov d77a775948 Fix liburing include/compat for older kernels 2025-08-24 16:55:58 +03:00
Vitaliy Filippov 8cc82bab39 Include liburing in static build 2025-08-24 16:55:58 +03:00
Vitaliy Filippov f9d5e33ddd Fix filtering OSDs by block size in monitor 2025-08-24 16:46:36 +03:00
f83418d93e Fix snapshots not being deleted during VM deletion in Proxmox plugin (#85)
Co-authored-by: zhu.chengzhen <zhu.chengzhen@jingjiamicro.com>

I accept Vitastor CLA agreement: https://git.yourcmc.ru/vitalif/vitastor/src/branch/master/CLA-en.md
2025-08-24 16:28:12 +03:00
Vitaliy Filippov fbf14fb0cb Allow to enforce permissions at the server side 2025-08-24 16:20:19 +03:00
Vitaliy Filippov fb1c3e00f4 Use default uid and gid from NFS AUTH_SYS when creating files 2025-08-24 16:03:40 +03:00
ston3luandVitaliy Filippov d8332171e9 According to the NFSv3 protocol: Directory traversal should start with cookie '0' - "." entry as the first entry, cookie should be '0' - '.." entry as the second entry, cookie should be '1' 2025-08-24 14:44:48 +03:00
Vitaliy Filippov c24cc9bf0b Bump antietcd version to 1.1.3 2025-08-23 17:45:56 +03:00
Vitaliy Filippov 9f57c75acf unordered_map flush_versions 2025-08-23 17:45:56 +03:00
Vitaliy Filippov 53b12641d1 Fix the fix :) 2025-08-23 17:45:56 +03:00
flynn.yangandVitaliy Filippov 5c5c8825dc Avoid multiple inflight overwrites to meta blocks
https://github.com/vitalif/vitastor/pull/83

By https://github.com/Flynn049

By submitting this pull request, I accept Vitastor CLA
2025-08-23 17:45:03 +03:00
Vitaliy Filippov 3a261ac3fc Support QEMU 10 2025-08-18 10:51:04 +03:00
Vitaliy Filippov 04514435de Fix checking config in qemu driver 2025-08-12 01:16:57 +03:00
Vitaliy Filippov 07303020fc Fix enabling checksums in blockstore-disk O_o 2025-08-10 18:04:52 +03:00
Vitaliy Filippov feaf7a15cf Use CRC32C implementation from ISA-L when available - enables VPCLMULQDQ version with AVX512 which is ~2x faster 2025-08-10 18:03:47 +03:00
Vitaliy Filippov 29dda5066f Stop doing cpuid repeatedly in runtime 2025-08-10 18:03:22 +03:00
Vitaliy Filippov 1de53ef7e6 Move crc32c_pad to util/crc32c.c 2025-08-10 18:03:17 +03:00
Vitaliy Filippov 4793dbe9c3 Use set_immediate() in osd_flush to prevent stack overflows on repeated errors 2025-08-10 17:58:01 +03:00
Vitaliy Filippov 918ea34af2 Remove "Only allow to overwrite part of the bitmap" blockstore API feature 2025-08-10 17:57:04 +03:00
Vitaliy Filippov 2db8184cd8 Fix bitmap calculation for EC N+1 & the new store and EC N+2+ for the old store 2025-08-10 17:57:04 +03:00
Vitaliy Filippov 0e964b3c8c Fix renaming from ddeb 2025-08-09 16:11:22 +03:00
Vitaliy Filippov 1b9296ff6c Add qemu_file_mirror_path option to the config to allow to trick Veeam
When qemu_file_mirror_path is set to "/dir/" in /etc/vitastor/vitastor.conf, QEMU driver
returns "/dir/image_name" as the filename in qemu-img info and in QAPI to trick software
like Veeam which expects file-based access. After that, vitastor-nfs can be mounted to that
directory to make backups work :-)
2025-08-06 01:15:54 +03:00
Vitaliy Filippov 6bf136c199 Add a note about /etc/apt/preferences to docs 2025-08-05 11:25:06 +03:00
Vitaliy Filippov b529f77264 Release 2.2.3
- Support Ubuntu 24.04 Noble
- Fix clients hanging on online PG count change with in-flight operations
- Fix volume_size_info in PVE VitastorPlugin
- Implement bdrv_refresh_filename for the QEMU driver
- Fix RDMA-CM broken in 2.2.0
- Fix possible "invalid %N$ use detected" OSD crashes
- Fix vitastor-nfs build without RDMA
- Fix docker build by adding the forgotten apt/preferences
- Fix libfio_blockstore.so
- Fix "trigger event loop automatically" API version check in the QEMU driver
- Add Content-Type header for prometheus metrics
- Add a patch for libvirt 11.5
2025-07-30 10:50:02 +03:00
Vitaliy Filippov bf9519dcdc docker -i 2025-07-30 10:50:02 +03:00
Vitaliy Filippov 4ba687738b Use archive for debian buster 2025-07-30 10:50:02 +03:00
Vitaliy Filippov 8427f6fe46 Use the new build method for RPM packages too 2025-07-30 02:38:52 +03:00
Vitaliy Filippov efa6bc3e70 Fix docker pull commands in docs 2025-07-30 02:38:52 +03:00
Vitaliy Filippov da33e9b12d Add noble repository 2025-07-30 02:38:52 +03:00
Vitaliy FilippovandVitaliy Filippov 265127c1a7 Fix vitastor-nfs build without RDMA 2025-07-30 02:38:52 +03:00
Vitaliy Filippov 2b30acfc1d Do not use podman and do not use build command to build packages (we are not building a docker image) 2025-07-30 02:38:52 +03:00
Vitaliy Filippov 7fbc38ef29 Add a build file for ubuntu 24.04 2025-07-29 02:51:59 +03:00
Vitaliy Filippov e5070e991a Add forgotten apt/preferences 2025-07-29 01:18:55 +03:00
Vitaliy Filippov 625552c441 Fix RDMA-CM broken in 2.2.0 2025-07-29 00:14:05 +03:00
Vitaliy Filippov 78c95c94f6 Use uint8_t* for BS buffers in OSD 2025-07-26 14:13:12 +03:00
Vitaliy Filippov 488e20bf55 Remove BS_OP_SYNC_STAB_ALL from OSD 2025-07-26 14:13:12 +03:00
Vitaliy Filippov 25d6281b3e Remove BS_OP_SYNC_STAB_ALL 2025-07-26 14:13:11 +03:00
Vitaliy Filippov 1676e50b3a Fix debug trace for fio_bs 2025-07-26 14:12:48 +03:00
Vitaliy Filippov 8049e3c14a Use OP_WRITE_STABLE in bs_fio 2025-07-26 14:12:48 +03:00
Vitaliy Filippov 93a30efd86 Do not use numbered printf args 2025-07-26 14:11:15 +03:00
Vitaliy Filippov 83fb121f36 Fix "Trigger event loop automatically" API version check 2025-07-20 17:14:22 +03:00
Vitaliy Filippov afc97b757b Implement bdrv_refresh_filename 2025-07-20 16:30:01 +03:00
Vitaliy Filippov 68905cbf41 Fix online PG count change, add a test for it 2025-07-15 02:33:28 +03:00
Vitaliy Filippov 3fff667f13 Add Content-Type header for prometheus metrics 2025-07-14 19:51:35 +03:00
Vitaliy Filippov 980aec1d9b Add patch for libvirt 11.5 2025-07-04 23:15:18 +03:00
Vitaliy Filippov f515fcce62 Fix volume_size_info in PVE VitastorPlugin 2025-06-18 12:38:09 +03:00
Vitaliy Filippov 97bb809b54 Release 2.2.2
- Fix a bug introduced in 2.2.0 - pg_locks weren't disabled for pools without local_reads
  correctly which could lead to inactive pools during various operations
- Fix an old bug where OSDs could send sub-operations to incorrect peer OSDs when their
  connections were stopped and reestablished quickly, in 2.2.0 it was usually leading
  to "sequencing broken" messages in OSD logs
- Fix debug use_sync_send_recv mode
2025-06-07 12:56:48 +03:00
Vitaliy Filippov 6022a61329 Decouple break_pg_locks from outbound OSD disconnections 2025-06-05 02:48:54 +03:00
Vitaliy Filippov a3c1996101 Do not accidentally clear incorrect osd_peer_fds entries 2025-06-05 02:22:13 +03:00
Vitaliy Filippov 8d2a1f0297 Fix PG lock auto-enabling/auto-disabling in the default configuration 2025-06-05 02:22:01 +03:00
Vitaliy Filippov 91cbc313c2 Change "on osd -123" logging to "on peer 123" for unknown connections 2025-06-05 02:22:01 +03:00
Vitaliy Filippov f0a025428e Postpone read/write handlers using timerfd in the debug use_sync_send_recv mode 2025-06-05 02:22:01 +03:00
Vitaliy Filippov 67071158bd Cancel outbound operations only in the osd_client_t destructor
This is required to prevent disconnected peers from sometimes receiving messages
suited for other peers - stop_client was freeing the operations even though they
were still references in the io_uring requests in progress. This was leading to
OSDs sometimes receiving garbage and "broken sequencing" errors in logs as the
memory was usually already reallocated for other operations
2025-06-05 02:09:41 +03:00
Vitaliy Filippov cd028612c8 Use a separate osd_client_t::in_osd_num for inbound OSD connections 2025-06-05 02:09:41 +03:00
Vitaliy Filippov f390e73dae Log broken sequence numbers in "sequencing" errors 2025-06-05 02:09:41 +03:00
Vitaliy Filippov de2539c491 Correct Proxmox version 2025-06-03 01:56:09 +03:00
Vitaliy Filippov 957a4fce7e Release 2.2.1
- Fix vitastor-disk purge broken after adding the "OSD is still running" check
- Fix iothreads hanging after adding zero-copy send support
- Fix enabling localized reads online (without restarting OSDs) in the default PG lock mode
2025-05-25 01:04:48 +03:00
Vitaliy Filippov f201ecdd51 Fix missing mutex unlock with zero-copy and iothreads O_o 2025-05-24 00:56:31 +03:00
Vitaliy Filippov 4afb617f59 Also zero-init sqe 2025-05-23 21:18:37 +03:00
Vitaliy Filippov d3fde0569f Add a test with enabled iothreads 2025-05-23 21:05:18 +03:00
Vitaliy Filippov 438b64f6c3 Allow to enable PG locks online when changing local_reads in pool configuration 2025-05-23 20:54:47 +03:00
Vitaliy Filippov 2b0a802ea1 Fix iothreads sometimes hanging after adding zerocopy support 2025-05-23 20:54:03 +03:00
Vitaliy Filippov 0dd49c1d67 Followup to "allow to purge running OSDs again" 2025-05-22 01:10:05 +03:00
Vitaliy Filippov 410170db96 Add notes about VNPL in English 2025-05-20 02:12:49 +03:00
Vitaliy Filippov 7d8523e0e5 Add more notes about VNPL in Russian 2025-05-19 02:41:34 +03:00
Vitaliy Filippov db915184c6 Allow to purge running OSDs again, as in 2.1.0 and earlier 2025-05-11 13:59:28 +03:00
Vitaliy Filippov 5ae6fea49c Add a note about local reads 2025-05-11 01:23:48 +03:00
Vitaliy Filippov 95ec750b8c Release 2.2.0
New features:

- [Localized read support](https://vitastor.io/docs/config/pool.html#local_reads) for multi-datacenter setups.
- io_uring-based [zero-copy send support](https://vitastor.io/docs/config/network.html#min_zerocopy_send_size) -
  read the instruction carefully for optimal performance!
- Improve and speedup data distribution, especially in cases of very large hosts (100 OSD+).
  Previously, PG optimization speed depended on the number of OSDs, now it only depends on
  the number of failure domains. Distribution over specific OSDs is now also more even and
  becomes strictly more even when you increase the number of PGs.
- Add a [very interesting instruction](https://vitastor.io/en/docs/usage/nfs.html#linux-nfs-write-size) to change NFS_MAX_FILE_IO_SIZE
- Check operation sequencing and stop connections when it breaks - should help catch some
  very rare RDMA packet loss problems.
- `vitastor-cli rm-osd` now refuses to remove OSDs which are still up and suggests to use `vitastor-disk purge`.
- Allow removal of direntries referring non-existent in VitastorFS.
- Change default vitastor-etcd data dir to /var/lib/etcd/vitastor.

Bug fixes:

- Fix compatibility with ISA-L 2.31+. ⚠️Very important: please upgrade Vitastor before upgrading ISA-L to 2.31+.
- Fix in-memory state cleanup for incomplete PGs.
- Fix monitor crash with non-existent node_placement nodes.
- Slightly speedup `vitastor-kv dump` command by adding output buffering.
- Fix theoretically possible slowdowns in OSD sub-operation failure handling code.
- Fix very rare stack overflows in vitastor-kv.
- Fix a possible crash in VitastorFS during handling of file creation race condition.
- Fix modify-pool -s PG_SIZE which didn't work without --pg_minsize.
- Fix marking peer OSDs as alive on receiving data from them via RDMA - in theory,
  the bug could result in instability with RDMA under high load with slow disks.
- Fix a rare OSD crash due to double handle_primary_subop() call.
- Fix latency aggregation in global stats (/vitastor/stats in etcd) - do not sum it.
- Hide "Ran out of journal space" log messages by default.
- Wait for RDMA-CM EVENT_ESTABLISHED after rdma_accept(), handle rdma_accept() before acking the event.
- Fix VitastorFS total & free numbers multiplied by extra 2.
- Fix systemd unit name in make-etcd.
- Do not allow reweight > 1 in vitastor-cli modify-osd.
- Fix docker build.
2025-05-11 00:26:08 +03:00
Vitaliy Filippov 90b1de307b Support local reads in client 2025-05-10 16:42:02 +03:00
Vitaliy Filippov 7e6a95c678 Support primary-reads from clean replicated PGs on secondary OSDs 2025-05-10 16:42:02 +03:00
Vitaliy Filippov b2416afb28 Lock PGs on secondary OSDs to allow local reads and guarantee splitbrain prevention 2025-05-10 15:18:00 +03:00
Vitaliy Filippov 66dc116f60 Cleanup PG_INCOMPLETE peering states 2025-05-10 02:51:26 +03:00
Vitaliy Filippov 0cb8629ab6 Remove finish_stop_pg shortcut 2025-05-08 16:14:35 +03:00
Vitaliy Filippov b7322a405a Move gethostname_str to utils 2025-05-05 02:16:06 +03:00
Vitaliy Filippov 5692630005 Move check_sequencing indication into config response features subkey 2025-05-05 02:07:50 +03:00
Vitaliy Filippov 00ced7cea7 Fix monitor crash with non-existing node_placement nodes 2025-05-05 02:05:04 +03:00
Vitaliy Filippov ebdb75e287 Fix typo in docker docs 2025-05-04 18:09:33 +03:00
Vitaliy Filippov f397fe9c6a Add compatibility with ISA-L 2.31+ 2025-05-04 18:09:33 +03:00
Vitaliy Filippov 28560b4ae5 Write K/V listings in buffered manner 2025-05-03 15:06:59 +03:00
Vitaliy Filippov 2d07449e74 Postpone cb() to set_immediate() to prevent stack overflows in kv_db 2025-05-03 15:06:59 +03:00
Vitaliy Filippov 80c4e8c20f Add missing wakeup in ringloop->set_immediate to prevent slowdowns in code using set_immediate 2025-05-03 14:40:48 +03:00
Vitaliy Filippov 2ab0ae3bc9 Check operation sequencing and stop clients when it breaks 2025-05-02 17:01:50 +03:00
Vitaliy Filippov 05e59c1b4f Fix MSG_WAITALL assertion added in the zero-copy patch 2025-05-02 17:01:43 +03:00
Vitaliy Filippov e6e1c5b962 Check if OSDs are still up in rm-osd 2025-05-02 13:05:59 +03:00
Vitaliy Filippov 9556eeae45 Implement io_uring zero-copy send support 2025-05-01 18:47:10 +03:00
Vitaliy Filippov 96b5a72630 Allow removal of bad direntries in VitastorFS (direntries referring non-existent inodes) 2025-05-01 01:14:23 +03:00
Vitaliy Filippov ef80f121f6 Fix "duplicate inode during create" deletion in VitastorFS 2025-04-30 20:37:49 +03:00
Vitaliy Filippov bbdd1f3aa7 Fix modify-pool -s PG_SIZE without --pg_minsize 2025-04-28 02:20:54 +03:00
Vitaliy Filippov 5dd37f519a Fix node folding in case of empty rules (pool with size 1), add a test 2025-04-28 02:16:49 +03:00
Vitaliy Filippov a2278be84d Improve data distribution: solve LP task on failure domains instead of individual OSDs
This greatly speeds up PG placement and makes it more uniform both because the LP task
becomes simpler and because the distribution of individual OSDs is optimised manually
2025-04-27 01:44:46 +03:00
Vitaliy Filippov 1393a2671c Change default vitastor-etcd data dir to /var/lib/etcd/vitastor 2025-04-27 01:44:46 +03:00
Vitaliy Filippov 9fa8ae5384 Reset OSD ping state on receiving data from it via RDMA 2025-04-26 14:17:09 +03:00
Vitaliy Filippov 169a35a067 Followup to latency aggregation fix 2025-04-26 01:32:46 +03:00
Vitaliy Filippov 2b2a10581d Prevent double handle_primary_subop in rare cases 2025-04-26 01:16:53 +03:00
Vitaliy Filippov 10fd51862a Fix latency aggregation in global stats (/vitastor/stats in etcd) 2025-04-25 00:08:10 +03:00
Vitaliy Filippov 15d0204f96 Hide "Ran out of journal space" log messages by default 2025-04-20 01:21:03 +03:00
Vitaliy Filippov 21d6e88a1b Add instructions to change NFS_MAX_FILE_IO_SIZE 2025-04-18 13:39:32 +03:00
Vitaliy Filippov df2847df2d Wait for RDMA-CM EVENT_ESTABLISHED after rdma_accept(), handle rdma_accept() before acking the event 2025-04-15 15:19:36 +03:00
Vitaliy Filippov 327c98a4b6 Fix index_tree 2025-04-13 16:13:44 +03:00
Vitaliy Filippov 3cc0abfd81 Fix NFS total & free multiplied by extra 2 2025-04-12 19:09:26 +03:00
Vitaliy Filippov 80e5f8ba76 Add missing WITH_RDMACM defines 2025-04-12 19:06:27 +03:00
Vitaliy Filippov 4b660f1ce8 Fix systemd unit name in make-etcd 2025-04-11 02:05:08 +03:00
Vitaliy Filippov dfde0e60f0 Do not allow reweight > 1 2025-04-05 12:21:14 +03:00
Vitaliy Filippov 013f688ffe Run check_peer_config on RDMA-CM connections too 2025-04-02 01:32:28 +03:00
Vitaliy Filippov cf9738ddbe Fix docker 2.1.0 build :) 2025-04-01 22:46:22 +03:00
Vitaliy Filippov 891b2811c7 Release 2.1.0
New features:

- Support separate OSD cluster network - [osd_cluster_network](https://vitastor.io/docs/config/network.html#osd_cluster_network)
  and, in general, multiple OSD networks, including RDMA
- Add an alternative RDMA implementation via RDMA-CM - [use_rdmacm](https://vitastor.io/docs/config/network.html#use_rdmacm),
  required for iWARP and, maybe, for some IB setups (but not for RoCE)
- Change default PG behaviour to wait for all "up" OSDs to be connected before starting it.
  The old behaviour may be returned by enabling a new [allow_net_split](https://vitastor.io/docs/config/osd.html#allow_net_split)
  option.
- Add a patch for QEMU 9.2

Bug fixes:

- Fix incorrect "has_xxx" PG state names in ls-pgs
- Fix possible QEMU crashes after detaching of Vitastor disks (and update all QEMU builds in Vitastor repos)
- Fix clients sometimes spamming OSDs with infinite reconnections when some PGs are offline
- Fall back to TCP on RDMA connection failures
- Add missing logging of RDMA ibv_modify_qp() errors
- Add a minimum interval for etcd_state_client to reload state
2025-04-01 20:16:27 +03:00
Vitaliy Filippov 01590df6da Update QEMU version in vitastor-csi Dockerfile 2025-04-01 20:16:27 +03:00
Vitaliy Filippov 3e5f0be52c Use separate port numbers for RDMA-CM 2025-04-01 16:16:03 +03:00
Vitaliy Filippov 58af897e73 s/listen on/listen to/ :) 2025-04-01 12:07:15 +03:00
Vitaliy Filippov dbf9ecd171 Move osd_network to config/network docs 2025-03-31 21:12:09 +03:00
Vitaliy Filippov 8508e78288 Add an alternative RDMA implementation via RDMA-CM
Required for non-RoCE cards: iWARP and, possibly, Infiniband
2025-03-31 21:01:25 +03:00
Vitaliy Filippov f32dea02bf Support multiple RDMA networks 2025-03-31 21:01:25 +03:00
Vitaliy Filippov a103065d12 Support multiple OSD networks and separate OSD cluster network 2025-03-31 21:01:15 +03:00
Vitaliy Filippov 5d2e28d4a9 Remove unused used_max_cqe from nfs_proxy_rdma 2025-03-30 02:13:24 +03:00
Vitaliy Filippov 18e14eed11 Fix --pg_count formula in docs/usage/cli 2025-03-29 17:54:53 +03:00
Vitaliy Filippov ccc32b9e68 Use TCP on RDMA connection failure 2025-03-23 12:04:23 +03:00
Vitaliy Filippov ebaf3fee79 Add an assertion to prevent sending message to TCP channel when switched to RDMA 2025-03-23 12:04:09 +03:00
Vitaliy Filippov 196d28e987 Fix typo 2025-03-23 12:00:20 +03:00
Vitaliy Filippov 8f243b2328 Fix qemu buster build and bullseye version 2025-03-23 02:46:52 +03:00
Vitaliy Filippov 7a835fcd8f Add allow_net_split parameter 2025-03-23 02:12:32 +03:00
Vitaliy Filippov 8b0389b4e8 Log RDMA ibv_modify_qp() errors 2025-03-22 15:58:13 +03:00
Vitaliy Filippov f544c350ba %l* -> %j* 2025-03-22 15:32:07 +03:00
Vitaliy Filippov 4eafb55b5c Add a patch for QEMU 9.2, fix debian bookworm QEMU build 2025-03-22 15:30:52 +03:00
Vitaliy Filippov 5030396f71 Clear QEMU eventfd handler on vitastor block driver destruction 2025-03-21 20:47:17 +03:00
Vitaliy Filippov be22c363ca Do not skip client_retry_interval on reconnecting OSDs to prevent OSD spam 2025-03-20 00:12:38 +03:00
Vitaliy Filippov 0f80c87b43 Add a minimum interval for etcd_state_client to reload state
(To prevent excessive load on etcd during outages)
2025-03-19 02:36:09 +03:00
Vitaliy Filippov e0953fd502 Wait for all "up" OSDs to be connected before starting PG 2025-03-19 02:36:09 +03:00
Vitaliy Filippov 6e0ae47938 Add Proxmox QEMU 9.2 patch 2025-03-19 02:36:02 +03:00
Vitaliy Filippov b8f19e85ad Fix pg state formatting in ls-pgs 2025-03-17 01:37:58 +03:00
Vitaliy Filippov b7636e595f Update version in docker docs 2025-03-16 16:53:57 +03:00
Vitaliy Filippov 48c026bfa0 Release 2.0.0
No breaking features, it's 2.0.0 just because it includes S3 and because
there are already too many 1.x releases :).

New features:

- S3 is finally available: https://vitastor.io/docs/installation/s3.html
- node.js addon is now packaged as a Debian package
- Support listing PGs by OSDs in `vitastor-cli ls-pgs`
- Implement offline TRIM support: [vitastor-disk trim](https://vitastor.io/docs/usage/disk.html#trim),
  [discard_on_start](https://vitastor.io/docs/config/osd.html#discard_on_start)
- Change used_for_fs pool option to used_for_app

Bug fixes:

- Fix several bugs in the node.js addon (a memory leak, an incorrectly triggered event loop)
- Fix a client crash (vitastor-cli rm) during deletion when writeback is enabled
- Fix PG object count statistics on deletion of non-existing objects
- Fix vitastor-nbd crash when mapping by ID instead of inode name
- Fix a client memory leak with enabled immediate_commit and write-back cache
- Add seccomp=unconfined for vitastor docker OSDs to not break io_uring
- Add udev and systemd to vitastor docker image
- Fix upgrading from pre-0.7.1 (very old) systemd units O_o
- Fix total object count calculation in rm_data
2025-03-16 14:34:31 +03:00
Vitaliy Filippov a73b2a26b6 Fix blockstore initialization after moving clean_dyn_size calc to calc_lengths 2025-03-16 13:44:02 +03:00
Vitaliy Filippov f3192b610d Fix vitastor-disk in Docker installations 2025-03-16 13:44:01 +03:00
Vitaliy Filippov a950889976 Add missing docs for discard_on_start 2025-03-16 12:29:22 +03:00
Vitaliy Filippov ef5194d93c Add S3 installation docs 2025-03-16 01:17:09 +03:00
Vitaliy Filippov f904576ab1 Fix total calculation in rm_data 2025-03-15 17:01:10 +03:00
Vitaliy Filippov 4f9b1f2f62 Support listing PGs by OSDs 2025-03-15 16:42:57 +03:00
Vitaliy Filippov 1d94afbd51 Implement offline TRIM support 2025-03-14 01:37:16 +03:00
Vitaliy Filippov 3634f005f1 Fix upgrading from pre-0.7.1 systemd units O_o 2025-03-14 01:37:16 +03:00
Vitaliy Filippov 263a3b5ad6 Rename allocator to allocator_t 2025-03-13 00:53:34 +03:00
Vitaliy Filippov b760951aa7 Add seccomp=unconfined for vitastor docker OSDs to not break io_uring 2025-03-11 00:42:10 +03:00
Vitaliy Filippov c8321b8ed1 Add udev and systemd to vitastor docker image 2025-03-11 00:40:39 +03:00
Vitaliy Filippov 21066a095b Fix a memory leak with enabled immediate_commit and write-back cache
Remove dirty buffers after writing when immediate_commit is on instead
of saving them for repeating later
2025-03-11 00:40:18 +03:00
Vitaliy Filippov a96900b696 Explicitly destroy Nan::Persistents, otherwise it leaks memory 2025-03-09 16:45:10 +03:00
Vitaliy Filippov 8a6e461322 Fix license (VNPL 1.1, not 2.0) 2025-03-08 17:17:23 +03:00
Vitaliy Filippov 0b6a0463a4 Save a reference to the buffer during write 2025-03-08 16:00:26 +03:00
Vitaliy Filippov 35d4047f46 Fix vitastor-nbd crash when mapping by ID instead of inode name 2025-03-08 15:52:57 +03:00
Vitaliy Filippov 819f1125ae Support used_for_app instead of used_for_fs 2025-03-07 01:03:43 +03:00
Vitaliy Filippov 108df7329f Fix PG object count statistics on deletion of non-existing objects 2025-03-04 00:40:56 +03:00
Vitaliy Filippov d32edf6cdf Fix deletion writeback 2025-03-04 00:40:35 +03:00
Vitaliy Filippov dca436d7e6 Trigger event loop automatically in libvitastor_c 2025-03-03 00:57:09 +03:00
Vitaliy Filippov 8129a0b4e3 Loop once after registering eventfd to prevent skipping previous events 2025-03-03 00:57:00 +03:00
Vitaliy Filippov 704c87d512 Trigger initial epoll when adding an FD 2025-03-03 00:56:17 +03:00
383 changed files with 40860 additions and 9888 deletions
+9 -8
View File
@@ -1,28 +1,29 @@
FROM node:16-bullseye
FROM node:16-bookworm
WORKDIR /root
ADD ./docker/vitastor.gpg /etc/apt/trusted.gpg.d
ADD ./docker/etc/apt/trusted.gpg.d /etc/apt/trusted.gpg.d
RUN echo 'deb http://deb.debian.org/debian bullseye-backports main' >> /etc/apt/sources.list; \
echo 'deb http://vitastor.io/debian bullseye main' >> /etc/apt/sources.list; \
RUN echo 'deb http://deb.debian.org/debian bookworm-backports main' >> /etc/apt/sources.list; \
echo 'deb http://vitastor.io/debian bookworm main' >> /etc/apt/sources.list; \
echo >> /etc/apt/preferences; \
echo 'Package: *' >> /etc/apt/preferences; \
echo 'Pin: release a=bullseye-backports' >> /etc/apt/preferences; \
echo 'Pin: release n=bookworm-backports' >> /etc/apt/preferences; \
echo 'Pin-Priority: 500' >> /etc/apt/preferences; \
echo >> /etc/apt/preferences; \
echo 'Package: *' >> /etc/apt/preferences; \
echo 'Pin: origin "vitastor.io"' >> /etc/apt/preferences; \
echo 'Pin-Priority: 1000' >> /etc/apt/preferences; \
perl -i -pe 's/Types: deb$/Types: deb deb-src/' /etc/apt/sources.list.d/debian.sources; \
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 etcd qemu-system-x86 qemu-block-extra qemu-utils fio libasan5 \
liburing1 liburing-dev libgoogle-perftools-dev devscripts libjerasure-dev cmake libibverbs-dev libisal-dev
RUN apt-get -y install etcd qemu-system-x86 qemu-block-extra qemu-utils fio libasan8 \
libgoogle-perftools-dev devscripts libjerasure-dev cmake libibverbs-dev libisal-dev
RUN apt-get -y build-dep fio qemu=`dpkg -s qemu-system-x86|grep ^Version:|awk '{print $2}'`
RUN apt-get update && apt-get -y install jq lp-solve sudo nfs-common fdisk parted
RUN apt-get update && apt-get -y install jq lp-solve sudo nfs-common fdisk parted libc-ares-dev udev
RUN apt-get --download-only source fio qemu=`dpkg -s qemu-system-x86|grep ^Version:|awk '{print $2}'`
RUN set -ex; \
File diff suppressed because it is too large Load Diff
+4
View File
@@ -38,6 +38,10 @@ for my $line (<>)
{
$test_name .= '_antietcd';
}
elsif ($1 eq 'OLD')
{
$test_name =~ s/^test_/test_old_/s;
}
else
{
$test_name .= '_'.lc($1).'_'.$2;
+15 -2
View File
@@ -1,7 +1,20 @@
cmake_minimum_required(VERSION 2.8.12)
cmake_minimum_required(VERSION 2.8...3.30)
project(vitastor)
set(VITASTOR_VERSION "1.11.0")
set(VITASTOR_VERSION "3.0.8")
include(CTest)
add_custom_target(build_tests)
set_property(TEST PROPERTY ENVIRONMENT LSAN_OPTIONS=suppressions=${CMAKE_CURRENT_BINARY_DIR}/lsan-suppress.txt)
add_test(gen_lsan_suppress
${CMAKE_COMMAND} -E echo leak:tcmalloc > "${CMAKE_CURRENT_BINARY_DIR}/lsan-suppress.txt"
)
set_tests_properties(gen_lsan_suppress PROPERTIES FIXTURES_SETUP f_lsan_suppress)
set_property(TEST PROPERTY FIXTURES_REQUIRED f_lsan_suppress)
# make -j16 -C ../../build test_heap && ../../build/src/test/test_heap
# make -j16 -C ../../build test_heap && rm -f $(find ../../build -name '*.gcda') && ctest -V -T test -T coverage -R heap --test-dir ../../build && (cd ../../build; gcovr -f ../src --html --html-nested -o coverage/index.html; cd ../src/test)
# make -j16 -C ../../build test_blockstore && rm -f $(find ../../build -name '*.gcda') && ctest -V -T test -T coverage -R blockstore --test-dir ../../build && (cd ../../build; gcovr -f ../src --html --html-nested -o coverage/index.html; cd ../src/test)
# kcov --include-path=../../../src ../../kcov ./test_blockstore
add_subdirectory(src)
+12 -6
View File
@@ -6,7 +6,7 @@
Вернём былую скорость кластерному блочному хранилищу!
Vitastor - распределённая блочная и файловая SDS (программная СХД), прямой аналог Ceph RBD и CephFS,
Vitastor - распределённая блочная, файловая и объектная SDS (программная СХД), прямой аналог Ceph RBD, CephFS и RGW,
а также внутренних СХД популярных облачных провайдеров. Однако, в отличие от них, Vitastor
быстрый и при этом простой. Только пока маленький :-).
@@ -19,18 +19,22 @@ Vitastor нацелен в первую очередь на SSD и SSD+HDD кл
TCP и RDMA и на хорошем железе может достигать задержки 4 КБ чтения и записи на уровне ~0.1 мс,
что примерно в 10 раз быстрее, чем Ceph и другие популярные программные СХД.
Vitastor поддерживает QEMU-драйвер, протоколы NBD и NFS, драйверы OpenStack, OpenNebula, Proxmox, Kubernetes.
Vitastor поддерживает QEMU-драйвер, протоколы UBLK, NBD и NFS, драйверы OpenStack, OpenNebula, Proxmox, Kubernetes.
Другие драйверы могут также быть легко реализованы.
Подробности смотрите в документации по ссылкам. Можете начать отсюда: [Быстрый старт](docs/intro/quickstart.ru.md).
## Презентации и записи докладов
- KuberConf'2025: [видео](https://vitastor.io/presentation/kuberconf.webm)
- Highload'2025: [видео](https://vitastor.io/presentation/hl2025/hl2025.webm),
[на youtube](https://www.youtube.com/watch?v=0R8MLjFtz7g), презентация
([на русском](https://vitastor.io/presentation/hl2025/), [на английском](https://vitastor.io/presentation/hl2025/en.html))
- Highload'2022: презентация ([на русском](https://vitastor.io/presentation/highload/highload.html)),
[видео](https://vitastor.io/presentation/highload/talk.webm)
- DevOpsConf'2021: презентация ([на русском](https://vitastor.io/presentation/devopsconf/devopsconf.html),
[на английском](https://vitastor.io/presentation/devopsconf/devopsconf_en.html)),
[видео](https://vitastor.io/presentation/devopsconf/talk.webm)
- Highload'2022: презентация ([на русском](https://vitastor.io/presentation/highload/highload.html)),
[видео](https://vitastor.io/presentation/highload/talk.webm)
## Документация
@@ -46,6 +50,7 @@ Vitastor поддерживает QEMU-драйвер, протоколы NBD и
- [OpenNebula](docs/installation/opennebula.ru.md)
- [OpenStack](docs/installation/openstack.ru.md)
- [Kubernetes CSI](docs/installation/kubernetes.ru.md)
- [S3](docs/installation/s3.ru.md)
- [Сборка из исходных кодов](docs/installation/source.ru.md)
- Конфигурация
- [Обзор](docs/config.ru.md)
@@ -63,8 +68,9 @@ Vitastor поддерживает QEMU-драйвер, протоколы NBD и
- [vitastor-cli](docs/usage/cli.ru.md) (консольный интерфейс)
- [vitastor-disk](docs/usage/disk.ru.md) (управление дисками)
- [fio](docs/usage/fio.ru.md) для тестов производительности
- [NBD](docs/usage/nbd.ru.md) для монтирования ядром
- [QEMU и qemu-img](docs/usage/qemu.ru.md)
- [UBLK](docs/usage/ublk.ru.md) для монтирования ядром
- [NBD](docs/usage/nbd.ru.md) - старый интерфейс для монтирования ядром
- [QEMU, qemu-img и VDUSE](docs/usage/qemu.ru.md)
- [NFS](docs/usage/nfs.ru.md) кластерная файловая система и псевдо-ФС прокси
- [Администрирование](docs/usage/admin.ru.md)
- Производительность
+12 -6
View File
@@ -6,7 +6,7 @@
Make Clustered Block Storage Fast Again.
Vitastor is a distributed block and file SDS, direct replacement of Ceph RBD and CephFS,
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 :-).
@@ -19,18 +19,22 @@ supports TCP and RDMA and may achieve 4 KB read and write latency as low as ~0.1
with proper hardware which is ~10 times faster than other popular SDS's like Ceph
or internal systems of public clouds.
Vitastor supports QEMU, NBD, NFS protocols, OpenStack, OpenNebula, Proxmox, Kubernetes drivers.
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](docs/intro/quickstart.en.md).
## Talks and presentations
- KuberConf'2025: [video](https://vitastor.io/presentation/kuberconf.webm)
- Highload'2025: [video](https://vitastor.io/presentation/hl2025/hl2025.webm),
[youtube](https://www.youtube.com/watch?v=0R8MLjFtz7g), presentation
([in Russian](https://vitastor.io/presentation/hl2025/), [in English](https://vitastor.io/presentation/hl2025/en.html))
- Highload'2022: presentation ([in Russian](https://vitastor.io/presentation/highload/highload.html)),
[video](https://vitastor.io/presentation/highload/talk.webm)
- DevOpsConf'2021: presentation ([in Russian](https://vitastor.io/presentation/devopsconf/devopsconf.html),
[in English](https://vitastor.io/presentation/devopsconf/devopsconf_en.html)),
[video](https://vitastor.io/presentation/devopsconf/talk.webm)
- Highload'2022: presentation ([in Russian](https://vitastor.io/presentation/highload/highload.html)),
[video](https://vitastor.io/presentation/highload/talk.webm)
## Documentation
@@ -46,6 +50,7 @@ Read more details in the documentation. You can start from here: [Quick Start](d
- [OpenNebula](docs/installation/opennebula.en.md)
- [OpenStack](docs/installation/openstack.en.md)
- [Kubernetes CSI](docs/installation/kubernetes.en.md)
- [S3](docs/installation/s3.en.md)
- [Building from Source](docs/installation/source.en.md)
- Configuration
- [Overview](docs/config.en.md)
@@ -63,8 +68,9 @@ Read more details in the documentation. You can start from here: [Quick Start](d
- [vitastor-cli](docs/usage/cli.en.md) (command-line interface)
- [vitastor-disk](docs/usage/disk.en.md) (disk management tool)
- [fio](docs/usage/fio.en.md) for benchmarks
- [NBD](docs/usage/nbd.en.md) for kernel mounts
- [QEMU and qemu-img](docs/usage/qemu.en.md)
- [UBLK](docs/usage/ublk.en.md) for kernel mounts
- [NBD](docs/usage/nbd.en.md) - old interface for kernel mounts
- [QEMU, qemu-img and VDUSE](docs/usage/qemu.en.md)
- [NFS](docs/usage/nfs.en.md) clustered file system and pseudo-FS proxy
- [Administration](docs/usage/admin.en.md)
- Performance
+8 -8
View File
@@ -1,5 +1,5 @@
# Compile stage
FROM golang:bookworm AS build
FROM golang:trixie AS build
ADD go.sum go.mod /app/
RUN cd /app; CGO_ENABLED=1 GOOS=linux GOARCH=amd64 go mod download -x
@@ -9,7 +9,7 @@ RUN perl -i -e '$/ = undef; while(<>) { s/\n\s*(\{\s*\n)/$1\n/g; s/\}(\s*\n\s*)e
CGO_ENABLED=1 GOOS=linux GOARCH=amd64 go build -o vitastor-csi
# Final stage
FROM debian:bookworm
FROM debian:trixie
LABEL maintainers="Vitaliy Filippov <vitalif@yourcmc.ru>"
LABEL description="Vitastor CSI Driver"
@@ -25,20 +25,20 @@ RUN apt-get update && \
# NFS mount dependencies
nfs-common netbase \
# dependencies of qemu-storage-daemon
libnuma1 liburing2 libglib2.0-0 libfuse3-3 libaio1 libzstd1 libnettle8 \
libgmp10 libhogweed6 libp11-kit0 libidn2-0 libunistring2 libtasn1-6 libpcre2-8-0 libffi8 && \
libaio1t64 libc6 libfuse3-4 libglib2.0-0t64 libgmp10 libgnutls30t64 \
libhogweed6t64 libnettle8t64 libnuma1 libselinux1 liburing2 libzstd1 zlib1g && \
apt-get clean && \
(echo options nbd nbds_max=128 > /etc/modprobe.d/nbd.conf)
COPY --from=build /app/vitastor-csi /bin/
RUN (echo deb http://vitastor.io/debian bookworm main > /etc/apt/sources.list.d/vitastor.list) && \
RUN (echo deb http://vitastor.io/debian trixie main > /etc/apt/sources.list.d/vitastor.list) && \
((echo 'Package: *'; echo 'Pin: origin "vitastor.io"'; echo 'Pin-Priority: 1000') > /etc/apt/preferences.d/vitastor.pref) && \
wget -q -O /etc/apt/trusted.gpg.d/vitastor.gpg https://vitastor.io/debian/pubkey.gpg && \
apt-get update && \
apt-get install -y vitastor-client && \
wget https://vitastor.io/archive/qemu/qemu-bookworm-8.1.2%2Bds-1%2Bvitastor1/qemu-utils_8.1.2%2Bds-1%2Bvitastor1_amd64.deb && \
wget https://vitastor.io/archive/qemu/qemu-bookworm-8.1.2%2Bds-1%2Bvitastor1/qemu-block-extra_8.1.2%2Bds-1%2Bvitastor1_amd64.deb && \
apt-get install -y vitastor-client ibverbs-providers && \
wget https://vitastor.io/archive/qemu/qemu-trixie-10.0.2%2Bds-2%2Bvitastor1/qemu-utils_10.0.2%2Bds-2%2Bvitastor1_amd64.deb && \
wget https://vitastor.io/archive/qemu/qemu-trixie-10.0.2%2Bds-2%2Bvitastor1/qemu-block-extra_10.0.2%2Bds-2%2Bvitastor1_amd64.deb && \
dpkg -x qemu-utils*.deb tmp1 && \
dpkg -x qemu-block-extra*.deb tmp1 && \
cp -a tmp1/usr/bin/qemu-storage-daemon /usr/bin/ && \
+49
View File
@@ -0,0 +1,49 @@
# Compile stage
FROM golang:trixie AS build
ADD go.sum go.mod /app/
RUN cd /app; CGO_ENABLED=1 GOOS=linux GOARCH=amd64 go mod download -x
ADD . /app
RUN perl -i -e '$/ = undef; while(<>) { s/\n\s*(\{\s*\n)/$1\n/g; s/\}(\s*\n\s*)else\b/$1} else/g; print; }' `find /app -name '*.go'` && \
cd /app && \
CGO_ENABLED=1 GOOS=linux GOARCH=amd64 go build -o vitastor-csi
# Final stage
FROM debian:trixie
LABEL maintainers="Vitaliy Filippov <vitalif@yourcmc.ru>"
LABEL description="Vitastor CSI Driver"
ENV NODE_ID=""
ENV CSI_ENDPOINT=""
RUN apt-get update && \
apt-get install -y wget && \
(echo "APT::Install-Recommends false;" > /etc/apt/apt.conf) && \
apt-get update && \
apt-get install -y e2fsprogs xfsprogs kmod iproute2 \
# NFS mount dependencies
nfs-common netbase \
# dependencies of qemu-storage-daemon
libnuma1 liburing2 libglib2.0-0 libfuse3-3 libaio1 libzstd1 libnettle8 \
libgmp10 libhogweed6 libp11-kit0 libidn2-0 libunistring2 libtasn1-6 libpcre2-8-0 libffi8 && \
apt-get clean && \
(echo options nbd nbds_max=128 > /etc/modprobe.d/nbd.conf)
COPY --from=build /app/vitastor-csi /bin/
ADD deb /deb
RUN apt-get update && \
apt-get -y install /deb/vitastor-client_*.deb && \
wget https://vitastor.io/archive/qemu/qemu-trixie-9.2.2%2Bds-1%2Bvitastor4/qemu-utils_9.2.2%2Bds-1%2Bvitastor4_amd64.deb && \
wget https://vitastor.io/archive/qemu/qemu-trixie-9.2.2%2Bds-1%2Bvitastor4/qemu-block-extra_9.2.2%2Bds-1%2Bvitastor4_amd64.deb && \
dpkg -x qemu-utils*.deb tmp1 && \
dpkg -x qemu-block-extra*.deb tmp1 && \
cp -a tmp1/usr/bin/qemu-storage-daemon /usr/bin/ && \
mkdir -p /usr/lib/x86_64-linux-gnu/qemu && \
cp -a tmp1/usr/lib/x86_64-linux-gnu/qemu/block-vitastor.so /usr/lib/x86_64-linux-gnu/qemu/ && \
rm -rf tmp1 *.deb && \
apt-get clean
ENTRYPOINT ["/bin/vitastor-csi"]
+1 -1
View File
@@ -1,4 +1,4 @@
VITASTOR_VERSION ?= v1.11.0
VITASTOR_VERSION ?= v3.0.8
all: build push
+1 -1
View File
@@ -49,7 +49,7 @@ spec:
capabilities:
add: ["SYS_ADMIN"]
allowPrivilegeEscalation: true
image: vitalif/vitastor-csi:v1.11.0
image: vitalif/vitastor-csi:v3.0.8
args:
- "--node=$(NODE_ID)"
- "--endpoint=$(CSI_ENDPOINT)"
+1 -1
View File
@@ -121,7 +121,7 @@ spec:
privileged: true
capabilities:
add: ["SYS_ADMIN"]
image: vitalif/vitastor-csi:v1.11.0
image: vitalif/vitastor-csi:v3.0.8
args:
- "--node=$(NODE_ID)"
- "--endpoint=$(CSI_ENDPOINT)"
+1 -1
View File
@@ -5,7 +5,7 @@ package vitastor
const (
vitastorCSIDriverName = "csi.vitastor.io"
vitastorCSIDriverVersion = "1.11.0"
vitastorCSIDriverVersion = "3.0.8"
)
// Config struct fills the parameters of request or user input
+115 -20
View File
@@ -33,7 +33,7 @@ import (
type NodeServer struct
{
*Driver
useVduse bool
method MountMethod
stateDir string
nfsStageDir string
mounter mount.Interface
@@ -81,16 +81,23 @@ func NewNodeServer(driver *Driver) *NodeServer
}
ns := &NodeServer{
Driver: driver,
useVduse: checkVduseSupport(),
method: selectMountMethod(),
stateDir: stateDir,
nfsStageDir: nfsStageDir,
mounter: mount.New(""),
volumeLocks: make(map[string]bool),
}
ns.cond = sync.NewCond(&ns.mu)
if (ns.useVduse)
if (ns.method == MOUNT_VDUSE)
{
ns.restoreVduseDaemons()
}
else if (ns.method == MOUNT_UBLK)
{
ns.restoreUblkDaemons()
}
if (ns.method == MOUNT_VDUSE || ns.method == MOUNT_UBLK)
{
dur, err := time.ParseDuration(os.Getenv("RESTART_INTERVAL"))
if (err != nil)
{
@@ -136,7 +143,14 @@ func (ns *NodeServer) restarter()
for
{
<-ticker.C
ns.restoreVduseDaemons()
if (ns.method == MOUNT_VDUSE)
{
ns.restoreVduseDaemons()
}
else if (ns.method == MOUNT_UBLK)
{
ns.restoreUblkDaemons()
}
}
}
@@ -231,6 +245,78 @@ func (ns *NodeServer) checkVduseState(stateFile string, devs map[string]interfac
}
}
func (ns *NodeServer) restoreUblkDaemons()
{
pattern := ns.stateDir+"vitastor-ublk-*.json"
stateFiles, err := filepath.Glob(pattern)
if (err != nil)
{
klog.Errorf("failed to list %s: %v", pattern, err)
}
if (len(stateFiles) == 0)
{
return
}
for _, stateFile := range stateFiles
{
deviceNum := stateFile[len(ns.stateDir) + len("vitastor-ublk-") :]
deviceNum = deviceNum[0:len(deviceNum)-5]
ns.checkUblkState(deviceNum)
}
}
func (ns *NodeServer) checkUblkState(deviceNum string)
{
// Check if the ublk daemon is still active
// Read state file
stateFile := ns.stateDir + "vitastor-ublk-" + deviceNum + ".json"
stateJSON, err := os.ReadFile(stateFile)
if (err != nil)
{
klog.Warningf("error reading state file %v: %v", stateFile, err)
return
}
var state DeviceState
err = json.Unmarshal(stateJSON, &state)
if (err != nil)
{
klog.Warningf("state file %v contains invalid JSON (error %v): %v", stateFile, err, string(stateJSON))
return
}
// Lock volume
ns.lockVolume(state.ConfigPath+":block:"+state.Image)
defer ns.unlockVolume(state.ConfigPath+":block:"+state.Image)
// Recheck state file after locking
_, err = os.ReadFile(stateFile)
if (err != nil)
{
klog.Warningf("state file %v disappeared, skipping volume", stateFile)
return
}
// Check if the vitastor-ublk process is still active
pidFile := ns.stateDir + "vitastor-ublk-" + deviceNum + ".pid"
exists := false
proc, err := findByPidFile(pidFile)
if (err == nil)
{
exists = proc.Signal(syscall.Signal(0)) == nil
}
if (!exists)
{
// Restart daemon
klog.Warningf("recovering UBLK device /dev/ublkb%v for volume %v", deviceNum, state.Image)
_, err = mapUblk(ns.stateDir, state.Image, state.ConfigPath, state.Readonly, "/dev/ublkb"+deviceNum)
if (err != nil)
{
klog.Warningf("failed to recover ublk device for volume %v: %v", state.Image, err)
}
}
}
func (ns *NodeServer) restoreNfsDaemons()
{
pattern := ns.stateDir+"vitastor-nfs-*.json"
@@ -417,14 +503,18 @@ func (ns *NodeServer) NodeStageVolume(ctx context.Context, req *csi.NodeStageVol
}
var devicePath, vdpaId string
if (!ns.useVduse)
if (ns.method == MOUNT_UBLK)
{
devicePath, err = mapNbd(volName, ctxVars, false)
devicePath, err = mapUblk(ns.stateDir, volName, ctxVars["configPath"], false, "")
}
else
else if (ns.method == MOUNT_VDUSE)
{
devicePath, vdpaId, err = mapVduse(ns.stateDir, volName, ctxVars, false)
}
else /* if (ns.method == MOUNT_NBD) */
{
devicePath, err = mapNbd(volName, ctxVars, false)
}
if (err != nil)
{
return nil, err
@@ -439,7 +529,8 @@ func (ns *NodeServer) NodeStageVolume(ctx context.Context, req *csi.NodeStageVol
else
{
// Check existing format
existingFormat, err := diskMounter.GetDiskFormat(devicePath)
var existingFormat string
existingFormat, err = diskMounter.GetDiskFormat(devicePath)
if (err != nil)
{
klog.Errorf("failed to get disk format for path %s, error: %v", err)
@@ -495,10 +586,6 @@ func (ns *NodeServer) NodeStageVolume(ctx context.Context, req *csi.NodeStageVol
case "xfs":
_, err = systemCombined("xfs_growfs", devicePath)
}
if (err != nil)
{
goto unmap
}
}
}
if (err != nil)
@@ -512,14 +599,18 @@ func (ns *NodeServer) NodeStageVolume(ctx context.Context, req *csi.NodeStageVol
return &csi.NodeStageVolumeResponse{}, nil
unmap:
if (!ns.useVduse || len(devicePath) >= 8 && devicePath[0:8] == "/dev/nbd")
if (ns.method == MOUNT_UBLK)
{
unmapNbd(devicePath)
unmapUblk(ns.stateDir, devicePath)
}
else
else if (ns.method == MOUNT_VDUSE)
{
unmapVduseById(ns.stateDir, vdpaId)
}
else /* if (ns.method == MOUNT_NBD) */
{
unmapNbd(devicePath)
}
return nil, err
}
@@ -545,7 +636,7 @@ func (ns *NodeServer) NodeUnstageVolume(ctx context.Context, req *csi.NodeUnstag
defer ns.unlockVolume(ctxVars["configPath"]+":block:"+volName)
targetPath := req.GetStagingTargetPath()
devicePath, _, err := mount.GetDeviceNameFromMount(ns.mounter, targetPath)
devicePath, err := GetDeviceNameFromMount(targetPath)
if (err != nil)
{
if (os.IsNotExist(err))
@@ -582,14 +673,18 @@ func (ns *NodeServer) NodeUnstageVolume(ctx context.Context, req *csi.NodeUnstag
// unmap device
if (len(refList) == 0)
{
if (!ns.useVduse)
if (ns.method == MOUNT_UBLK)
{
unmapNbd(devicePath)
unmapUblk(ns.stateDir, devicePath)
}
else
else if (ns.method == MOUNT_VDUSE)
{
unmapVduse(ns.stateDir, devicePath)
}
else /* if (ns.method == MOUNT_NBD) */
{
unmapNbd(devicePath)
}
}
return &csi.NodeUnstageVolumeResponse{}, nil
@@ -897,7 +992,7 @@ func (ns *NodeServer) NodeUnpublishVolume(ctx context.Context, req *csi.NodeUnpu
}
targetPath := req.GetTargetPath()
devicePath, _, err := mount.GetDeviceNameFromMount(ns.mounter, targetPath)
devicePath, err := GetDeviceNameFromMount(targetPath)
if (err != nil)
{
if (os.IsNotExist(err))
+205 -26
View File
@@ -16,10 +16,20 @@ import (
"syscall"
"k8s.io/klog"
"k8s.io/utils/mount"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/status"
)
type MountMethod int
const (
MOUNT_NBD MountMethod = 0
MOUNT_VDUSE MountMethod = 1
MOUNT_UBLK MountMethod = 2
)
func Contains(list []string, s string) bool
{
for i := 0; i < len(list); i++
@@ -32,29 +42,26 @@ func Contains(list []string, s string) bool
return false
}
func checkVduseSupport() bool
func selectMountMethod() MountMethod
{
// Check UBLK support (ublk_drv kernel module)
if (checkModule("ublk_drv"))
{
klog.Infof("UBLK support enabled successfully")
return MOUNT_UBLK
}
klog.Errorf(
"Your host apparently has no UBLK support. UBLK support disabled."+
" For UBLK you need at least Linux 6.0 and the ublk_drv kernel module.",
)
// Check VDUSE support (vdpa, vduse, virtio-vdpa kernel modules)
vduse := true
for _, mod := range []string{"vdpa", "vduse", "virtio-vdpa"}
{
_, err := os.Stat("/sys/module/"+mod)
if (err != nil)
if (!checkModule(mod))
{
if (!errors.Is(err, os.ErrNotExist))
{
klog.Errorf("failed to check /sys/module/%s: %v", mod, err)
}
c := exec.Command("/sbin/modprobe", mod)
c.Stdout = os.Stderr
c.Stderr = os.Stderr
err := c.Run()
if (err != nil)
{
klog.Errorf("/sbin/modprobe %s failed: %v", mod, err)
vduse = false
break
}
vduse = false
break
}
}
// Check that vdpa tool functions
@@ -69,18 +76,38 @@ func checkVduseSupport() bool
vduse = false
}
}
if (!vduse)
{
klog.Errorf(
"Your host apparently has no VDUSE support. VDUSE support disabled, NBD will be used to map devices."+
" For VDUSE you need at least Linux 5.15 and the following kernel modules: vdpa, virtio-vdpa, vduse.",
)
}
else
if (vduse)
{
klog.Infof("VDUSE support enabled successfully")
return MOUNT_VDUSE
}
return vduse
klog.Errorf(
"Your host apparently has no VDUSE support. VDUSE support disabled, NBD will be used to map devices."+
" For VDUSE you need at least Linux 5.15 and the following kernel modules: vdpa, virtio-vdpa, vduse.",
)
return MOUNT_NBD
}
func checkModule(mod string) bool
{
_, err := os.Stat("/sys/module/"+mod)
if (err != nil)
{
if (!errors.Is(err, os.ErrNotExist))
{
klog.Errorf("failed to check /sys/module/%s: %v", mod, err)
}
c := exec.Command("/sbin/modprobe", mod)
c.Stdout = os.Stderr
c.Stderr = os.Stderr
err := c.Run()
if (err != nil)
{
klog.Errorf("/sbin/modprobe %s failed: %v", mod, err)
return false
}
}
return true
}
func mapNbd(volName string, ctxVars map[string]string, readonly bool) (string, error)
@@ -217,6 +244,7 @@ func mapVduse(stateDir string, volName string, ctxVars map[string]string, readon
stateJSON, _ := json.Marshal(&DeviceState{
ConfigPath: ctxVars["configPath"],
VdpaId: vdpaId,
Image: volName,
Blockdev: blockdev,
Readonly: readonly,
@@ -309,6 +337,117 @@ func unmapVduseById(stateDir, vdpaId string)
}
}
func mapUblk(stateDir string, volName string, configPath string, readonly bool, recoverDev string) (string, error)
{
pidFile := ""
if (recoverDev != "")
{
if (len(recoverDev) < 10 || recoverDev[0:10] != "/dev/ublkb")
{
return "", fmt.Errorf("recover: %s does not start with /dev/ublkb", recoverDev)
}
pidFile = stateDir + "vitastor-ublk-" + recoverDev[10:] + ".pid"
}
else
{
pidFd, err := os.CreateTemp(stateDir, "vitastor-tmp-*.pid")
if (err != nil)
{
return "", err
}
pidFile = pidFd.Name()
pidFd.Close()
}
// Map device via vitastor-ublk
args := []string{
"map", "--image", volName, "--pidfile", pidFile,
}
if (configPath != "")
{
args = append(args, "--config_path", configPath)
}
if (readonly)
{
args = append(args, "--readonly")
}
if (recoverDev != "")
{
args = append(args, "--recover", recoverDev)
}
stdout, stderr, err := system("/usr/bin/vitastor-ublk", args...)
if (err != nil)
{
return "", err
}
devicePath := strings.TrimSpace(string(stdout))
if (devicePath == "")
{
return "", fmt.Errorf("vitastor-ublk did not return the name of the device. output: %s", stderr)
}
if (len(devicePath) >= 10 && devicePath[0:10] == "/dev/ublkb")
{
// Generate state file
devNum := devicePath[10:]
pidNew := stateDir + "vitastor-ublk-" + devNum + ".pid"
if (pidFile != pidNew)
{
err := os.Rename(pidFile, pidNew)
if (err != nil)
{
klog.Errorf("Failed to rename PID file %s to %s: %v", pidFile, pidNew, err)
}
else
{
pidFile = pidNew
}
}
stateFile := stateDir + "vitastor-ublk-" + devNum + ".json"
stateJSON, _ := json.Marshal(&DeviceState{
ConfigPath: configPath,
Image: volName,
Readonly: readonly,
PidFile: pidFile,
})
err = os.WriteFile(stateFile, stateJSON, 0600)
if (err == nil)
{
klog.Infof("Attached volume %s via UBLK as %s", volName, devicePath)
return devicePath, nil
}
os.Remove(stateFile)
}
killErr := killByPidFile(pidFile)
if (killErr != nil)
{
klog.Errorf("Failed to kill started vitastor-ublk: %v", killErr)
}
os.Remove(pidFile)
return "", err
}
func unmapUblk(stateDir, devicePath string)
{
if (len(devicePath) < 10 || devicePath[0:10] != "/dev/ublkb")
{
klog.Errorf("%s does not start with /dev/ublkb", devicePath)
return
}
unmapOut, unmapErr := exec.Command("/usr/bin/vitastor-ublk", "unmap", devicePath).CombinedOutput()
if (unmapErr != nil)
{
klog.Errorf("failed to unmap UBLK device %s: %s, error: %v", devicePath, unmapOut, unmapErr)
}
for _, ext := range []string{"json", "pid"}
{
fn := stateDir + "vitastor-ublk-" + devicePath[10:] + "." + ext
err := os.Remove(fn)
if (err != nil)
{
klog.Errorf("failed to remove %s: %v", fn, err)
}
}
}
func system(program string, args ...string) ([]byte, []byte, error)
{
klog.Infof("Running "+program+" "+strings.Join(args, " "))
@@ -340,3 +479,43 @@ func systemCombined(program string, args ...string) ([]byte, error)
}
return out.Bytes(), nil
}
func GetDeviceNameFromMount(mountPath string) (string, error)
{
// Use /proc/self/mountinfo to correctly parse bind mounts for block device files
mps, err := mount.ParseMountInfo("/proc/self/mountinfo")
if (err != nil)
{
return "", err
}
slTarget, err := filepath.EvalSymlinks(mountPath)
if (err != nil)
{
slTarget = mountPath
}
device := ""
for _, mp := range mps
{
if (mp.MountPoint == slTarget)
{
device = mp.Source
if (device[0] != '/' && mp.Root != "/")
{
// Handle {Source=udev Root=/vdb MountPoint=/var/lib/kubelet/tralaleylo/tralala}
for _, other := range mps
{
if (other.Root == "/" && other.Source == mp.Source)
{
device = other.MountPoint + mp.Root
break
}
}
}
break
}
}
return device, nil
}
+2 -5
View File
@@ -1,7 +1,4 @@
#!/bin/bash
cat < vitastor.Dockerfile > ../Dockerfile
cd ..
mkdir -p packages
sudo podman build --build-arg DISTRO=debian --build-arg REL=bookworm -v `pwd`/packages:/root/packages -f Dockerfile .
rm Dockerfile
docker build --build-arg DISTRO=debian --build-arg REL=bookworm -t vitastor-buildenv:bookworm -f vitastor-buildenv.Dockerfile .
docker run -it --rm -e REL=bookworm -v `dirname $0`/../:/root/vitastor vitastor-buildenv:bookworm /root/vitastor/debian/vitastor-build.sh
+2 -5
View File
@@ -1,7 +1,4 @@
#!/bin/bash
cat < vitastor.Dockerfile > ../Dockerfile
cd ..
mkdir -p packages
sudo podman build --build-arg DISTRO=debian --build-arg REL=bullseye -v `pwd`/packages:/root/packages -f Dockerfile .
rm Dockerfile
docker build --build-arg DISTRO=debian --build-arg REL=bullseye -t vitastor-buildenv:bullseye -f vitastor-buildenv.Dockerfile .
docker run -it --rm -e REL=bullseye -v `dirname $0`/../:/root/vitastor vitastor-buildenv:bullseye /root/vitastor/debian/vitastor-build.sh
+2 -5
View File
@@ -1,7 +1,4 @@
#!/bin/bash
cat < vitastor.Dockerfile > ../Dockerfile
cd ..
mkdir -p packages
sudo podman build --build-arg DISTRO=debian --build-arg REL=buster -v `pwd`/packages:/root/packages -f Dockerfile .
rm Dockerfile
docker build --build-arg DISTRO=debian --build-arg REL=buster -t vitastor-buildenv:buster -f vitastor-buildenv.Dockerfile .
docker run -it --rm -e REL=buster -v `dirname $0`/../:/root/vitastor vitastor-buildenv:buster /root/vitastor/debian/vitastor-build.sh
+4
View File
@@ -0,0 +1,4 @@
#!/bin/bash
docker build --build-arg DISTRO=debian --build-arg REL=trixie -t vitastor-buildenv:trixie -f vitastor-buildenv.Dockerfile .
docker run -it --rm -e REL=trixie -v `dirname $0`/../:/root/vitastor vitastor-buildenv:trixie /root/vitastor/debian/vitastor-build.sh
+3 -5
View File
@@ -1,7 +1,5 @@
#!/bin/bash
# Ubuntu 22.04 Jammy Jellyfish
cat < vitastor.Dockerfile > ../Dockerfile
cd ..
mkdir -p packages
sudo podman build --build-arg DISTRO=ubuntu --build-arg REL=jammy -v `pwd`/packages:/root/packages -f Dockerfile .
rm Dockerfile
docker build --build-arg DISTRO=ubuntu --build-arg REL=jammy -t vitastor-buildenv:jammy -f vitastor-buildenv.Dockerfile .
docker run -it --rm -e REL=jammy -v `dirname $0`/../:/root/vitastor vitastor-buildenv:jammy /root/vitastor/debian/vitastor-build.sh
+5
View File
@@ -0,0 +1,5 @@
#!/bin/bash
# 24.04 Noble Numbat
docker build --build-arg DISTRO=ubuntu --build-arg REL=noble -t vitastor-buildenv:noble -f vitastor-buildenv.Dockerfile .
docker run -it --rm -e REL=noble -v `dirname $0`/../:/root/vitastor vitastor-buildenv:noble /root/vitastor/debian/vitastor-build.sh
+5
View File
@@ -0,0 +1,5 @@
#!/bin/bash
# 25.10 Questing quokka
docker build --build-arg DISTRO=ubuntu --build-arg REL=questing -t vitastor-buildenv:questing -f vitastor-buildenv.Dockerfile .
docker run -it --rm -e REL=questing -v `dirname $0`/../:/root/vitastor vitastor-buildenv:questing /root/vitastor/debian/vitastor-build.sh
+5
View File
@@ -0,0 +1,5 @@
#!/bin/bash
# 26.04 Resolute Raccoon
docker build --build-arg DISTRO=ubuntu --build-arg REL=resolute -t vitastor-buildenv:resolute -f vitastor-buildenv.Dockerfile .
docker run -it --rm -e REL=resolute -v `dirname $0`/../:/root/vitastor vitastor-buildenv:resolute /root/vitastor/debian/vitastor-build.sh
+1 -1
View File
@@ -1,4 +1,4 @@
vitastor (1.11.0-1) unstable; urgency=medium
vitastor (3.0.8-1) unstable; urgency=medium
* Bugfixes
+2 -2
View File
@@ -2,9 +2,9 @@ Source: vitastor
Section: admin
Priority: optional
Maintainer: Vitaliy Filippov <vitalif@yourcmc.ru>
Build-Depends: debhelper, liburing-dev (>= 0.6), g++ (>= 8), libstdc++6 (>= 8),
Build-Depends: debhelper, g++ (>= 8), libstdc++6 (>= 8),
linux-libc-dev, libgoogle-perftools-dev, libjerasure-dev, libgf-complete-dev,
libibverbs-dev, libisal-dev, cmake, pkg-config, libnl-3-dev, libnl-genl-3-dev,
libibverbs-dev, librdmacm-dev, libisal-dev, cmake, pkg-config, libnl-3-dev, libnl-genl-3-dev,
node-bindings <!nocheck>, node-gyp, node-nan
Standards-Version: 4.5.0
Homepage: https://vitastor.io/
+8 -4
View File
@@ -10,10 +10,14 @@ ARG REL=
WORKDIR /root
RUN if [ "$REL" = "buster" -o "$REL" = "bullseye" -o "$REL" = "bookworm" ]; then \
echo "deb http://deb.debian.org/debian $REL-backports main" >> /etc/apt/sources.list; \
if [ "$REL" = "buster" ]; then \
echo "deb http://archive.debian.org/debian $REL-backports main" >> /etc/apt/sources.list; \
else \
echo "deb http://deb.debian.org/debian $REL-backports main" >> /etc/apt/sources.list; \
fi; \
echo >> /etc/apt/preferences; \
echo 'Package: *' >> /etc/apt/preferences; \
echo "Pin: release a=$REL-backports" >> /etc/apt/preferences; \
echo "Pin: release n=$REL-backports" >> /etc/apt/preferences; \
echo 'Pin-Priority: 500' >> /etc/apt/preferences; \
fi; \
grep '^deb ' /etc/apt/sources.list | perl -pe 's/^deb/deb-src/' >> /etc/apt/sources.list; \
@@ -22,7 +26,7 @@ RUN if [ "$REL" = "buster" -o "$REL" = "bullseye" -o "$REL" = "bookworm" ]; then
echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
RUN apt-get update
RUN DEBIAN_FRONTEND=noninteractive TZ=Europe/Moscow apt-get -y install fio liburing-dev libgoogle-perftools-dev devscripts
RUN DEBIAN_FRONTEND=noninteractive TZ=Europe/Moscow apt-get -y install fio libgoogle-perftools-dev devscripts
RUN DEBIAN_FRONTEND=noninteractive TZ=Europe/Moscow apt-get -y build-dep qemu
# To build a custom version
#RUN cp /root/packages/qemu-orig/* /root
@@ -56,7 +60,7 @@ RUN set -e; \
quilt add block/vitastor.c; \
cp /root/qemu_driver.c block/vitastor.c; \
quilt refresh; \
V=$(head -n1 debian/changelog | perl -pe 's/5\.2\+dfsg-9/5.2+dfsg-11/; s/^.*\((.*?)(~bpo[\d\+]*)?\).*$/$1/')+vitastor4; \
V=$(head -n1 debian/changelog | perl -pe 's/5\.2\+dfsg-9/5.2+dfsg-11/; s/^.*\((.*?)(\+deb\d+u\d+)?(~bpo[\d\+]*)?\).*$/$1/')+vitastor5; \
if [ "$REL" = bullseye ]; then V=${V}bullseye; fi; \
DEBEMAIL="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v $V 'Plug Vitastor block driver'; \
DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \
Vendored Executable
+65
View File
@@ -0,0 +1,65 @@
#!/bin/bash
# To be ran inside buildenv docker
set -e -x
[ -e /usr/lib/x86_64-linux-gnu/pkgconfig/libisal.pc ] || cp /root/vitastor/debian/libisal.pc /usr/lib/x86_64-linux-gnu/pkgconfig
mkdir -p /root/fio-build/
cd /root/fio-build/
rm -rf /root/fio-build/*
dpkg-source -x /root/fio*.dsc
FULLVER=`head -n1 /root/vitastor/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'`
VER=${FULLVER%%-*}
rm -rf /root/vitastor-$VER
mkdir /root/vitastor-$VER
cd /root/vitastor
cp -a $(ls | grep -v packages) /root/vitastor-$VER
rm -rf /root/vitastor/packages/vitastor-$REL
mkdir -p /root/vitastor/packages/vitastor-$REL
mv /root/vitastor-$VER /root/vitastor/packages/vitastor-$REL/
cd /root/vitastor/packages/vitastor-$REL/vitastor-$VER
rm -rf fio
ln -s /root/fio-build/fio-*/ ./fio
FIO=`head -n1 fio/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'`
ls /usr/include/linux/raw.h || cp ./debian/raw.h /usr/include/linux/raw.h
sh copy-fio-includes.sh
rm fio
mkdir -p a b debian/patches
mv fio-copy b/fio
diff -NaurpbB a b > debian/patches/fio-headers.patch || true
echo fio-headers.patch >> debian/patches/series
rm -rf a b
echo "dep:fio=$FIO" > debian/fio_version
cd /root/vitastor/packages/vitastor-$REL/vitastor-$VER
mkdir mon/node_modules
cd mon/node_modules
curl -s https://git.yourcmc.ru/vitalif/antietcd/archive/master.tar.gz | tar -zx
curl -s https://git.yourcmc.ru/vitalif/tinyraft/archive/master.tar.gz | tar -zx
cd /root/vitastor/packages/vitastor-$REL
if [[ ( "$REL" = "trixie" || "$REL" = "resolute" ) && -e ../vitastor-bookworm/vitastor_$VER.orig.tar.xz ]]; then
# Fucking shit, archives differ between bookworm (xz 5.4.1) and trixie (xz 5.8.1)
cp ../vitastor-bookworm/vitastor_$VER.orig.tar.xz .
else
tar --sort=name --mtime='2020-01-01' --owner=0 --group=0 --exclude=debian -cJf vitastor_$VER.orig.tar.xz vitastor-$VER
fi
cd vitastor-$VER
DEBEMAIL="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v "$FULLVER""$REL" "Rebuild for $REL"
DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa
rm -rf /root/vitastor/packages/vitastor-$REL/vitastor-*/
# Why does ubuntu rename debug packages to *.ddeb?
cd /root/vitastor/packages/vitastor-$REL
if ls *.ddeb >/dev/null; then
perl -i -pe 's/\.ddeb/.deb/' *.buildinfo *.changes
for i in *.ddeb; do
mv $i ${i%%.ddeb}.deb
done
fi
+31
View File
@@ -0,0 +1,31 @@
# Build environment for building Vitastor packages for Debian inside a container
# cd ..
# docker build --build-arg DISTRO=debian --build-arg REL=bullseye -f debian/vitastor.Dockerfile -t vitastor-buildenv:bullseye .
# docker run --rm -e REL=bullseye -v ./:/root/vitastor /root/vitastor/debian/vitastor-build.sh
ARG DISTRO=debian
ARG REL=
FROM $DISTRO:$REL
ARG DISTRO=debian
ARG REL=
WORKDIR /root
RUN set -e -x; \
if [ "$REL" = "buster" ]; then \
perl -i -pe 's/deb.debian.org/archive.debian.org/' /etc/apt/sources.list; \
apt-get update; \
apt-get -y install wget; \
wget https://vitastor.io/debian/pubkey.gpg -O /etc/apt/trusted.gpg.d/vitastor.gpg; \
echo "deb https://vitastor.io/debian $REL main" >> /etc/apt/sources.list; \
fi; \
grep '^deb ' /etc/apt/sources.list | perl -pe 's/^deb/deb-src/' >> /etc/apt/sources.list; \
perl -i -pe 's/Types: deb$/Types: deb deb-src/' /etc/apt/sources.list.d/*.sources || true; \
echo 'APT::Install-Recommends false;' >> /etc/apt/apt.conf; \
echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
RUN apt-get update && \
apt-get -y install fio libgoogle-perftools-dev devscripts libjerasure-dev cmake \
libibverbs-dev librdmacm-dev libisal-dev libnl-3-dev libnl-genl-3-dev curl nodejs npm node-nan node-bindings && \
apt-get -y build-dep fio && \
apt-get --download-only source fio
+1
View File
@@ -2,6 +2,7 @@ usr/bin/vita
usr/bin/vitastor-cli
usr/bin/vitastor-rm
usr/bin/vitastor-nbd
usr/bin/vitastor-ublk
usr/bin/vitastor-nfs
usr/bin/vitastor-kv
usr/bin/vitastor-kv-stress
-65
View File
@@ -1,65 +0,0 @@
# Build Vitastor packages for Debian inside a container
# cd ..; podman build --build-arg DISTRO=debian --build-arg REL=bullseye -v `pwd`/packages:/root/packages -f debian/vitastor.Dockerfile .
ARG DISTRO=debian
ARG REL=
FROM $DISTRO:$REL
ARG DISTRO=debian
ARG REL=
WORKDIR /root
RUN set -e -x; \
if [ "$REL" = "buster" ]; then \
apt-get update; \
apt-get -y install wget; \
wget https://vitastor.io/debian/pubkey.gpg -O /etc/apt/trusted.gpg.d/vitastor.gpg; \
echo "deb https://vitastor.io/debian $REL main" >> /etc/apt/sources.list; \
fi; \
grep '^deb ' /etc/apt/sources.list | perl -pe 's/^deb/deb-src/' >> /etc/apt/sources.list; \
perl -i -pe 's/Types: deb$/Types: deb deb-src/' /etc/apt/sources.list.d/debian.sources || true; \
echo 'APT::Install-Recommends false;' >> /etc/apt/apt.conf; \
echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
RUN apt-get update && \
apt-get -y install fio liburing-dev libgoogle-perftools-dev devscripts libjerasure-dev cmake \
libibverbs-dev librdmacm-dev libisal-dev libnl-3-dev libnl-genl-3-dev curl nodejs npm node-nan node-bindings && \
apt-get -y build-dep fio && \
apt-get --download-only source fio
ADD . /root/vitastor
RUN set -e -x; \
[ -e /usr/lib/x86_64-linux-gnu/pkgconfig/libisal.pc ] || cp /root/vitastor/debian/libisal.pc /usr/lib/x86_64-linux-gnu/pkgconfig; \
mkdir -p /root/fio-build/; \
cd /root/fio-build/; \
rm -rf /root/fio-build/*; \
dpkg-source -x /root/fio*.dsc; \
mkdir -p /root/packages/vitastor-$REL; \
rm -rf /root/packages/vitastor-$REL/*; \
cd /root/packages/vitastor-$REL; \
FULLVER=$(head -n1 /root/vitastor/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
VER=${FULLVER%%-*}; \
cp -r /root/vitastor vitastor-$VER; \
cd vitastor-$VER; \
ln -s /root/fio-build/fio-*/ ./fio; \
FIO=$(head -n1 fio/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
ls /usr/include/linux/raw.h || cp ./debian/raw.h /usr/include/linux/raw.h; \
sh copy-fio-includes.sh; \
rm fio; \
mkdir -p a b debian/patches; \
mv fio-copy b/fio; \
diff -NaurpbB a b > debian/patches/fio-headers.patch || true; \
echo fio-headers.patch >> debian/patches/series; \
rm -rf a b; \
echo "dep:fio=$FIO" > debian/fio_version; \
cd /root/packages/vitastor-$REL/vitastor-$VER; \
mkdir mon/node_modules; \
cd mon/node_modules; \
curl -s https://git.yourcmc.ru/vitalif/antietcd/archive/master.tar.gz | tar -zx; \
curl -s https://git.yourcmc.ru/vitalif/tinyraft/archive/master.tar.gz | tar -zx; \
cd /root/packages/vitastor-$REL; \
tar --sort=name --mtime='2020-01-01' --owner=0 --group=0 --exclude=debian -cJf vitastor_$VER.orig.tar.xz vitastor-$VER; \
cd vitastor-$VER; \
DEBFULLNAME="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v "$FULLVER""$REL" "Rebuild for $REL"; \
DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \
rm -rf /root/packages/vitastor-$REL/vitastor-*/
+2 -2
View File
@@ -1,9 +1,9 @@
# Build Docker image with Vitastor packages
FROM debian:bookworm
FROM debian:trixie
ADD etc/apt /etc/apt/
RUN apt-get update && apt-get -y install vitastor qemu-system-x86 qemu-system-common qemu-block-extra qemu-utils jq nfs-common && apt-get clean
RUN apt-get update && apt-get -y install vitastor ibverbs-providers udev systemd qemu-system-x86 qemu-system-common qemu-block-extra qemu-utils jq nfs-common && apt-get clean
ADD sleep.sh /usr/bin/
ADD install.sh /usr/bin/
ADD scripts /opt/scripts/
+2 -2
View File
@@ -1,9 +1,9 @@
VITASTOR_VERSION ?= v1.11.0
VITASTOR_VERSION ?= v3.0.8
all: build push
build:
@docker build --rm -t vitalif/vitastor:$(VITASTOR_VERSION) .
@docker build --no-cache --rm -t vitalif/vitastor:$(VITASTOR_VERSION) .
push:
@docker push vitalif/vitastor:$(VITASTOR_VERSION)
+3
View File
@@ -0,0 +1,3 @@
Package: *
Pin: release n=trixie-backports
Pin-Priority: 500
+2 -1
View File
@@ -1 +1,2 @@
deb http://vitastor.io/debian bookworm main
deb http://vitastor.io/debian trixie main
#deb http://http.debian.net/debian/ trixie-backports main
@@ -7,8 +7,8 @@ PartOf=vitastor.target
[Service]
Restart=always
EnvironmentFile=/etc/vitastor/docker.conf
ExecStart=bash -c 'docker run --rm -i -v /etc/vitastor:/etc/vitastor -v /dev:/dev \
--privileged --log-driver none --network host --name vitastor vitastor:$VITASTOR_VERSION \
ExecStart=bash -c 'docker run --rm -i -v /etc/vitastor:/etc/vitastor -v /dev:/dev -v /run:/run -e SYSTEMD_IN_CHROOT=0 \
--security-opt seccomp=unconfined --privileged --pid=host --log-driver none --network host --name vitastor vitastor:$VITASTOR_VERSION \
sleep.sh'
ExecStartPost=udevadm trigger
ExecStop=docker stop vitastor
@@ -12,7 +12,8 @@ EnvironmentFile=/etc/vitastor/docker.conf
SyslogIdentifier=vitastor-osd%i
ExecStart=bash -c 'docker run --rm -i -v /etc/vitastor:/etc/vitastor -v /dev:/dev \
$(for i in $(ls /dev/vitastor/osd%i-*); do echo --device $i:$i; done) \
--log-driver none --network host --ulimit nofile=1048576 --ulimit memlock=-1 $CONTAINER_OPTIONS --name vitastor-osd%i \
--log-driver none --network host --ulimit nofile=1048576 --ulimit memlock=-1 \
--security-opt seccomp=unconfined $CONTAINER_OPTIONS --name vitastor-osd%i \
vitastor:$VITASTOR_VERSION vitastor-disk exec-osd /dev/vitastor/osd%i-data'
ExecStartPre=+docker exec vitastor vitastor-disk pre-exec /dev/vitastor/osd%i-data
ExecStop=docker stop vitastor-etcd%i
+1 -1
View File
@@ -4,7 +4,7 @@
#
# Desired Vitastor version
VITASTOR_VERSION=1.11.0
VITASTOR_VERSION=v3.0.8
# Additional arguments for all containers
# For example, you may want to specify a custom logging driver here
+2 -3
View File
@@ -2,8 +2,7 @@
set -e
cp -urv /etc/default /host-etc/
cp -urv /etc/systemd /host-etc/
cp -urv /etc/udev /host-etc/
cp -urv /etc/systemd/system/vitastor* /host-etc/systemd/system/
cp -urv /etc/udev/rules.d /host-etc/udev/
cp -urnv /etc/vitastor /host-etc/
cp -urnv /opt/scripts/* /host-bin/
+52
View File
@@ -13,6 +13,7 @@ affect their interaction with the cluster.
- [client_retry_interval](#client_retry_interval)
- [client_eio_retry_interval](#client_eio_retry_interval)
- [client_retry_enospc](#client_retry_enospc)
- [client_wait_up_timeout](#client_wait_up_timeout)
- [client_max_dirty_bytes](#client_max_dirty_bytes)
- [client_max_dirty_ops](#client_max_dirty_ops)
- [client_enable_writeback](#client_enable_writeback)
@@ -23,6 +24,10 @@ affect their interaction with the cluster.
- [nbd_max_devices](#nbd_max_devices)
- [nbd_max_part](#nbd_max_part)
- [osd_nearfull_ratio](#osd_nearfull_ratio)
- [hostname](#hostname)
- [ublk_queue_depth](#ublk_queue_depth)
- [ublk_max_io_size](#ublk_max_io_size)
- [qemu_file_mirror_path](#qemu_file_mirror_path)
## client_iothread_count
@@ -70,6 +75,19 @@ and clients are not blocked and just get EIO error code instead.
Retry writes on out of space errors to wait until some space is freed on
OSDs.
## client_wait_up_timeout
- Type: seconds
- Default: 16
- Can be changed online: yes
Wait for this number of seconds until PGs are up when doing operations
which require all PGs to be up. Currently only used by object listings
in delete and merge-based commands ([vitastor-cli rm](../usage/cli.en.md#rm), merge and so on).
The default value is calculated as `1 + OSD lease timeout`, which is
`1 + etcd_report_interval + max_etcd_attempts*2*etcd_quick_timeout`.
## client_max_dirty_bytes
- Type: integer
@@ -201,3 +219,37 @@ just one OSD becomes 100 % full!
However, unlike in Ceph, 100 % full Vitastor OSDs don't crash (in Ceph they're
unable to start at all), so you'll be able to recover from "out of space" errors
without destroying and recreating OSDs.
## hostname
- Type: string
- Can be changed online: yes
Clients use host name to find their distance to OSDs when [localized reads](pool.en.md#local_reads)
are enabled. By default, standard [gethostname](https://man7.org/linux/man-pages/man2/gethostname.2.html)
function is used to determine host name, but you can also override it with this parameter.
## ublk_queue_depth
- Type: integer
- Default: 256
Default queue depth for [Vitastor ublk servers](../usage/ublk.en.md).
## ublk_max_io_size
- Type: integer
Default maximum I/O size for Vitastor [ublk servers](../usage/ublk.en.md).
The largest of 1 MB and pool block size multiplied by EC data chunk count is used if not specified.
## qemu_file_mirror_path
- Type: string
When set to an FS directory path (for example, `/mnt/vitastor/`), `qemu-img info` and similar
QAPI commands return the name of the image inside this directory instead of normal
`vitastor://?image=abc` URI as `filename`.
This allows to then mount this path using [vitastor-nfs](../usage/nfs.en.md) and trick
third-party systems like Veeam which rely on `filename` in the image info but don't support Vitastor.
+55
View File
@@ -13,6 +13,7 @@
- [client_retry_interval](#client_retry_interval)
- [client_eio_retry_interval](#client_eio_retry_interval)
- [client_retry_enospc](#client_retry_enospc)
- [client_wait_up_timeout](#client_wait_up_timeout)
- [client_max_dirty_bytes](#client_max_dirty_bytes)
- [client_max_dirty_ops](#client_max_dirty_ops)
- [client_enable_writeback](#client_enable_writeback)
@@ -23,6 +24,10 @@
- [nbd_max_devices](#nbd_max_devices)
- [nbd_max_part](#nbd_max_part)
- [osd_nearfull_ratio](#osd_nearfull_ratio)
- [hostname](#hostname)
- [ublk_queue_depth](#ublk_queue_depth)
- [ublk_max_io_size](#ublk_max_io_size)
- [qemu_file_mirror_path](#qemu_file_mirror_path)
## client_iothread_count
@@ -72,6 +77,19 @@ RDMA и хотите повысить пиковую производитель
Повторять запросы записи, завершившиеся с ошибками нехватки места, т.е.
ожидать, пока на OSD не освободится место.
## client_wait_up_timeout
- Тип: секунды
- Значение по умолчанию: 16
- Можно менять на лету: да
Время ожидания поднятия PG при операциях, требующих активности всех PG.
В данный момент используется листингами объектов в командах, использующих
удаление и слияние ([vitastor-cli rm](../usage/cli.ru.md#rm), merge и подобные).
Значение по умолчанию вычисляется как `1 + время lease OSD`, равное
`1 + etcd_report_interval + max_etcd_attempts*2*etcd_quick_timeout`.
## client_max_dirty_bytes
- Тип: целое число
@@ -205,3 +223,40 @@ RDMA и хотите повысить пиковую производитель
заполненные на 100% OSD вообще не могут стартовать), так что вы сможете
восстановить работу кластера после ошибок отсутствия свободного места
без уничтожения и пересоздания OSD.
## hostname
- Тип: строка
- Можно менять на лету: да
Клиенты используют имя хоста для определения расстояния до OSD, когда включены
[локальные чтения](pool.ru.md#local_reads). По умолчанию для определения имени
хоста используется стандартная функция [gethostname](https://man7.org/linux/man-pages/man2/gethostname.2.html),
но вы также можете задать имя хоста вручную данным параметром.
## ublk_queue_depth
- Тип: целое число
- Значение по умолчанию: 256
Глубина очереди по умолчанию для [ublk-серверов Vitastor](../usage/ublk.ru.md).
## ublk_max_io_size
- Тип: целое число
Максимальный размер запроса ввода-вывода для [ublk-серверов Vitastor](../usage/ublk.ru.md).
Если не задан, используется максимум из 1 МБ и размера блока пула, умноженного на число частей
данных EC-пула.
## qemu_file_mirror_path
- Тип: строка
Если установить эту опцию равной пути к каталогу в ФС, команда `qemu-img info` и подобные
команды QAPI будут возвращать в поле `filename` имя образа внутри заданного каталога вместо
обычного адреса типа `vitastor://?image=abc`.
Это позволяет смонтировать этот путь с помощью [vitastor-nfs](../usage/nfs.ru.md) и обмануть
сторонние системы типа Veeam, которые полагаются на поле `filename` в информации об образе QEMU,
но не поддерживают Vitastor.
+16
View File
@@ -9,6 +9,7 @@
These parameters apply to OSDs, are fixed at the moment of OSD drive
initialization and can't be changed after it without losing data.
- [meta_format](#meta_format)
- [data_device](#data_device)
- [meta_device](#meta_device)
- [journal_device](#journal_device)
@@ -27,6 +28,21 @@ initialization and can't be changed after it without losing data.
- [data_csum_type](#data_csum_type)
- [csum_block_size](#csum_block_size)
## meta_format
- Type: integer
- Default: 3
OSD store implementation version and on-disk metadata format.
Three versions are currently supported: 3, 2 and 1.
- 3 the new log-structured store, it's overall faster, has lower Write
Amplification, which may be even close to 1 (i.e. almost no extra writes)
if your SSDs support atomic writes (see [atomic_write_size](osd.en.md#atomic_write_size)).
- 2 is the old stable store from Vitastor 0.9-2.x.
- 1 is the same old store but with a legacy metadata format from Vitastor
versions to up 0.8.x, without any support for checksums.
## data_device
- Type: string
+18
View File
@@ -10,6 +10,7 @@
дисковые параметры, задаются в момент инициализации дисков OSD и не могут быть
изменены после этого без потери данных.
- [meta_format](#meta_format)
- [data_device](#data_device)
- [meta_device](#meta_device)
- [journal_device](#journal_device)
@@ -28,6 +29,23 @@
- [data_csum_type](#data_csum_type)
- [csum_block_size](#csum_block_size)
## meta_format
- Тип: целое число
- Значение по умолчанию: 3
Версия реализации дискового хранилища OSD и дискового формата метаданных.
Поддерживаются три версии: 3, 2 и 1.
- 3 - новое лог-структурированное хранилище, в целом более быстрое, со
сниженным фактором амплификации записи, который может составлять около 1
(то есть, практически без лишней служебной записи), если ваши SSD
поддерживают атомарную запись (см. [atomic_write_size](osd.ru.md#atomic_write_size)).
- 2 - старое стабильное хранилище из версий Vitastor 0.9-2.x.
- 1 - то же самое стабильное хранилище, но с ещё более старым форматом
метаданных из версий Vitastor до 0.8.x, без какой-либо поддержки
контрольных сумм.
## data_device
- Тип: строка
+2 -2
View File
@@ -74,13 +74,13 @@ Grafana dashboard suitable for this exporter is here: [Vitastor-Grafana-6+.json]
- Type: integer
- Default: 8060
HTTP port for monitors to listen on (including metrics exporter)
HTTP port for monitors to listen to (including metrics exporter)
## mon_http_ip
- Type: string
IP address for monitors to listen on (all addresses by default)
IP address for monitors to listen to (all addresses by default)
## mon_https_cert
+132 -54
View File
@@ -9,9 +9,11 @@
These parameters apply to clients and OSDs and affect network connection logic
between clients, OSDs and etcd.
- [tcp_header_buffer_size](#tcp_header_buffer_size)
- [use_sync_send_recv](#use_sync_send_recv)
- [osd_network](#osd_network)
- [osd_cluster_network](#osd_cluster_network)
- [use_rdma](#use_rdma)
- [use_rdmacm](#use_rdmacm)
- [disable_tcp](#disable_tcp)
- [rdma_device](#rdma_device)
- [rdma_port_num](#rdma_port_num)
- [rdma_gid_index](#rdma_gid_index)
@@ -20,7 +22,6 @@ between clients, OSDs and etcd.
- [rdma_max_msg](#rdma_max_msg)
- [rdma_max_recv](#rdma_max_recv)
- [rdma_max_send](#rdma_max_send)
- [rdma_odp](#rdma_odp)
- [peer_connect_interval](#peer_connect_interval)
- [peer_connect_timeout](#peer_connect_timeout)
- [osd_idle_timeout](#osd_idle_timeout)
@@ -30,38 +31,63 @@ between clients, OSDs and etcd.
- [etcd_slow_timeout](#etcd_slow_timeout)
- [etcd_keepalive_timeout](#etcd_keepalive_timeout)
- [etcd_ws_keepalive_interval](#etcd_ws_keepalive_interval)
- [etcd_min_reload_interval](#etcd_min_reload_interval)
- [tcp_header_buffer_size](#tcp_header_buffer_size)
- [min_zerocopy_send_size](#min_zerocopy_send_size)
- [use_sync_send_recv](#use_sync_send_recv)
## tcp_header_buffer_size
## osd_network
- Type: integer
- Default: 65536
- Type: string or array of strings
Size of the buffer used to read data using an additional copy. Vitastor
packet headers are 128 bytes, payload is always at least 4 KB, so it is
usually beneficial to try to read multiple packets at once even though
it requires to copy the data an additional time. The rest of each packet
is received without an additional copy. You can try to play with this
parameter and see how it affects random iops and linear bandwidth if you
want.
Network mask of public OSD network(s) (IPv4 or IPv6). Each OSD listens to all
addresses of UP + RUNNING interfaces matching one of these networks, on the
same port. Port is auto-selected except if [bind_port](osd.en.md#bind_port) is
explicitly specified. Bind address(es) may also be overridden manually by
specifying [bind_address](osd.en.md#bind_address). If OSD networks are not specified
at all, OSD just listens to a wildcard address (0.0.0.0).
## use_sync_send_recv
## osd_cluster_network
- Type: boolean
- Default: false
- Type: string or array of strings
If true, synchronous send/recv syscalls are used instead of io_uring for
socket communication. Useless for OSDs because they require io_uring anyway,
but may be required for clients with old kernel versions.
Network mask of separate network(s) (IPv4 or IPv6) to use for OSD
cluster connections. I.e. OSDs will always attempt to use these networks
to connect to other OSDs, while clients will attempt to use networks from
[osd_network](#osd_network).
## use_rdma
- Type: boolean
- Default: true
Try to use RDMA for communication if it's available. Disable if you don't
want Vitastor to use RDMA. TCP-only clients can also talk to an RDMA-enabled
cluster, so disabling RDMA may be needed if clients have RDMA devices,
but they are not connected to the cluster.
Try to use RDMA through libibverbs for communication if it's available.
Disable if you don't want Vitastor to use RDMA. TCP-only clients can also
talk to an RDMA-enabled cluster, so disabling RDMA may be needed if clients
have RDMA devices, but they are not connected to the cluster.
`use_rdma` works with RoCEv1/RoCEv2 networks, but not with iWARP and,
maybe, with some Infiniband configurations which require RDMA-CM.
Consider `use_rdmacm` for such networks.
## use_rdmacm
- Type: boolean
- Default: false
Use an alternative implementation of RDMA through RDMA-CM (Connection
Manager). Works with all RDMA networks: Infiniband, iWARP and
RoCEv1/RoCEv2, and even allows to disable TCP and run only with RDMA.
OSDs always use random port numbers for RDMA-CM listeners, different
from their TCP ports. `use_rdma` is automatically disabled when
`use_rdmacm` is enabled.
## disable_tcp
- Type: boolean
- Default: true
Fully disable TCP and only use RDMA-CM for OSD communication.
## rdma_device
@@ -75,11 +101,6 @@ found or if `osd_network` is not specified. Auto-selection is also
unsupported with old libibverbs < v32, like in Debian 10 Buster or
CentOS 7.
Vitastor supports all adapters, even ones without ODP support, like
Mellanox ConnectX-3 and non-Mellanox cards. Versions up to Vitastor
1.2.0 required ODP which is only present in Mellanox ConnectX >= 4.
See also [rdma_odp](#rdma_odp).
Run `ibv_devinfo -v` as root to list available RDMA devices and their
features.
@@ -89,15 +110,33 @@ the manual of your network vendor for details about setting up the switch
for RoCEv2 correctly. Usually it means setting up Lossless Ethernet with
PFC (Priority Flow Control) and ECN (Explicit Congestion Notification).
Vitastor supports all adapters, even ones without ODP (On-Demand Paging)
support, like Mellanox ConnectX-3 and non-Mellanox cards. ODP is only present
in Mellanox ConnectX >= 4 adapters and allows to skip memory registration
for RDMA and thus, in theory, avoid memory copying.
Versions up to Vitastor 1.2.0 required ODP, then it was disabled by default,
but it was still supported up to 3.0.3. Now ODP support is removed because it
actually only hurts performance: an example 3-node cluster with 8 NVMe in each
node and 2*25 GBit/s ConnectX-6 RDMA network pushed 3950000 read iops without
ODP, but only 239000 iops with ODP.
This happens because Mellanox ODP implementation seems to be based on
message retransmissions when the adapter doesn't know about the buffer yet -
it likely uses standard "RNR retransmissions" (RNR = receiver not ready)
which is generally slow in RDMA/RoCE networks. Here's a presentation about
it from ISPASS-2021 conference: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
## rdma_port_num
- Type: integer
- Default: 1
RDMA device port number to use. Only for devices that have more than 1 port.
See `phys_port_cnt` in `ibv_devinfo -v` output to determine how many ports
your device has.
Not relevant for RDMA-CM (use_rdmacm).
## rdma_gid_index
- Type: integer
@@ -113,13 +152,14 @@ GID auto-selection is unsupported with libibverbs < v32.
A correct rdma_gid_index for RoCEv2 is usually 1 (IPv6) or 3 (IPv4).
Not relevant for RDMA-CM (use_rdmacm).
## rdma_mtu
- Type: integer
- Default: 4096
RDMA Path MTU to use. Must be 1024, 2048 or 4096. There is usually no
sense to change it from the default 4096.
RDMA Path MTU to use. Must be 1024, 2048 or 4096. Default is to use the
RDMA device's MTU.
## rdma_max_sge
@@ -158,28 +198,6 @@ less than `rdma_max_recv` so the receiving side doesn't run out of buffers.
Doesn't affect memory usage - additional memory isn't allocated for send
operations.
## rdma_odp
- Type: boolean
- Default: false
Use RDMA with On-Demand Paging. ODP is currently only available on Mellanox
ConnectX-4 and newer adapters. ODP allows to not register memory explicitly
for RDMA adapter to be able to use it. This, in turn, allows to skip memory
copying during sending. One would think this should improve performance, but
**in reality** RDMA performance with ODP is **drastically** worse. Example
3-node cluster with 8 NVMe in each node and 2*25 GBit/s ConnectX-6 RDMA network
without ODP pushes 3950000 read iops, but only 239000 iops with ODP...
This happens because Mellanox ODP implementation seems to be based on
message retransmissions when the adapter doesn't know about the buffer yet -
it likely uses standard "RNR retransmissions" (RNR = receiver not ready)
which is generally slow in RDMA/RoCE networks. Here's a presentation about
it from ISPASS-2021 conference: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
ODP support is retained in the code just in case a good ODP implementation
appears one day.
## peer_connect_interval
- Type: seconds
@@ -261,3 +279,63 @@ etcd_report_interval to guarantee that keepalive actually works.
etcd websocket ping interval required to keep the connection alive and
detect disconnections quickly.
## etcd_min_reload_interval
- Type: milliseconds
- Default: 1000
- Can be changed online: yes
Minimum interval for full etcd state reload. Introduced to prevent
excessive load on etcd during outages when etcd can't keep up with event
streams and cancels them.
## tcp_header_buffer_size
- Type: integer
- Default: 65536
Size of the buffer used to read data using an additional copy. Vitastor
packet headers are 128 bytes, payload is always at least 4 KB, so it is
usually beneficial to try to read multiple packets at once even though
it requires to copy the data an additional time. The rest of each packet
is received without an additional copy. You can try to play with this
parameter and see how it affects random iops and linear bandwidth if you
want.
## min_zerocopy_send_size
- Type: integer
- Default: 32768
OSDs and clients will attempt to use io_uring-based zero-copy TCP send
for buffers larger than this number of bytes. Zero-copy send with io_uring is
supported since Linux kernel version 6.1. Support is auto-detected and disabled
automatically when not available. It can also be disabled explicitly by setting
this parameter to a negative value.
⚠️ Warning! Zero-copy send performance may vary greatly from CPU to CPU and from
one kernel version to another. Generally, it tends to only make benefit with larger
messages. With smaller messages (say, 4 KB), it may actually be slower. 32 KB is
enough for almost all CPUs, but even smaller values are optimal for some of them.
For example, 4 KB is OK for EPYC Milan/Genoa and 12 KB is OK for Xeon Ice Lake
(but verify it yourself please).
Verification instructions:
1. Add `iommu=pt` into your Linux kernel command line and reboot.
2. Upgrade your kernel. For example, it's very important to use 6.11+ with recent AMD EPYCs.
3. Run some tests with the [send-zerocopy liburing example](https://github.com/axboe/liburing/blob/master/examples/send-zerocopy.c)
to find the minimal message size for which zero-copy is optimal.
Use `./send-zerocopy tcp -4 -R` at the server side and
`time ./send-zerocopy tcp -4 -b 0 -s BUFFER_SIZE -D SERVER_IP` at the client side with
`-z 0` (no zero-copy) and `-z 1` (zero-copy), and compare MB/s and used CPU time
(user+system).
## use_sync_send_recv
- Type: boolean
- Default: false
If true, synchronous send/recv syscalls are used instead of io_uring for
socket communication. Useless for OSDs because they require io_uring anyway,
but may be required for clients with old kernel versions.
+134 -58
View File
@@ -9,9 +9,11 @@
Данные параметры используются клиентами и OSD и влияют на логику сетевого
взаимодействия между клиентами, OSD, а также etcd.
- [tcp_header_buffer_size](#tcp_header_buffer_size)
- [use_sync_send_recv](#use_sync_send_recv)
- [osd_network](#osd_network)
- [osd_cluster_network](#osd_cluster_network)
- [use_rdma](#use_rdma)
- [use_rdmacm](#use_rdmacm)
- [disable_tcp](#disable_tcp)
- [rdma_device](#rdma_device)
- [rdma_port_num](#rdma_port_num)
- [rdma_gid_index](#rdma_gid_index)
@@ -20,7 +22,6 @@
- [rdma_max_msg](#rdma_max_msg)
- [rdma_max_recv](#rdma_max_recv)
- [rdma_max_send](#rdma_max_send)
- [rdma_odp](#rdma_odp)
- [peer_connect_interval](#peer_connect_interval)
- [peer_connect_timeout](#peer_connect_timeout)
- [osd_idle_timeout](#osd_idle_timeout)
@@ -30,41 +31,63 @@
- [etcd_slow_timeout](#etcd_slow_timeout)
- [etcd_keepalive_timeout](#etcd_keepalive_timeout)
- [etcd_ws_keepalive_interval](#etcd_ws_keepalive_interval)
- [etcd_min_reload_interval](#etcd_min_reload_interval)
- [tcp_header_buffer_size](#tcp_header_buffer_size)
- [min_zerocopy_send_size](#min_zerocopy_send_size)
- [use_sync_send_recv](#use_sync_send_recv)
## tcp_header_buffer_size
## osd_network
- Тип: целое число
- Значение по умолчанию: 65536
- Тип: строка или массив строк
Размер буфера для чтения данных с дополнительным копированием. Пакеты
Vitastor содержат 128-байтные заголовки, за которыми следуют данные размером
от 4 КБ и для мелких операций ввода-вывода обычно выгодно за 1 вызов читать
сразу несколько пакетов, даже не смотря на то, что это требует лишний раз
скопировать данные. Часть каждого пакета за пределами значения данного
параметра читается без дополнительного копирования. Вы можете попробовать
поменять этот параметр и посмотреть, как он влияет на производительность
случайного и линейного доступа.
Маски подсетей (IPv4 или IPv6) публичной сети или сетей OSD. Каждый OSD слушает
один и тот же порт на всех адресах поднятых (UP + RUNNING) сетевых интерфейсов,
соответствующих одной из указанных сетей. Порт выбирается автоматически, если
только [bind_port](osd.ru.md#bind_port) не задан явно. Адреса для подключений можно
также переопределить явно, задав [bind_address](osd.ru.md#bind_address). Если сети OSD
не заданы вообще, OSD слушает все адреса (0.0.0.0).
## use_sync_send_recv
## osd_cluster_network
- Тип: булево (да/нет)
- Значение по умолчанию: false
- Тип: строка или массив строк
Если установлено в истину, то вместо io_uring для передачи данных по сети
будут использоваться обычные синхронные системные вызовы send/recv. Для OSD
это бессмысленно, так как OSD в любом случае нуждается в io_uring, но, в
принципе, это может применяться для клиентов со старыми версиями ядра.
Маски подсетей (IPv4 или IPv6) отдельной кластерной сети или сетей OSD.
То есть, OSD будут всегда стараться использовать эти сети для соединений
с другими OSD, а клиенты будут стараться использовать сети из [osd_network](#osd_network).
## use_rdma
- Тип: булево (да/нет)
- Значение по умолчанию: true
Пытаться использовать RDMA для связи при наличии доступных устройств.
Отключите, если вы не хотите, чтобы Vitastor использовал RDMA.
TCP-клиенты также могут работать с RDMA-кластером, так что отключать
RDMA может быть нужно только если у клиентов есть RDMA-устройства,
но они не имеют соединения с кластером Vitastor.
Попробовать использовать RDMA через libibverbs для связи при наличии
доступных устройств. Отключите, если вы не хотите, чтобы Vitastor
использовал RDMA. TCP-клиенты также могут работать с RDMA-кластером,
так что отключать RDMA может быть нужно, только если у клиентов есть
RDMA-устройства, но они не имеют соединения с кластером Vitastor.
`use_rdma` работает с RoCEv1/RoCEv2 сетями, но не работает с iWARP и
может не работать с частью конфигураций Infiniband, требующих RDMA-CM.
Рассмотрите включение `use_rdmacm` для таких сетей.
## use_rdmacm
- Тип: булево (да/нет)
- Значение по умолчанию: false
Использовать альтернативную реализацию RDMA на основе RDMA-CM (Connection
Manager). Работает со всеми типами RDMA-сетей: Infiniband, iWARP и
RoCEv1/RoCEv2, и даже позволяет полностью отключить TCP и работать
только на RDMA. OSD используют случайные номера портов для ожидания
соединений через RDMA-CM, отличающиеся от их TCP-портов. Также при
включении `use_rdmacm` автоматически отключается опция `use_rdma`.
## disable_tcp
- Тип: булево (да/нет)
- Значение по умолчанию: true
Полностью отключить TCP и использовать только RDMA-CM для соединений с OSD.
## rdma_device
@@ -77,12 +100,6 @@ RDMA может быть нужно только если у клиентов е
не задана. Также автовыбор не поддерживается со старыми версиями библиотеки
libibverbs < v32, например в Debian 10 Buster или CentOS 7.
Vitastor поддерживает все модели адаптеров, включая те, у которых
нет поддержки ODP, то есть вы можете использовать RDMA с ConnectX-3 и
картами производства не Mellanox. Версии Vitastor до 1.2.0 включительно
требовали ODP, который есть только на Mellanox ConnectX 4 и более новых.
См. также [rdma_odp](#rdma_odp).
Запустите `ibv_devinfo -v` от имени суперпользователя, чтобы посмотреть
список доступных RDMA-устройств, их параметры и возможности.
@@ -93,16 +110,35 @@ Vitastor поддерживает все модели адаптеров, вкл
подразумевает настройку сети без потерь на основе PFC (Priority Flow
Control) и ECN (Explicit Congestion Notification).
Vitastor поддерживает все модели адаптеров, включая те, у которых нет
поддержки ODP (On-Demand Paging), например, ConnectX-3 и карты производства
не Mellanox. Функция ODP доступна только на адаптерах Mellanox ConnectX-4 и
более новых и позволяет не регистрировать память для её использования RDMA-картой,
благодаря чему в теории можно избежать лишних копирований памяти.
Версии Vitastor до 1.2.0 включительно требовали ODP, потом функция был отключена
по умолчанию, но поддерживалась вплоть до версии 3.0.3. Сейчас поддержка ODP
полностью удалена, так как на самом деле она только портит производительность:
например, на 3-узловом кластере с 8 NVMe в каждом узле и сетью 2*25 Гбит/с на
чтение с RDMA без ODP удаётся снять 3950000 iops, а с ODP - всего 239000 iops.
Это происходит из-за того, что реализация ODP у Mellanox неоптимальная и
основана на повторной передаче сообщений, когда карте не известен буфер -
вероятно, на стандартных "RNR retransmission" (RNR = receiver not ready).
А данные повторные передачи в RDMA/RoCE - всегда очень медленная штука.
Презентация на эту тему с конференции ISPASS-2021: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
## rdma_port_num
- Тип: целое число
- Значение по умолчанию: 1
Номер порта RDMA-устройства, который следует использовать. Имеет смысл
только для устройств, у которых более 1 порта. Чтобы узнать, сколько портов
у вашего адаптера, посмотрите `phys_port_cnt` в выводе команды
`ibv_devinfo -v`.
Опция неприменима к RDMA-CM (use_rdmacm).
## rdma_gid_index
- Тип: целое число
@@ -119,13 +155,14 @@ libibverbs < v32.
Правильный rdma_gid_index для RoCEv2, как правило, 1 (IPv6) или 3 (IPv4).
Опция неприменима к RDMA-CM (use_rdmacm).
## rdma_mtu
- Тип: целое число
- Значение по умолчанию: 4096
Максимальная единица передачи (Path MTU) для RDMA. Должно быть равно 1024,
2048 или 4096. Обычно нет смысла менять значение по умолчанию, равное 4096.
2048 или 4096. По умолчанию используется значение MTU RDMA-устройства.
## rdma_max_sge
@@ -166,29 +203,6 @@ OSD в любом случае согласовывают реальное зн
Не влияет на потребление памяти - дополнительная память на операции отправки
не выделяется.
## rdma_odp
- Тип: булево (да/нет)
- Значение по умолчанию: false
Использовать RDMA с On-Demand Paging. ODP - функция, доступная пока что
исключительно на адаптерах Mellanox ConnectX-4 и более новых. ODP позволяет
не регистрировать память для её использования RDMA-картой. Благодаря этому
можно не копировать данные при отправке их в сеть и, казалось бы, это должно
улучшать производительность - но **по факту** получается так, что
производительность только ухудшается, причём сильно. Пример - на 3-узловом
кластере с 8 NVMe в каждом узле и сетью 2*25 Гбит/с на чтение с RDMA без ODP
удаётся снять 3950000 iops, а с ODP - всего 239000 iops...
Это происходит из-за того, что реализация ODP у Mellanox неоптимальная и
основана на повторной передаче сообщений, когда карте не известен буфер -
вероятно, на стандартных "RNR retransmission" (RNR = receiver not ready).
А данные повторные передачи в RDMA/RoCE - всегда очень медленная штука.
Презентация на эту тему с конференции ISPASS-2021: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
Возможность использования ODP сохранена в коде на случай, если вдруг в один
прекрасный день появится хорошая реализация ODP.
## peer_connect_interval
- Тип: секунды
@@ -271,3 +285,65 @@ etcd_report_interval, чтобы keepalive гарантированно рабо
- Можно менять на лету: да
Интервал проверки живости вебсокет-подключений к etcd.
## etcd_min_reload_interval
- Тип: миллисекунды
- Значение по умолчанию: 1000
- Можно менять на лету: да
Минимальный интервал полной перезагрузки состояния из etcd. Добавлено для
предотвращения избыточной нагрузки на etcd во время отказов, когда etcd не
успевает рассылать потоки событий и отменяет их.
## tcp_header_buffer_size
- Тип: целое число
- Значение по умолчанию: 65536
Размер буфера для чтения данных с дополнительным копированием. Пакеты
Vitastor содержат 128-байтные заголовки, за которыми следуют данные размером
от 4 КБ и для мелких операций ввода-вывода обычно выгодно за 1 вызов читать
сразу несколько пакетов, даже не смотря на то, что это требует лишний раз
скопировать данные. Часть каждого пакета за пределами значения данного
параметра читается без дополнительного копирования. Вы можете попробовать
поменять этот параметр и посмотреть, как он влияет на производительность
случайного и линейного доступа.
## min_zerocopy_send_size
- Тип: целое число
- Значение по умолчанию: 32768
OSD и клиенты будут пробовать использовать TCP-отправку без копирования (zero-copy) на
основе io_uring для буферов, больших, чем это число байт. Отправка без копирования
поддерживается в io_uring, начиная с версии ядра Linux 6.1. Наличие поддержки
проверяется автоматически и zero-copy отключается, когда поддержки нет. Также
её можно отключить явно, установив данный параметр в отрицательное значение.
⚠️ Внимание! Производительность данной функции может сильно отличаться на разных
процессорах и на разных версиях ядра Linux. В целом, zero-copy обычно быстрее с
большими сообщениями, а с мелкими (например, 4 КБ) zero-copy может быть даже
медленнее. 32 КБ достаточно почти для всех процессоров, но для каких-то можно
использовать даже меньшие значения. Например, для EPYC Milan/Genoa подходит 4 КБ,
а для Xeon Ice Lake - 12 КБ (но, пожалуйста, перепроверьте это сами).
Инструкция по проверке:
1. Добавьте `iommu=pt` в командную строку загрузки вашего ядра Linux и перезагрузитесь.
2. Обновите ядро. Например, для AMD EPYC очень важно использовать версию 6.11+.
3. Позапускайте тесты с помощью [send-zerocopy из примеров liburing](https://github.com/axboe/liburing/blob/master/examples/send-zerocopy.c),
чтобы найти минимальный размер сообщения, для которого zero-copy отправка оптимальна.
Запускайте `./send-zerocopy tcp -4 -R` на стороне сервера и
`time ./send-zerocopy tcp -4 -b 0 -s РАЗМЕР_БУФЕРА -D АДРЕС_СЕРВЕРА` на стороне клиента
с опцией `-z 0` (обычная отправка) и `-z 1` (отправка без копирования), и сравнивайте
скорость в МБ/с и занятое процессорное время (user+system).
## use_sync_send_recv
- Тип: булево (да/нет)
- Значение по умолчанию: false
Если установлено в истину, то вместо io_uring для передачи данных по сети
будут использоваться обычные синхронные системные вызовы send/recv. Для OSD
это бессмысленно, так как OSD в любом случае нуждается в io_uring, но, в
принципе, это может применяться для клиентов со старыми версиями ядра.
+166 -42
View File
@@ -7,16 +7,15 @@
# Runtime OSD Parameters
These parameters only apply to OSDs, are not fixed at the moment of OSD drive
initialization and can be changed - either with an OSD restart or, for some of
them, even without restarting by updating configuration in etcd.
initialization and can be changed - in /etc/vitastor/vitastor.conf or [vitastor-disk update-sb](../usage/disk.en.md#update-sb)
with an OSD restart or, for some of them, even without restarting by updating configuration in etcd.
- [bind_address](#bind_address)
- [bind_port](#bind_port)
- [osd_iothread_count](#osd_iothread_count)
- [etcd_report_interval](#etcd_report_interval)
- [etcd_stats_interval](#etcd_stats_interval)
- [run_primary](#run_primary)
- [osd_network](#osd_network)
- [bind_address](#bind_address)
- [bind_port](#bind_port)
- [autosync_interval](#autosync_interval)
- [autosync_writes](#autosync_writes)
- [recovery_queue_depth](#recovery_queue_depth)
@@ -39,6 +38,7 @@ them, even without restarting by updating configuration in etcd.
- [journal_io](#journal_io)
- [journal_sector_buffer_count](#journal_sector_buffer_count)
- [journal_no_same_sector_overwrites](#journal_no_same_sector_overwrites)
- [skip_corrupted_meta_entries](#skip_corrupted_meta_entries)
- [throttle_small_writes](#throttle_small_writes)
- [throttle_target_iops](#throttle_target_iops)
- [throttle_target_mbs](#throttle_target_mbs)
@@ -61,6 +61,32 @@ them, even without restarting by updating configuration in etcd.
- [recovery_tune_agg_interval](#recovery_tune_agg_interval)
- [recovery_tune_sleep_min_us](#recovery_tune_sleep_min_us)
- [recovery_tune_sleep_cutoff_us](#recovery_tune_sleep_cutoff_us)
- [discard_on_start](#discard_on_start)
- [min_discard_size](#min_discard_size)
- [allow_net_split](#allow_net_split)
- [enable_pg_locks](#enable_pg_locks)
- [pg_lock_retry_interval_ms](#pg_lock_retry_interval_ms)
- [atomic_write_size](#atomic_write_size)
- [use_atomic_flag](#use_atomic_flag)
- [pg_reshard_chunk_size](#pg_reshard_chunk_size)
- [pg_reshard_chunk_pause_ms](#pg_reshard_chunk_pause_ms)
## bind_address
- Type: string or array of strings
Instead of the network masks ([osd_network](network.en.md#osd_network) and
[osd_cluster_network](network.en.md#osd_cluster_network)), you can also set
OSD listen addresses explicitly using this parameter. May be useful if you
want to start OSDs on interfaces that are not UP + RUNNING.
## bind_port
- Type: integer
By default, OSDs pick random ports to use for incoming connections
automatically. With this option you can set a specific port for a specific
OSD by hand.
## osd_iothread_count
@@ -104,34 +130,6 @@ debugging purposes. It's possible to implement additional feature for the
monitor which may allow to separate primary and secondary OSDs, but it's
unclear why anyone could need it, so it's not implemented.
## osd_network
- Type: string or array of strings
Network mask of the network (IPv4 or IPv6) to use for OSDs. Note that
although it's possible to specify multiple networks here, this does not
mean that OSDs will create multiple listening sockets - they'll only
pick the first matching address of an UP + RUNNING interface. Separate
networks for cluster and client connections are also not implemented, but
they are mostly useless anyway, so it's not a big deal.
## bind_address
- Type: string
- Default: 0.0.0.0
Instead of the network mask, you can also set OSD listen address explicitly
using this parameter. May be useful if you want to start OSDs on interfaces
that are not UP + RUNNING.
## bind_port
- Type: integer
By default, OSDs pick random ports to use for incoming connections
automatically. With this option you can set a specific port for a specific
OSD by hand.
## autosync_interval
- Type: seconds
@@ -282,13 +280,19 @@ Maximum number of journal flushers (see above min_flusher_count).
- Type: boolean
- Default: true
This parameter makes Vitastor always keep metadata area of the block device
in memory. It's required for good performance because it allows to avoid
additional read-modify-write cycles during metadata modifications. Metadata
area size is currently roughly 224 MB per 1 TB of data. You can turn it off
to reduce memory usage by this value, but it will hurt performance. This
restriction is likely to be removed in the future along with the upgrade
of the metadata storage scheme.
Only for the old store ([meta_format](layout-osd.en.md#meta_format) 2).
This parameter makes Vitastor keep a copy of metadata area in memory as it is
on disk, in addition to the metadata database. When the option is enabled, every
metadata entry is effectively stored in RAM twice. It's required for good performance
because it allows to avoid additional read-modify-write cycles during metadata
modifications. Metadata area size with the old store is roughly 224 MB per 1 TB
of data. You can turn the option off to reduce memory usage by this value, but
it will reduce performance.
For the new store ([meta_format](layout-osd.en.md#meta_format) 3), the option
may be changed in the future to support operation without loading full metadata
database in memory.
## inmemory_journal
@@ -316,7 +320,7 @@ for hot data and slower disks - HDDs and maybe SATA SSDs - but will slightly
decrease write performance for fast disks because page cache is an overhead
itself.
Choose "directsync" to use [immediate_commit](layout-cluster.ru.md#immediate_commit)
Choose "directsync" to use [immediate_commit](layout-cluster.en.md#immediate_commit)
(which requires disable_data_fsync) with drives having write-back cache
which can't be turned off, for example, Intel Optane. Also note that *some*
desktop SSDs (for example, HP EX950) may ignore O_SYNC thus making
@@ -367,6 +371,8 @@ blocks. The only situation when you should increase it to a larger value
is when you enable journal_no_same_sector_overwrites. In this case set
it to, for example, 1024.
Not applicable to the new store ([meta_format](layout-osd.en.md#meta_format) 3).
## journal_no_same_sector_overwrites
- Type: boolean
@@ -380,6 +386,17 @@ journal after writing it instead of possibly overwriting it the second time.
Most (99%) other SSDs don't need this option.
Not applicable to the new store ([meta_format](layout-osd.en.md#meta_format) 3).
## skip_corrupted_meta_entries
- Type: boolean
- Default: false
Only for the new store ([meta_format](layout-osd.en.md#meta_format) 3).
Allow OSD to start when some metadata entries or blocks are corrupted by
skipping them. Should be only used as an emergency measure.
## throttle_small_writes
- Type: boolean
@@ -498,7 +515,7 @@ Can be used to slow down scrubbing if it affects user load too much.
## scrub_list_limit
- Type: integer
- Default: 1000
- Default: 262144
- Can be changed online: yes
Number of objects to list in one listing operation during scrub.
@@ -629,3 +646,110 @@ are changed to 0.
Maximum possible value for auto-tuned recovery_sleep_us. Higher values
are treated as outliers and ignored in aggregation.
## discard_on_start
- Type: boolean
Discard (SSD TRIM) unused data device blocks on every OSD startup.
## min_discard_size
- Type: integer
- Default: 1048576
Minimum consecutive block size to TRIM it.
## allow_net_split
- Type: boolean
- Default: false
Allow "safe" cases of network splits/partitions - allow to start PGs without
connections to some OSDs currently registered as alive in etcd, if the number
of actually connected PG OSDs is at least pg_minsize. That is, allow some OSDs to lose
connectivity with some other OSDs as long as it doesn't break pg_minsize guarantees.
The downside is that it increases the probability of writing data into just pg_minsize
OSDs during failover which can lead to PGs becoming incomplete after additional outages.
The old behaviour in versions up to 2.0.0 was equal to enabled allow_net_split.
## enable_pg_locks
- Type: boolean
Vitastor 2.2.0 introduces a new layer of split-brain prevention mechanism in
addition to etcd: PG locks. They prevent split-brain even in abnormal theoretical cases
when etcd is extremely laggy. As a new feature, by default, PG locks are only enabled
for pools where they're required - pools with [localized reads](pool.en.md#local_reads).
Use this parameter to enable or disable this function for all pools.
## pg_lock_retry_interval_ms
- Type: milliseconds
- Default: 100
Retry interval for failed PG lock attempts.
## atomic_write_size
- Type: integer
- Default: 4096
Maximum data device atomic write size allowed for OSD to use.
Atomic writes allow to reduce the Write Amplification factor with the new store
([meta_format](layout-osd.en.md#meta_format)=3) to almost 1 (i.e. almost no extra writes)
with replicated pools and reach the best possible write performance.
Default value is auto-detected during OSD initialization from
`/sys/block/xx/queue/atomic_write_max_bytes` or assumed to be 4096 bytes
because all known disks support 4 KB atomic writes. Auto-detection is only used for
NVMe disks because SAS disks require the explicit WRITE ATOMIC command which requires
RWF_ATOMIC (see below [#use_atomic_flag]) but that flag works incorrectly in current
Linux versions.
You can also check if your NVMe drives support atomic writes by running
the command `nvme id-ctrl /dev/nvme0n1 | grep awupf`. If the reported value,
plus 1, multiplied by the currently selected block size of the NVMe,
is more than 4 KB, then the new store can utilize it for better performance.
The only drives known to support it currently are [Micron and Kioxia](../intro/quickstart.en.md).
Atomic writes allow to skip double data writes in replicated pools, thus
reducing Write Amplification and improving write performance up to 2 times.
## use_atomic_flag
- Type: boolean
This option controls whether Vitastor OSDs use RWF_ATOMIC write flag with atomic writes.
This flag is supported since Linux 6.11 and adds some safety to atomic writes - the kernel
guarantees to not fragment write requests with it and also to check them against the actual
device atomic write capabilities.
However, the option is disabled by default because the flag is currently UNUSABLE - Linux
incorrectly requires writes with that flag to be of power-of-2 length and length-aligned.
I.e., for example, 12 KB writes and not-8-KB aligned 8 KB writes are forbidden by the kernel,
even though the NVMe specification allows them.
For NVMe disks with `scheduler=none` writes aren't fragmented anyway so it's not a big deal.
However, you can rebuild your kernel with [this patch](../../patches/linux-fix-atomic-write-checks.diff)
and turn this option on. It will make your atomic writes a bit safer.
## pg_reshard_chunk_size
- Type: integer
- Default: 100000
Pool PG count change is a CPU-intensive operation because OSDs store the full object database
in memory and have to move all entries between old and new PGs. Thus it's performed in chunks,
with pauses between chunks to prevent blocking OSD's event loop and other clients' operations.
This option sets the maximum number of object is a chunk. Moving 100k objects usually takes
50-100ms. Chunk size equal to 0 means unlimited.
## pg_reshard_chunk_pause_ms
- Type: milliseconds
- Default: 100
This option sets the interval between handling two PG count change chunks.
+174 -41
View File
@@ -8,16 +8,15 @@
Данные параметры используются только OSD, но, в отличие от дисковых параметров,
не фиксируются в момент инициализации дисков OSD и могут быть изменены в любой
момент с помощью перезапуска OSD, а некоторые и без перезапуска, с помощью
изменения конфигурации в etcd.
момент с перезапуском OSD в /etc/vitastor/vitastor.conf или [vitastor-disk update-sb](../usage/disk.ru.md#update-sb),
а некоторые и без перезапуска, с помощью изменения конфигурации в etcd.
- [bind_address](#bind_address)
- [bind_port](#bind_port)
- [osd_iothread_count](#osd_iothread_count)
- [etcd_report_interval](#etcd_report_interval)
- [etcd_stats_interval](#etcd_stats_interval)
- [run_primary](#run_primary)
- [osd_network](#osd_network)
- [bind_address](#bind_address)
- [bind_port](#bind_port)
- [autosync_interval](#autosync_interval)
- [autosync_writes](#autosync_writes)
- [recovery_queue_depth](#recovery_queue_depth)
@@ -40,6 +39,7 @@
- [journal_io](#journal_io)
- [journal_sector_buffer_count](#journal_sector_buffer_count)
- [journal_no_same_sector_overwrites](#journal_no_same_sector_overwrites)
- [skip_corrupted_meta_entries](#skip_corrupted_meta_entries)
- [throttle_small_writes](#throttle_small_writes)
- [throttle_target_iops](#throttle_target_iops)
- [throttle_target_mbs](#throttle_target_mbs)
@@ -62,6 +62,32 @@
- [recovery_tune_agg_interval](#recovery_tune_agg_interval)
- [recovery_tune_sleep_min_us](#recovery_tune_sleep_min_us)
- [recovery_tune_sleep_cutoff_us](#recovery_tune_sleep_cutoff_us)
- [discard_on_start](#discard_on_start)
- [min_discard_size](#min_discard_size)
- [allow_net_split](#allow_net_split)
- [enable_pg_locks](#enable_pg_locks)
- [pg_lock_retry_interval_ms](#pg_lock_retry_interval_ms)
- [atomic_write_size](#atomic_write_size)
- [use_atomic_flag](#use_atomic_flag)
- [pg_reshard_chunk_size](#pg_reshard_chunk_size)
- [pg_reshard_chunk_pause_ms](#pg_reshard_chunk_pause_ms)
## bind_address
- Тип: строка или массив строк
Вместо использования масок подсети ([osd_network](network.ru.md#osd_network) и
[osd_cluster_network](network.ru.md#osd_cluster_network)), вы также можете явно
задать адрес(а), на которых будут ожидать соединений OSD, с помощью данного
параметра. Это может быть полезно, например, чтобы запускать OSD на неподнятых
интерфейсах (не UP + RUNNING).
## bind_port
- Тип: целое число
По умолчанию OSD сами выбирают случайные порты для входящих подключений.
С помощью данной опции вы можете задать порт для отдельного OSD вручную.
## osd_iothread_count
@@ -107,34 +133,6 @@ max_etcd_attempts * etcd_quick_timeout.
первичные OSD от вторичных, но пока не понятно, зачем это может кому-то
понадобиться, поэтому это не реализовано.
## osd_network
- Тип: строка или массив строк
Маска подсети (IPv4 или IPv6) для использования для соединений с OSD.
Имейте в виду, что хотя сейчас и можно передать в этот параметр несколько
подсетей, это не означает, что OSD будут создавать несколько слушающих
сокетов - они лишь будут выбирать адрес первого поднятого (состояние UP +
RUNNING), подходящий под заданную маску. Также не реализовано разделение
кластерной и публичной сетей OSD. Правда, от него обычно всё равно довольно
мало толку, так что особенной проблемы в этом нет.
## bind_address
- Тип: строка
- Значение по умолчанию: 0.0.0.0
Этим параметром можно явным образом задать адрес, на котором будет ожидать
соединений OSD (вместо использования маски подсети). Может быть полезно,
например, чтобы запускать OSD на неподнятых интерфейсах (не UP + RUNNING).
## bind_port
- Тип: целое число
По умолчанию OSD сами выбирают случайные порты для входящих подключений.
С помощью данной опции вы можете задать порт для отдельного OSD вручную.
## autosync_interval
- Тип: секунды
@@ -290,13 +288,19 @@ Flusher - это микро-поток (корутина), которая коп
- Тип: булево (да/нет)
- Значение по умолчанию: true
Данный параметр заставляет Vitastor всегда держать область метаданных диска
в памяти. Это нужно, чтобы избегать дополнительных операций чтения с диска
при записи. Размер области метаданных на данный момент составляет примерно
224 МБ на 1 ТБ данных. При включении потребление памяти снизится примерно
на эту величину, но при этом также снизится и производительность. В будущем,
после обновления схемы хранения метаданных, это ограничение, скорее всего,
будет ликвидировано.
Только для старого хранилища ([meta_format](layout-osd.en.md#meta_format) 2).
Данный параметр заставляет Vitastor всегда держать копию области метаданных
в памяти в том же виде, как она лежит на диске, в дополнение к БД метаданных.
То есть, с включённой опцией каждая запись метаданных хранится в памяти дважды.
Это нужно, чтобы избегать дополнительных операций чтения с диска при записи.
Размер области метаданных в старом хранилище составляет примерно 224 МБ на
1 ТБ данных. Вы можете отключить опцию, чтобы снизить потребление памяти
примерно на эту величину, но при этом также снизится и производительность.
Для нового хранилища ([meta_format](layout-osd.en.md#meta_format) 3) опция,
возможно, будет переработана в будущем для поддержки работы без полной
загрузки метаданных в памяти.
## inmemory_journal
@@ -379,6 +383,8 @@ fsync небезопасным даже с режимом "directsync".
нужно менять - это если вы включаете journal_no_same_sector_overwrites. В
этом случае установите данный параметр, например, в 1024.
Неприменимо к новому хранилищу ([meta_format](layout-osd.en.md#meta_format) 3).
## journal_no_same_sector_overwrites
- Тип: булево (да/нет)
@@ -394,6 +400,18 @@ fsync небезопасным даже с режимом "directsync".
Почти все другие SSD (99% моделей) не требуют данной опции.
Неприменимо к новому хранилищу ([meta_format](layout-osd.en.md#meta_format) 3).
## skip_corrupted_meta_entries
- Тип: булево (да/нет)
- Значение по умолчанию: false
Только для нового хранилища ([meta_format](layout-osd.en.md#meta_format) 3).
Разрешить OSD запускаться, даже если часть блоков или записей метаданных
повреждена, пропуская их. Опция предназначена для использования только в
целях аварийного восстановления.
## throttle_small_writes
- Тип: булево (да/нет)
@@ -521,7 +539,7 @@ fsync небезопасным даже с режимом "directsync".
## scrub_list_limit
- Тип: целое число
- Значение по умолчанию: 1000
- Значение по умолчанию: 262144
- Можно менять на лету: да
Размер загружаемых за одну операцию списков объектов в процессе фоновой
@@ -660,3 +678,118 @@ EC (кодов коррекции ошибок) с более, чем 1 диск
Максимальное возможное значение авто-подстроенного recovery_sleep_us.
Большие значения считаются случайными выбросами и игнорируются в
усреднении.
## discard_on_start
- Тип: булево (да/нет)
Освобождать (SSD TRIM) неиспользуемые блоки диска данных при каждом запуске OSD.
## min_discard_size
- Тип: целое число
- Значение по умолчанию: 1048576
Минимальный размер последовательного блока данных, чтобы освобождать его через TRIM.
## allow_net_split
- Тип: булево (да/нет)
- Значение по умолчанию: false
Разрешить "безопасные" случаи разделений сети - разрешить активировать PG без
соединений к некоторым OSD, помеченным активными в etcd, если общее число активных
OSD в PG составляет как минимум pg_minsize. То есть, разрешать некоторым OSD терять
соединения с некоторыми другими OSD, если это не нарушает гарантий pg_minsize.
Минус такого разрешения в том, что оно повышает вероятность записи данных ровно в
pg_minsize OSD во время переключений, что может потом привести к тому, что PG станут
неполными (incomplete), если упадут ещё какие-то OSD.
Старое поведение в версиях до 2.0.0 было идентично включённому allow_net_split.
## enable_pg_locks
- Тип: булево (да/нет)
В Vitastor 2.2.0 появился новый слой защиты от сплитбрейна в дополнение к etcd -
блокировки PG. Они гарантируют порядок даже в теоретических ненормальных случаях,
когда etcd очень сильно тормозит. Так как функция новая, по умолчанию она включается
только для пулов, в которых она необходима - а именно, в пулах с включёнными
[локальными чтениями](pool.ru.md#local_reads). Ну а с помощью данного параметра
можно включить блокировки PG для всех пулов.
## pg_lock_retry_interval_ms
- Тип: миллисекунды
- Значение по умолчанию: 100
Интервал повтора неудачных попыток блокировки PG.
## atomic_write_size
- Тип: целое число
- Значение по умолчанию: 4096
Максимальный размер атомарной записи на диск данных, который OSD разрешено использовать.
Поддержка атомарной записи позволяет снизить мультипликатор записи (Write Amplification)
на диск с новым хранилищем ([meta_format](layout-osd.ru.md#meta_format)=3)
практически до 1 (то есть, почти до нулевого объёма лишней записи) в реплицированных
пулах и достигнуть наилучшей возможной производительности записи.
Значение по умолчанию авто-определяется во время инициализации OSD из
`/sys/block/xx/queue/atomic_write_max_bytes` либо принимается равным 4096,
так как все известные диски поддерживают атомарную запись 4 КБ блоков.
Автоопределение применяется только для NVMe-дисков, так как SAS диски требуют
использования отдельной команды WRITE ATOMIC, а для неё нужен флаг RWF_ATOMIC
(см. ниже [#use_atomic_flag]), а он в текущих версиях Linux работает некорректно.
Вы также можете проверить, поддерживают ли ваши NVMe-диски атомарную запись,
с помощью команды `nvme id-ctrl /dev/nvme0n1 | grep awupf`. Если значение awupf
плюс 1, умноженное на текущий выбранный размер блока NVMe-диска, больше 4 КБ,
то новое хранилище может использовать атомарные записи для достижения лучшей
производительности. Единственные известные диски, которые поддерживают это сейчас -
[Micron и Kioxia](../intro/quickstart.ru.md).
Атомарная запись позволяет не использовать двойную запись данных (в журнал и на
устройство данных) в реплицированных пулах и таким образом снижает амплификацию
записи (объём служебной записи на диск) и улучшает производительность записи
вплоть до 2-х кратного прироста.
## use_atomic_flag
- Тип: булево (да/нет)
Данная опция контролирует использование Vitastor OSD флага RWF_ATOMIC при атомарной записи
блоков. Этот флаг поддерживается, начиная с версии ядра Linux 6.11 и добавляет немного корректности
атомарным записям - ядро гарантирует отсутствие фрагментации запросов записи с этим флагом и
проверяет их на соответствие реальным возможностям устройства.
Однако, данная опция по умолчанию отключена, так как флаг в текущих версиях Linux работает
абсолютно НЕКОРРЕКТНО - при нём Linux требует, чтобы запросы записи имели длину, равную
степени двойки и были выровнены на эту длину. То есть, например, 12 КБ запросы записи, а также
8 КБ запросы записи по не-кратному 8 КБ смещению запрещаются ядром, хотя спецификация NVMe их
разрешает.
Для NVMe-дисков с `scheduler=none` запросы записи и так не фрагментируются, так что это не так
уж и важно, однако вы можете пересобрать своё ядро с [этим патчем](../../patches/linux-fix-atomic-write-checks.diff)
и включить данную опцию. Это сделает вашу атомарную запись капельку безопаснее.
## pg_reshard_chunk_size
- Тип: целое число
- Значение по умолчанию: 100000
Изменение числа PG в пуле заметно загружает процессор, так как OSD хранят полную базу данных
объектов в памяти и им приходится перемещать все записи объектов между старыми и новыми PG.
Поэтому изменение применяется порциями, с паузами между порциями, чтобы не блокировать обработку
событий OSD и операции остальных клиентов. Данная опция задаёт максимальное число объектов
в порции. Перемещение 100 тысяч объектов (значение по умолчанию) обычно занимает порядка
50-100 миллисекунд. Значение опции 0 отключает лимит размера порции.
## pg_reshard_chunk_pause_ms
- Тип: миллисекунды
- Значение по умолчанию: 100
Данная опция задаёт интервал между обработкой двух порций изменения числа PG пулов.
+54 -12
View File
@@ -34,6 +34,7 @@ Parameters:
- [failure_domain](#failure_domain)
- [level_placement](#level_placement)
- [raw_placement](#raw_placement)
- [local_reads](#local_reads)
- [max_osd_combinations](#max_osd_combinations)
- [block_size](#block_size)
- [bitmap_granularity](#bitmap_granularity)
@@ -43,7 +44,7 @@ Parameters:
- [osd_tags](#osd_tags)
- [primary_affinity_tags](#primary_affinity_tags)
- [scrub_interval](#scrub_interval)
- [used_for_fs](#used_for_fs)
- [used_for_app](#used_for_app)
Examples:
@@ -133,8 +134,8 @@ Pool name.
## scheme
- Type: string
- Required
- One of: "replicated", "xor", "ec" or "jerasure"
- Required
Redundancy scheme used for data in this pool. "jerasure" is an alias for "ec",
both use Reed-Solomon-Vandermonde codes based on ISA-L or jerasure libraries.
@@ -189,6 +190,9 @@ So, pg_minsize regulates the number of failures that a pool can tolerate
without temporary downtime for [osd_out_time](monitor.en.md#osd_out_time),
but at a cost of slightly reduced storage reliability.
See also [allow_net_split](osd.en.md#allow_net_split) and
[PG state descriptions](../usage/admin.en.md#pg-states).
FIXME: pg_minsize behaviour may be changed in the future to only make PGs
read-only instead of deactivating them.
@@ -286,6 +290,30 @@ Examples:
- EC 4+2 in 3 DC: `any, dc=1 host!=1, dc!=1, dc=3 host!=3, dc!=(1,3), dc=5 host!=5`
- 1 replica in fixed DC + 2 in random DCs: `dc?=meow, dc!=1, dc!=(1,2)`
## local_reads
- Type: string
- One of: "primary", "nearest" or "random"
- Default: primary
By default, Vitastor serves all read and write requests from the primary OSD of each PG.
But it can also serve read requests for replicated pools from secondary OSDs in clean PGs
(active or active+left_on_dead) which may be useful if you have OSDs with different network
latency to the client - for example, if you have a cross-datacenter setup.
If you set this parameter to "nearest", clients will try to read from the nearest OSD
in the [Placement Tree](#placement-tree), i.e. from an OSD from the same host or datacenter.
Distance to different OSDs will be calculated based on client hostname, determined
automatically or set manually in the [hostname](client.en.md#hostname) parameter.
If you set this parameter to "random", clients will try to distribute read requests over
all available secondary OSDs. This mode is mainly useful for tests, but, probably, not
really required in production setups.
[PG locks](osd.en.md#enable_pg_locks) are required for local reads to function. However,
PG locks are enabled automatically by default for pools with enabled local reads, so you
don't have to enable them explicitly.
## max_osd_combinations
- Type: integer
@@ -321,7 +349,8 @@ Read more about this parameter in [Cluster-Wide Disk Layout Parameters](layout-c
## immediate_commit
- Type: string, one of "all", "small" and "none"
- Type: string
- One of: "all", "small" or "none"
- Default: none
Immediate commit setting for this pool. The value from /vitastor/config/global
@@ -377,24 +406,37 @@ of the OSDs containing a data chunk for a PG.
Automatic scrubbing interval for this pool. Overrides
[global scrub_interval setting](osd.en.md#scrub_interval).
## used_for_fs
## used_for_app
- Type: string
If non-empty, the pool is marked as used for VitastorFS with metadata stored
in block image (regular Vitastor volume) named as the value of this pool parameter.
If non-empty, the pool is marked as used for a separate application, for example,
VitastorFS or S3, which allocates Vitastor volume IDs by itself and does not use
image/inode metadata in etcd.
When a pool is marked as used for VitastorFS, regular block volume creation in it
When a pool is marked as used for such app, regular block volume creation in it
is disabled (vitastor-cli refuses to create images without --force) to protect
the user from block volume and FS file ID collisions and data loss.
the user from block volume and FS/S3 volume ID collisions and data loss.
[vitastor-nfs](../usage/nfs.ru.md), in its turn, refuses to use pools not marked
Also such pools do not calculate per-inode space usage statistics in etcd because
using it for an external application implies that it may contain a very large
number of volumes and their statistics may take too much space in etcd.
Setting used_for_app to `fs:<name>` tells Vitastor that the pool is used for VitastorFS
with VitastorKV metadata base stored in a block image (regular Vitastor volume) named
`<name>`.
[vitastor-nfs](../usage/nfs.en.md), in its turn, refuses to use pools not marked
for the corresponding FS when starting. This also implies that you can use one
pool only for one VitastorFS.
The second thing that is disabled for VitastorFS pools is reporting per-inode space
usage statistics in etcd because a FS pool may store a very large number of files
and statistics for them all would take a lot of space in etcd.
If you plan to use the pool for S3, set its used_for_app to `s3:<name>`. `<name>` may
be basically anything you want (for example, `s3:standard`) - it's not validated
by Vitastor S3 components in any way.
All other values except prefixed with `fs:` or `s3:` may be used freely and don't
mean anything special for Vitastor core components. For now, you can use them as
you wish.
# Examples
+60 -18
View File
@@ -33,6 +33,7 @@
- [failure_domain](#failure_domain)
- [level_placement](#level_placement)
- [raw_placement](#raw_placement)
- [local_reads](#local_reads)
- [max_osd_combinations](#max_osd_combinations)
- [block_size](#block_size)
- [bitmap_granularity](#bitmap_granularity)
@@ -42,7 +43,7 @@
- [osd_tags](#osd_tags)
- [primary_affinity_tags](#primary_affinity_tags)
- [scrub_interval](#scrub_interval)
- [used_for_fs](#used_for_fs)
- [used_for_app](#used_for_app)
Примеры:
@@ -133,8 +134,8 @@ OSD игнорируется и OSD не удаляется из распред
## scheme
- Тип: строка
- Обязательный
- Возможные значения: "replicated", "xor", "ec" или "jerasure"
- Обязательный
Схема избыточности, используемая в данном пуле. "jerasure" - синоним для "ec",
в обеих схемах используются коды Рида-Соломона-Вандермонда, реализованные на
@@ -256,7 +257,7 @@ PG в Vitastor эферемерны, то есть вы можете менят
## raw_placement
- Type: string
- Тип: строка
Низкоуровневые правила генерации PG в форме DSL (доменно-специфичного языка).
Используйте, только если действительно знаете, зачем вам это надо :)
@@ -287,6 +288,30 @@ meow недоступен".
- EC 4+2 в 3 датацентрах: `any, dc=1 host!=1, dc!=1, dc=3 host!=3, dc!=(1,3), dc=5 host!=5`
- 1 копия в фиксированном ДЦ + 2 в других ДЦ: `dc?=meow, dc!=1, dc!=(1,2)`
## local_reads
- Тип: строка
- Возможные значения: "primary", "nearest" или "random"
- По умолчанию: primary
По умолчанию Vitastor обслуживает все запросы чтения и записи с первичного OSD каждой PG.
Однако, в чистых PG (active или active+left_on_dead) реплицированных пулов также есть
возможность обслуживать запросы чтения с вторичных OSD, что может быть полезно, если
у вас сильно отличается время сетевого обращения от клиента к разным OSD - например,
если у вас несколько дата-центров.
Если данный параметр установлен в значение "nearest", клиенты будут стараться читать с
ближайших по [Дереву размещения](#дерево-размещения) OSD, то есть, с OSD с того же хоста
или датацентра. Расстояние до разных OSD будет рассчитываться с помощью имени хоста клиента,
определяемого автоматически или заданного вручную параметром [hostname](client.ru.md#hostname).
Если данный параметр установлен в значение "random", клиенты будут стараться распределять
запросы чтения по всем доступным вторичным OSD. Этот режим в основном полезен для тестов,
но, скорее всего, редко нужен в реальных инсталляциях.
Для работы локальных чтений требуются [блокировки PG](osd.ru.md#enable_pg_locks). Включать
их явно не нужно - они включаются автоматически для пулов с включёнными локальными чтениями.
## max_osd_combinations
- Тип: целое число
@@ -324,7 +349,8 @@ meow недоступен".
## immediate_commit
- Тип: строка "all", "small" или "none"
- Тип: строка
- Возможные значения: "all", "small" или "none"
- По умолчанию: none
Настройка мгновенного коммита для данного пула. Если не задана, используется
@@ -383,26 +409,42 @@ OSD с "all".
Интервал скраба, то есть, автоматической фоновой проверки данных для данного пула.
Переопределяет [глобальную настройку scrub_interval](osd.ru.md#scrub_interval).
## used_for_fs
## used_for_app
- Type: string
- Тип: строка
Если непусто, пул помечается как используемый для файловой системы VitastorFS с
метаданными, хранимыми в блочном образе Vitastor с именем, равным значению
этого параметра.
Если непусто, пул помечается как используемый для отдельного приложения, например,
для VitastorFS или S3, которое распределяет ID образов в пуле само и не использует
метаданные образов/инодов в etcd.
Когда пул помечается как используемый для VitastorFS, создание обычных блочных
образов в нём отключается (vitastor-cli отказывается создавать образы без --force),
чтобы защитить пользователя от коллизий ID файлов и блочных образов и, таким
образом, от потери данных.
Когда пул помечается используемым для такого приложения, создание обычных блочных
образов в нём запрещается (vitastor-cli отказывается создавать образы без --force),
чтобы защитить пользователя от коллизий ID блочных образов и томов ФС/S3, и,
таким образом, от потери данных.
Также для таких пулов отключается передача статистики в etcd по отдельным инодам,
так как использование для внешнего приложения подразумевает, что пул может содержать
очень много томов и их статистика может занять слишком много места в etcd.
Установка used_for_app в значение `fs:<name>` сообщает о том, что пул используется
для VitastorFS с базой метаданных VitastorKV, хранимой в блочном образе с именем
`<name>`.
[vitastor-nfs](../usage/nfs.ru.md), в свою очередь, при запуске отказывается
использовать для ФС пулы, не выделенные для неё. Это также означает, что один
пул может использоваться только для одной VitastorFS.
использовать для ФС пулы, не помеченные, как используемые для неё. Это также
означает, что один пул может использоваться только для одной VitastorFS.
Также для ФС-пулов отключается передача статистики в etcd по отдельным инодам,
так как ФС-пул может содержать очень много файлов и статистика по ним всем
заняла бы очень много места в etcd.
Если же вы планируете использовать пул для данных S3, установите его used_for_app
в значение `s3:<name>`, где `<name>` - любое название по вашему усмотрению
(например, `s3:standard`) - конкретное содержимое `<name>` пока никак не проверяется
компонентами Vitastor S3.
Смотрите также [allow_net_split](osd.ru.md#allow_net_split) и
[документацию по состояниям PG](../usage/admin.ru.md#состояния-pg).
Все остальные значения used_for_app, кроме начинающихся на `fs:` или `s3:`, не
означают ничего особенного для основных компонентов Vitastor. Поэтому сейчас вы
можете использовать их свободно любым желаемым способом.
# Примеры
+45
View File
@@ -271,3 +271,48 @@
заполненные на 100% OSD вообще не могут стартовать), так что вы сможете
восстановить работу кластера после ошибок отсутствия свободного места
без уничтожения и пересоздания OSD.
- name: hostname
type: string
online: true
info: |
Clients use host name to find their distance to OSDs when [localized reads](pool.en.md#local_reads)
are enabled. By default, standard [gethostname](https://man7.org/linux/man-pages/man2/gethostname.2.html)
function is used to determine host name, but you can also override it with this parameter.
info_ru: |
Клиенты используют имя хоста для определения расстояния до OSD, когда включены
[локальные чтения](pool.ru.md#local_reads). По умолчанию для определения имени
хоста используется стандартная функция [gethostname](https://man7.org/linux/man-pages/man2/gethostname.2.html),
но вы также можете задать имя хоста вручную данным параметром.
- name: ublk_queue_depth
type: int
default: 256
online: false
info: Default queue depth for [Vitastor ublk servers](../usage/ublk.en.md).
info_ru: Глубина очереди по умолчанию для [ublk-серверов Vitastor](../usage/ublk.ru.md).
- name: ublk_max_io_size
type: int
online: false
info: |
Default maximum I/O size for Vitastor [ublk servers](../usage/ublk.en.md).
The largest of 1 MB and pool block size multiplied by EC data chunk count is used if not specified.
info_ru: |
Максимальный размер запроса ввода-вывода для [ublk-серверов Vitastor](../usage/ublk.ru.md).
Если не задан, используется максимум из 1 МБ и размера блока пула, умноженного на число частей
данных EC-пула.
- name: qemu_file_mirror_path
type: string
info: |
When set to an FS directory path (for example, `/mnt/vitastor/`), `qemu-img info` and similar
QAPI commands return the name of the image inside this directory instead of normal
`vitastor://?image=abc` URI as `filename`.
This allows to then mount this path using [vitastor-nfs](../usage/nfs.en.md) and trick
third-party systems like Veeam which rely on `filename` in the image info but don't support Vitastor.
info_ru: |
Если установить эту опцию равной пути к каталогу в ФС, команда `qemu-img info` и подобные
команды QAPI будут возвращать в поле `filename` имя образа внутри заданного каталога вместо
обычного адреса типа `vitastor://?image=abc`.
Это позволяет смонтировать этот путь с помощью [vitastor-nfs](../usage/nfs.ru.md) и обмануть
сторонние системы типа Veeam, которые полагаются на поле `filename` в информации об образе QEMU,
но не поддерживают Vitastor.
+4
View File
@@ -24,6 +24,8 @@
{{../../installation/kubernetes.en.md}}
{{../../installation/s3.en.md}}
{{../../installation/source.en.md}}
{{../../config.en.md|indent=1}}
@@ -54,6 +56,8 @@
{{../../usage/fio.en.md}}
{{../../usage/ublk.en.md}}
{{../../usage/nbd.en.md}}
{{../../usage/qemu.en.md}}
+4
View File
@@ -26,6 +26,8 @@
{{../../installation/source.ru.md}}
{{../../installation/s3.ru.md}}
{{../../config.ru.md|indent=1}}
{{../../config/common.ru.md|indent=2}}
@@ -54,6 +56,8 @@
{{../../usage/fio.ru.md}}
{{../../usage/ublk.ru.md}}
{{../../usage/nbd.ru.md}}
{{../../usage/qemu.ru.md}}
+25
View File
@@ -1,3 +1,28 @@
- name: meta_format
type: int
default: 3
info: |
OSD store implementation version and on-disk metadata format.
Three versions are currently supported: 3, 2 and 1.
- 3 the new log-structured store, it's overall faster, has lower Write
Amplification, which may be even close to 1 (i.e. almost no extra writes)
if your SSDs support atomic writes (see [atomic_write_size](osd.en.md#atomic_write_size)).
- 2 is the old stable store from Vitastor 0.9-2.x.
- 1 is the same old store but with a legacy metadata format from Vitastor
versions to up 0.8.x, without any support for checksums.
info_ru: |
Версия реализации дискового хранилища OSD и дискового формата метаданных.
Поддерживаются три версии: 3, 2 и 1.
- 3 - новое лог-структурированное хранилище, в целом более быстрое, со
сниженным фактором амплификации записи, который может составлять около 1
(то есть, практически без лишней служебной записи), если ваши SSD
поддерживают атомарную запись (см. [atomic_write_size](osd.ru.md#atomic_write_size)).
- 2 - старое стабильное хранилище из версий Vitastor 0.9-2.x.
- 1 - то же самое стабильное хранилище, но с ещё более старым форматом
метаданных из версий Vitastor до 0.8.x, без какой-либо поддержки
контрольных сумм.
- name: data_device
type: string
info: |
+2 -2
View File
@@ -75,11 +75,11 @@
- name: mon_http_port
type: int
default: 8060
info: HTTP port for monitors to listen on (including metrics exporter)
info: HTTP port for monitors to listen to (including metrics exporter)
info_ru: Порт, на котором мониторы принимают HTTP-соединения (в том числе для отдачи метрик)
- name: mon_http_ip
type: string
info: IP address for monitors to listen on (all addresses by default)
info: IP address for monitors to listen to (all addresses by default)
info_ru: IP-адрес, на котором мониторы принимают HTTP-соединения (по умолчанию все адреса)
- name: mon_https_cert
type: string
+205 -92
View File
@@ -1,49 +1,78 @@
- name: tcp_header_buffer_size
type: int
default: 65536
- name: osd_network
type: string or array of strings
type_ru: строка или массив строк
info: |
Size of the buffer used to read data using an additional copy. Vitastor
packet headers are 128 bytes, payload is always at least 4 KB, so it is
usually beneficial to try to read multiple packets at once even though
it requires to copy the data an additional time. The rest of each packet
is received without an additional copy. You can try to play with this
parameter and see how it affects random iops and linear bandwidth if you
want.
Network mask of public OSD network(s) (IPv4 or IPv6). Each OSD listens to all
addresses of UP + RUNNING interfaces matching one of these networks, on the
same port. Port is auto-selected except if [bind_port](osd.en.md#bind_port) is
explicitly specified. Bind address(es) may also be overridden manually by
specifying [bind_address](osd.en.md#bind_address). If OSD networks are not specified
at all, OSD just listens to a wildcard address (0.0.0.0).
info_ru: |
Размер буфера для чтения данных с дополнительным копированием. Пакеты
Vitastor содержат 128-байтные заголовки, за которыми следуют данные размером
от 4 КБ и для мелких операций ввода-вывода обычно выгодно за 1 вызов читать
сразу несколько пакетов, даже не смотря на то, что это требует лишний раз
скопировать данные. Часть каждого пакета за пределами значения данного
параметра читается без дополнительного копирования. Вы можете попробовать
поменять этот параметр и посмотреть, как он влияет на производительность
случайного и линейного доступа.
- name: use_sync_send_recv
type: bool
default: false
Маски подсетей (IPv4 или IPv6) публичной сети или сетей OSD. Каждый OSD слушает
один и тот же порт на всех адресах поднятых (UP + RUNNING) сетевых интерфейсов,
соответствующих одной из указанных сетей. Порт выбирается автоматически, если
только [bind_port](osd.ru.md#bind_port) не задан явно. Адреса для подключений можно
также переопределить явно, задав [bind_address](osd.ru.md#bind_address). Если сети OSD
не заданы вообще, OSD слушает все адреса (0.0.0.0).
- name: osd_cluster_network
type: string or array of strings
type_ru: строка или массив строк
info: |
If true, synchronous send/recv syscalls are used instead of io_uring for
socket communication. Useless for OSDs because they require io_uring anyway,
but may be required for clients with old kernel versions.
Network mask of separate network(s) (IPv4 or IPv6) to use for OSD
cluster connections. I.e. OSDs will always attempt to use these networks
to connect to other OSDs, while clients will attempt to use networks from
[osd_network](#osd_network).
info_ru: |
Если установлено в истину, то вместо io_uring для передачи данных по сети
будут использоваться обычные синхронные системные вызовы send/recv. Для OSD
это бессмысленно, так как OSD в любом случае нуждается в io_uring, но, в
принципе, это может применяться для клиентов со старыми версиями ядра.
Маски подсетей (IPv4 или IPv6) отдельной кластерной сети или сетей OSD.
То есть, OSD будут всегда стараться использовать эти сети для соединений
с другими OSD, а клиенты будут стараться использовать сети из [osd_network](#osd_network).
- name: use_rdma
type: bool
default: true
info: |
Try to use RDMA for communication if it's available. Disable if you don't
want Vitastor to use RDMA. TCP-only clients can also talk to an RDMA-enabled
cluster, so disabling RDMA may be needed if clients have RDMA devices,
but they are not connected to the cluster.
Try to use RDMA through libibverbs for communication if it's available.
Disable if you don't want Vitastor to use RDMA. TCP-only clients can also
talk to an RDMA-enabled cluster, so disabling RDMA may be needed if clients
have RDMA devices, but they are not connected to the cluster.
`use_rdma` works with RoCEv1/RoCEv2 networks, but not with iWARP and,
maybe, with some Infiniband configurations which require RDMA-CM.
Consider `use_rdmacm` for such networks.
info_ru: |
Пытаться использовать RDMA для связи при наличии доступных устройств.
Отключите, если вы не хотите, чтобы Vitastor использовал RDMA.
TCP-клиенты также могут работать с RDMA-кластером, так что отключать
RDMA может быть нужно только если у клиентов есть RDMA-устройства,
но они не имеют соединения с кластером Vitastor.
Попробовать использовать RDMA через libibverbs для связи при наличии
доступных устройств. Отключите, если вы не хотите, чтобы Vitastor
использовал RDMA. TCP-клиенты также могут работать с RDMA-кластером,
так что отключать RDMA может быть нужно, только если у клиентов есть
RDMA-устройства, но они не имеют соединения с кластером Vitastor.
`use_rdma` работает с RoCEv1/RoCEv2 сетями, но не работает с iWARP и
может не работать с частью конфигураций Infiniband, требующих RDMA-CM.
Рассмотрите включение `use_rdmacm` для таких сетей.
- name: use_rdmacm
type: bool
default: false
info: |
Use an alternative implementation of RDMA through RDMA-CM (Connection
Manager). Works with all RDMA networks: Infiniband, iWARP and
RoCEv1/RoCEv2, and even allows to disable TCP and run only with RDMA.
OSDs always use random port numbers for RDMA-CM listeners, different
from their TCP ports. `use_rdma` is automatically disabled when
`use_rdmacm` is enabled.
info_ru: |
Использовать альтернативную реализацию RDMA на основе RDMA-CM (Connection
Manager). Работает со всеми типами RDMA-сетей: Infiniband, iWARP и
RoCEv1/RoCEv2, и даже позволяет полностью отключить TCP и работать
только на RDMA. OSD используют случайные номера портов для ожидания
соединений через RDMA-CM, отличающиеся от их TCP-портов. Также при
включении `use_rdmacm` автоматически отключается опция `use_rdma`.
- name: disable_tcp
type: bool
default: true
info: |
Fully disable TCP and only use RDMA-CM for OSD communication.
info_ru: |
Полностью отключить TCP и использовать только RDMA-CM для соединений с OSD.
- name: rdma_device
type: string
info: |
@@ -55,11 +84,6 @@
unsupported with old libibverbs < v32, like in Debian 10 Buster or
CentOS 7.
Vitastor supports all adapters, even ones without ODP support, like
Mellanox ConnectX-3 and non-Mellanox cards. Versions up to Vitastor
1.2.0 required ODP which is only present in Mellanox ConnectX >= 4.
See also [rdma_odp](#rdma_odp).
Run `ibv_devinfo -v` as root to list available RDMA devices and their
features.
@@ -68,6 +92,23 @@
the manual of your network vendor for details about setting up the switch
for RoCEv2 correctly. Usually it means setting up Lossless Ethernet with
PFC (Priority Flow Control) and ECN (Explicit Congestion Notification).
Vitastor supports all adapters, even ones without ODP (On-Demand Paging)
support, like Mellanox ConnectX-3 and non-Mellanox cards. ODP is only present
in Mellanox ConnectX >= 4 adapters and allows to skip memory registration
for RDMA and thus, in theory, avoid memory copying.
Versions up to Vitastor 1.2.0 required ODP, then it was disabled by default,
but it was still supported up to 3.0.3. Now ODP support is removed because it
actually only hurts performance: an example 3-node cluster with 8 NVMe in each
node and 2*25 GBit/s ConnectX-6 RDMA network pushed 3950000 read iops without
ODP, but only 239000 iops with ODP.
This happens because Mellanox ODP implementation seems to be based on
message retransmissions when the adapter doesn't know about the buffer yet -
it likely uses standard "RNR retransmissions" (RNR = receiver not ready)
which is generally slow in RDMA/RoCE networks. Here's a presentation about
it from ISPASS-2021 conference: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
info_ru: |
Название RDMA-устройства для связи с Vitastor OSD (например, "rocep5s0f0").
Если не указано, Vitastor попробует найти RoCE-устройство, соответствующее
@@ -76,12 +117,6 @@
не задана. Также автовыбор не поддерживается со старыми версиями библиотеки
libibverbs < v32, например в Debian 10 Buster или CentOS 7.
Vitastor поддерживает все модели адаптеров, включая те, у которых
нет поддержки ODP, то есть вы можете использовать RDMA с ConnectX-3 и
картами производства не Mellanox. Версии Vitastor до 1.2.0 включительно
требовали ODP, который есть только на Mellanox ConnectX 4 и более новых.
См. также [rdma_odp](#rdma_odp).
Запустите `ibv_devinfo -v` от имени суперпользователя, чтобы посмотреть
список доступных RDMA-устройств, их параметры и возможности.
@@ -91,18 +126,39 @@
коммутатора для RoCEv2 ищите в документации производителя. Обычно это
подразумевает настройку сети без потерь на основе PFC (Priority Flow
Control) и ECN (Explicit Congestion Notification).
Vitastor поддерживает все модели адаптеров, включая те, у которых нет
поддержки ODP (On-Demand Paging), например, ConnectX-3 и карты производства
не Mellanox. Функция ODP доступна только на адаптерах Mellanox ConnectX-4 и
более новых и позволяет не регистрировать память для её использования RDMA-картой,
благодаря чему в теории можно избежать лишних копирований памяти.
Версии Vitastor до 1.2.0 включительно требовали ODP, потом функция был отключена
по умолчанию, но поддерживалась вплоть до версии 3.0.3. Сейчас поддержка ODP
полностью удалена, так как на самом деле она только портит производительность:
например, на 3-узловом кластере с 8 NVMe в каждом узле и сетью 2*25 Гбит/с на
чтение с RDMA без ODP удаётся снять 3950000 iops, а с ODP - всего 239000 iops.
Это происходит из-за того, что реализация ODP у Mellanox неоптимальная и
основана на повторной передаче сообщений, когда карте не известен буфер -
вероятно, на стандартных "RNR retransmission" (RNR = receiver not ready).
А данные повторные передачи в RDMA/RoCE - всегда очень медленная штука.
Презентация на эту тему с конференции ISPASS-2021: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
- name: rdma_port_num
type: int
default: 1
info: |
RDMA device port number to use. Only for devices that have more than 1 port.
See `phys_port_cnt` in `ibv_devinfo -v` output to determine how many ports
your device has.
Not relevant for RDMA-CM (use_rdmacm).
info_ru: |
Номер порта RDMA-устройства, который следует использовать. Имеет смысл
только для устройств, у которых более 1 порта. Чтобы узнать, сколько портов
у вашего адаптера, посмотрите `phys_port_cnt` в выводе команды
`ibv_devinfo -v`.
Опция неприменима к RDMA-CM (use_rdmacm).
- name: rdma_gid_index
type: int
info: |
@@ -116,6 +172,8 @@
GID auto-selection is unsupported with libibverbs < v32.
A correct rdma_gid_index for RoCEv2 is usually 1 (IPv6) or 3 (IPv4).
Not relevant for RDMA-CM (use_rdmacm).
info_ru: |
Номер глобального идентификатора адреса RDMA-устройства, который следует
использовать. Разным gid_index могут соответствовать разные протоколы связи:
@@ -128,15 +186,16 @@
libibverbs < v32.
Правильный rdma_gid_index для RoCEv2, как правило, 1 (IPv6) или 3 (IPv4).
Опция неприменима к RDMA-CM (use_rdmacm).
- name: rdma_mtu
type: int
default: 4096
info: |
RDMA Path MTU to use. Must be 1024, 2048 or 4096. There is usually no
sense to change it from the default 4096.
RDMA Path MTU to use. Must be 1024, 2048 or 4096. Default is to use the
RDMA device's MTU.
info_ru: |
Максимальная единица передачи (Path MTU) для RDMA. Должно быть равно 1024,
2048 или 4096. Обычно нет смысла менять значение по умолчанию, равное 4096.
2048 или 4096. По умолчанию используется значение MTU RDMA-устройства.
- name: rdma_max_sge
type: int
default: 128
@@ -183,45 +242,6 @@
у принимающей стороны в процессе работы не заканчивались буферы на приём.
Не влияет на потребление памяти - дополнительная память на операции отправки
не выделяется.
- name: rdma_odp
type: bool
default: false
online: false
info: |
Use RDMA with On-Demand Paging. ODP is currently only available on Mellanox
ConnectX-4 and newer adapters. ODP allows to not register memory explicitly
for RDMA adapter to be able to use it. This, in turn, allows to skip memory
copying during sending. One would think this should improve performance, but
**in reality** RDMA performance with ODP is **drastically** worse. Example
3-node cluster with 8 NVMe in each node and 2*25 GBit/s ConnectX-6 RDMA network
without ODP pushes 3950000 read iops, but only 239000 iops with ODP...
This happens because Mellanox ODP implementation seems to be based on
message retransmissions when the adapter doesn't know about the buffer yet -
it likely uses standard "RNR retransmissions" (RNR = receiver not ready)
which is generally slow in RDMA/RoCE networks. Here's a presentation about
it from ISPASS-2021 conference: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
ODP support is retained in the code just in case a good ODP implementation
appears one day.
info_ru: |
Использовать RDMA с On-Demand Paging. ODP - функция, доступная пока что
исключительно на адаптерах Mellanox ConnectX-4 и более новых. ODP позволяет
не регистрировать память для её использования RDMA-картой. Благодаря этому
можно не копировать данные при отправке их в сеть и, казалось бы, это должно
улучшать производительность - но **по факту** получается так, что
производительность только ухудшается, причём сильно. Пример - на 3-узловом
кластере с 8 NVMe в каждом узле и сетью 2*25 Гбит/с на чтение с RDMA без ODP
удаётся снять 3950000 iops, а с ODP - всего 239000 iops...
Это происходит из-за того, что реализация ODP у Mellanox неоптимальная и
основана на повторной передаче сообщений, когда карте не известен буфер -
вероятно, на стандартных "RNR retransmission" (RNR = receiver not ready).
А данные повторные передачи в RDMA/RoCE - всегда очень медленная штука.
Презентация на эту тему с конференции ISPASS-2021: https://tkygtr6.github.io/pub/ISPASS21_slides.pdf
Возможность использования ODP сохранена в коде на случай, если вдруг в один
прекрасный день появится хорошая реализация ODP.
- name: peer_connect_interval
type: sec
min: 1
@@ -306,3 +326,96 @@
detect disconnections quickly.
info_ru: |
Интервал проверки живости вебсокет-подключений к etcd.
- name: etcd_min_reload_interval
type: ms
default: 1000
online: true
info: |
Minimum interval for full etcd state reload. Introduced to prevent
excessive load on etcd during outages when etcd can't keep up with event
streams and cancels them.
info_ru: |
Минимальный интервал полной перезагрузки состояния из etcd. Добавлено для
предотвращения избыточной нагрузки на etcd во время отказов, когда etcd не
успевает рассылать потоки событий и отменяет их.
- name: tcp_header_buffer_size
type: int
default: 65536
info: |
Size of the buffer used to read data using an additional copy. Vitastor
packet headers are 128 bytes, payload is always at least 4 KB, so it is
usually beneficial to try to read multiple packets at once even though
it requires to copy the data an additional time. The rest of each packet
is received without an additional copy. You can try to play with this
parameter and see how it affects random iops and linear bandwidth if you
want.
info_ru: |
Размер буфера для чтения данных с дополнительным копированием. Пакеты
Vitastor содержат 128-байтные заголовки, за которыми следуют данные размером
от 4 КБ и для мелких операций ввода-вывода обычно выгодно за 1 вызов читать
сразу несколько пакетов, даже не смотря на то, что это требует лишний раз
скопировать данные. Часть каждого пакета за пределами значения данного
параметра читается без дополнительного копирования. Вы можете попробовать
поменять этот параметр и посмотреть, как он влияет на производительность
случайного и линейного доступа.
- name: min_zerocopy_send_size
type: int
default: 32768
info: |
OSDs and clients will attempt to use io_uring-based zero-copy TCP send
for buffers larger than this number of bytes. Zero-copy send with io_uring is
supported since Linux kernel version 6.1. Support is auto-detected and disabled
automatically when not available. It can also be disabled explicitly by setting
this parameter to a negative value.
⚠️ Warning! Zero-copy send performance may vary greatly from CPU to CPU and from
one kernel version to another. Generally, it tends to only make benefit with larger
messages. With smaller messages (say, 4 KB), it may actually be slower. 32 KB is
enough for almost all CPUs, but even smaller values are optimal for some of them.
For example, 4 KB is OK for EPYC Milan/Genoa and 12 KB is OK for Xeon Ice Lake
(but verify it yourself please).
Verification instructions:
1. Add `iommu=pt` into your Linux kernel command line and reboot.
2. Upgrade your kernel. For example, it's very important to use 6.11+ with recent AMD EPYCs.
3. Run some tests with the [send-zerocopy liburing example](https://github.com/axboe/liburing/blob/master/examples/send-zerocopy.c)
to find the minimal message size for which zero-copy is optimal.
Use `./send-zerocopy tcp -4 -R` at the server side and
`time ./send-zerocopy tcp -4 -b 0 -s BUFFER_SIZE -D SERVER_IP` at the client side with
`-z 0` (no zero-copy) and `-z 1` (zero-copy), and compare MB/s and used CPU time
(user+system).
info_ru: |
OSD и клиенты будут пробовать использовать TCP-отправку без копирования (zero-copy) на
основе io_uring для буферов, больших, чем это число байт. Отправка без копирования
поддерживается в io_uring, начиная с версии ядра Linux 6.1. Наличие поддержки
проверяется автоматически и zero-copy отключается, когда поддержки нет. Также
её можно отключить явно, установив данный параметр в отрицательное значение.
⚠️ Внимание! Производительность данной функции может сильно отличаться на разных
процессорах и на разных версиях ядра Linux. В целом, zero-copy обычно быстрее с
большими сообщениями, а с мелкими (например, 4 КБ) zero-copy может быть даже
медленнее. 32 КБ достаточно почти для всех процессоров, но для каких-то можно
использовать даже меньшие значения. Например, для EPYC Milan/Genoa подходит 4 КБ,
а для Xeon Ice Lake - 12 КБ (но, пожалуйста, перепроверьте это сами).
Инструкция по проверке:
1. Добавьте `iommu=pt` в командную строку загрузки вашего ядра Linux и перезагрузитесь.
2. Обновите ядро. Например, для AMD EPYC очень важно использовать версию 6.11+.
3. Позапускайте тесты с помощью [send-zerocopy из примеров liburing](https://github.com/axboe/liburing/blob/master/examples/send-zerocopy.c),
чтобы найти минимальный размер сообщения, для которого zero-copy отправка оптимальна.
Запускайте `./send-zerocopy tcp -4 -R` на стороне сервера и
`time ./send-zerocopy tcp -4 -b 0 -s РАЗМЕР_БУФЕРА -D АДРЕС_СЕРВЕРА` на стороне клиента
с опцией `-z 0` (обычная отправка) и `-z 1` (отправка без копирования), и сравнивайте
скорость в МБ/с и занятое процессорное время (user+system).
- name: use_sync_send_recv
type: bool
default: false
info: |
If true, synchronous send/recv syscalls are used instead of io_uring for
socket communication. Useless for OSDs because they require io_uring anyway,
but may be required for clients with old kernel versions.
info_ru: |
Если установлено в истину, то вместо io_uring для передачи данных по сети
будут использоваться обычные синхронные системные вызовы send/recv. Для OSD
это бессмысленно, так как OSD в любом случае нуждается в io_uring, но, в
принципе, это может применяться для клиентов со старыми версиями ядра.
+2 -2
View File
@@ -1,5 +1,5 @@
# Runtime OSD Parameters
These parameters only apply to OSDs, are not fixed at the moment of OSD drive
initialization and can be changed - either with an OSD restart or, for some of
them, even without restarting by updating configuration in etcd.
initialization and can be changed - in /etc/vitastor/vitastor.conf or [vitastor-disk update-sb](../usage/disk.en.md#update-sb)
with an OSD restart or, for some of them, even without restarting by updating configuration in etcd.
+2 -2
View File
@@ -2,5 +2,5 @@
Данные параметры используются только OSD, но, в отличие от дисковых параметров,
не фиксируются в момент инициализации дисков OSD и могут быть изменены в любой
момент с помощью перезапуска OSD, а некоторые и без перезапуска, с помощью
изменения конфигурации в etcd.
момент с перезапуском OSD в /etc/vitastor/vitastor.conf или [vitastor-disk update-sb](../usage/disk.ru.md#update-sb),
а некоторые и без перезапуска, с помощью изменения конфигурации в etcd.
+227 -54
View File
@@ -1,3 +1,26 @@
- name: bind_address
type: string or array of strings
type_ru: строка или массив строк
info: |
Instead of the network masks ([osd_network](network.en.md#osd_network) and
[osd_cluster_network](network.en.md#osd_cluster_network)), you can also set
OSD listen addresses explicitly using this parameter. May be useful if you
want to start OSDs on interfaces that are not UP + RUNNING.
info_ru: |
Вместо использования масок подсети ([osd_network](network.ru.md#osd_network) и
[osd_cluster_network](network.ru.md#osd_cluster_network)), вы также можете явно
задать адрес(а), на которых будут ожидать соединений OSD, с помощью данного
параметра. Это может быть полезно, например, чтобы запускать OSD на неподнятых
интерфейсах (не UP + RUNNING).
- name: bind_port
type: int
info: |
By default, OSDs pick random ports to use for incoming connections
automatically. With this option you can set a specific port for a specific
OSD by hand.
info_ru: |
По умолчанию OSD сами выбирают случайные порты для входящих подключений.
С помощью данной опции вы можете задать порт для отдельного OSD вручную.
- name: osd_iothread_count
type: int
default: 0
@@ -56,44 +79,6 @@
реализовать дополнительный режим для монитора, который позволит отделять
первичные OSD от вторичных, но пока не понятно, зачем это может кому-то
понадобиться, поэтому это не реализовано.
- name: osd_network
type: string or array of strings
type_ru: строка или массив строк
info: |
Network mask of the network (IPv4 or IPv6) to use for OSDs. Note that
although it's possible to specify multiple networks here, this does not
mean that OSDs will create multiple listening sockets - they'll only
pick the first matching address of an UP + RUNNING interface. Separate
networks for cluster and client connections are also not implemented, but
they are mostly useless anyway, so it's not a big deal.
info_ru: |
Маска подсети (IPv4 или IPv6) для использования для соединений с OSD.
Имейте в виду, что хотя сейчас и можно передать в этот параметр несколько
подсетей, это не означает, что OSD будут создавать несколько слушающих
сокетов - они лишь будут выбирать адрес первого поднятого (состояние UP +
RUNNING), подходящий под заданную маску. Также не реализовано разделение
кластерной и публичной сетей OSD. Правда, от него обычно всё равно довольно
мало толку, так что особенной проблемы в этом нет.
- name: bind_address
type: string
default: "0.0.0.0"
info: |
Instead of the network mask, you can also set OSD listen address explicitly
using this parameter. May be useful if you want to start OSDs on interfaces
that are not UP + RUNNING.
info_ru: |
Этим параметром можно явным образом задать адрес, на котором будет ожидать
соединений OSD (вместо использования маски подсети). Может быть полезно,
например, чтобы запускать OSD на неподнятых интерфейсах (не UP + RUNNING).
- name: bind_port
type: int
info: |
By default, OSDs pick random ports to use for incoming connections
automatically. With this option you can set a specific port for a specific
OSD by hand.
info_ru: |
По умолчанию OSD сами выбирают случайные порты для входящих подключений.
С помощью данной опции вы можете задать порт для отдельного OSD вручную.
- name: autosync_interval
type: sec
default: 5
@@ -268,21 +253,33 @@
type: bool
default: true
info: |
This parameter makes Vitastor always keep metadata area of the block device
in memory. It's required for good performance because it allows to avoid
additional read-modify-write cycles during metadata modifications. Metadata
area size is currently roughly 224 MB per 1 TB of data. You can turn it off
to reduce memory usage by this value, but it will hurt performance. This
restriction is likely to be removed in the future along with the upgrade
of the metadata storage scheme.
Only for the old store ([meta_format](layout-osd.en.md#meta_format) 2).
This parameter makes Vitastor keep a copy of metadata area in memory as it is
on disk, in addition to the metadata database. When the option is enabled, every
metadata entry is effectively stored in RAM twice. It's required for good performance
because it allows to avoid additional read-modify-write cycles during metadata
modifications. Metadata area size with the old store is roughly 224 MB per 1 TB
of data. You can turn the option off to reduce memory usage by this value, but
it will reduce performance.
For the new store ([meta_format](layout-osd.en.md#meta_format) 3), the option
may be changed in the future to support operation without loading full metadata
database in memory.
info_ru: |
Данный параметр заставляет Vitastor всегда держать область метаданных диска
в памяти. Это нужно, чтобы избегать дополнительных операций чтения с диска
при записи. Размер области метаданных на данный момент составляет примерно
224 МБ на 1 ТБ данных. При включении потребление памяти снизится примерно
на эту величину, но при этом также снизится и производительность. В будущем,
после обновления схемы хранения метаданных, это ограничение, скорее всего,
будет ликвидировано.
Только для старого хранилища ([meta_format](layout-osd.en.md#meta_format) 2).
Данный параметр заставляет Vitastor всегда держать копию области метаданных
в памяти в том же виде, как она лежит на диске, в дополнение к БД метаданных.
То есть, с включённой опцией каждая запись метаданных хранится в памяти дважды.
Это нужно, чтобы избегать дополнительных операций чтения с диска при записи.
Размер области метаданных в старом хранилище составляет примерно 224 МБ на
1 ТБ данных. Вы можете отключить опцию, чтобы снизить потребление памяти
примерно на эту величину, но при этом также снизится и производительность.
Для нового хранилища ([meta_format](layout-osd.en.md#meta_format) 3) опция,
возможно, будет переработана в будущем для поддержки работы без полной
загрузки метаданных в памяти.
- name: inmemory_journal
type: bool
default: true
@@ -315,7 +312,7 @@
decrease write performance for fast disks because page cache is an overhead
itself.
Choose "directsync" to use [immediate_commit](layout-cluster.ru.md#immediate_commit)
Choose "directsync" to use [immediate_commit](layout-cluster.en.md#immediate_commit)
(which requires disable_data_fsync) with drives having write-back cache
which can't be turned off, for example, Intel Optane. Also note that *some*
desktop SSDs (for example, HP EX950) may ignore O_SYNC thus making
@@ -401,11 +398,15 @@
blocks. The only situation when you should increase it to a larger value
is when you enable journal_no_same_sector_overwrites. In this case set
it to, for example, 1024.
Not applicable to the new store ([meta_format](layout-osd.en.md#meta_format) 3).
info_ru: |
Максимальное число буферов, разрешённых для использования под записываемые
в журнал блоки метаданных. Единственная ситуация, в которой этот параметр
нужно менять - это если вы включаете journal_no_same_sector_overwrites. В
этом случае установите данный параметр, например, в 1024.
Неприменимо к новому хранилищу ([meta_format](layout-osd.en.md#meta_format) 3).
- name: journal_no_same_sector_overwrites
type: bool
default: false
@@ -417,6 +418,8 @@
journal after writing it instead of possibly overwriting it the second time.
Most (99%) other SSDs don't need this option.
Not applicable to the new store ([meta_format](layout-osd.en.md#meta_format) 3).
info_ru: |
Включайте данную опцию для SSD вроде Intel D3-S4510 и D3-S4610, которые
ОЧЕНЬ не любят, когда ПО перезаписывает один и тот же сектор несколько раз
@@ -427,6 +430,20 @@
самого сектора.
Почти все другие SSD (99% моделей) не требуют данной опции.
Неприменимо к новому хранилищу ([meta_format](layout-osd.en.md#meta_format) 3).
- name: skip_corrupted_meta_entries
type: bool
default: false
info: |
Only for the new store ([meta_format](layout-osd.en.md#meta_format) 3).
Allow OSD to start when some metadata entries or blocks are corrupted by
skipping them. Should be only used as an emergency measure.
info_ru: |
Только для нового хранилища ([meta_format](layout-osd.en.md#meta_format) 3).
Разрешить OSD запускаться, даже если часть блоков или записей метаданных
повреждена, пропуская их. Опция предназначена для использования только в
целях аварийного восстановления.
- name: throttle_small_writes
type: bool
default: false
@@ -581,7 +598,7 @@
сильно влияет на пользовательскую нагрузку.
- name: scrub_list_limit
type: int
default: 1000
default: 262144
online: true
info: |
Number of objects to list in one listing operation during scrub.
@@ -765,3 +782,159 @@
Максимальное возможное значение авто-подстроенного recovery_sleep_us.
Большие значения считаются случайными выбросами и игнорируются в
усреднении.
- name: discard_on_start
type: bool
info: Discard (SSD TRIM) unused data device blocks on every OSD startup.
info_ru: Освобождать (SSD TRIM) неиспользуемые блоки диска данных при каждом запуске OSD.
- name: min_discard_size
type: int
default: 1048576
info: Minimum consecutive block size to TRIM it.
info_ru: Минимальный размер последовательного блока данных, чтобы освобождать его через TRIM.
- name: allow_net_split
type: bool
default: false
info: |
Allow "safe" cases of network splits/partitions - allow to start PGs without
connections to some OSDs currently registered as alive in etcd, if the number
of actually connected PG OSDs is at least pg_minsize. That is, allow some OSDs to lose
connectivity with some other OSDs as long as it doesn't break pg_minsize guarantees.
The downside is that it increases the probability of writing data into just pg_minsize
OSDs during failover which can lead to PGs becoming incomplete after additional outages.
The old behaviour in versions up to 2.0.0 was equal to enabled allow_net_split.
info_ru: |
Разрешить "безопасные" случаи разделений сети - разрешить активировать PG без
соединений к некоторым OSD, помеченным активными в etcd, если общее число активных
OSD в PG составляет как минимум pg_minsize. То есть, разрешать некоторым OSD терять
соединения с некоторыми другими OSD, если это не нарушает гарантий pg_minsize.
Минус такого разрешения в том, что оно повышает вероятность записи данных ровно в
pg_minsize OSD во время переключений, что может потом привести к тому, что PG станут
неполными (incomplete), если упадут ещё какие-то OSD.
Старое поведение в версиях до 2.0.0 было идентично включённому allow_net_split.
- name: enable_pg_locks
type: bool
info: |
Vitastor 2.2.0 introduces a new layer of split-brain prevention mechanism in
addition to etcd: PG locks. They prevent split-brain even in abnormal theoretical cases
when etcd is extremely laggy. As a new feature, by default, PG locks are only enabled
for pools where they're required - pools with [localized reads](pool.en.md#local_reads).
Use this parameter to enable or disable this function for all pools.
info_ru: |
В Vitastor 2.2.0 появился новый слой защиты от сплитбрейна в дополнение к etcd -
блокировки PG. Они гарантируют порядок даже в теоретических ненормальных случаях,
когда etcd очень сильно тормозит. Так как функция новая, по умолчанию она включается
только для пулов, в которых она необходима - а именно, в пулах с включёнными
[локальными чтениями](pool.ru.md#local_reads). Ну а с помощью данного параметра
можно включить блокировки PG для всех пулов.
- name: pg_lock_retry_interval_ms
type: ms
default: 100
info: Retry interval for failed PG lock attempts.
info_ru: Интервал повтора неудачных попыток блокировки PG.
- name: atomic_write_size
type: int
default: 4096
info: |
Maximum data device atomic write size allowed for OSD to use.
Atomic writes allow to reduce the Write Amplification factor with the new store
([meta_format](layout-osd.en.md#meta_format)=3) to almost 1 (i.e. almost no extra writes)
with replicated pools and reach the best possible write performance.
Default value is auto-detected during OSD initialization from
`/sys/block/xx/queue/atomic_write_max_bytes` or assumed to be 4096 bytes
because all known disks support 4 KB atomic writes. Auto-detection is only used for
NVMe disks because SAS disks require the explicit WRITE ATOMIC command which requires
RWF_ATOMIC (see below [#use_atomic_flag]) but that flag works incorrectly in current
Linux versions.
You can also check if your NVMe drives support atomic writes by running
the command `nvme id-ctrl /dev/nvme0n1 | grep awupf`. If the reported value,
plus 1, multiplied by the currently selected block size of the NVMe,
is more than 4 KB, then the new store can utilize it for better performance.
The only drives known to support it currently are [Micron and Kioxia](../intro/quickstart.en.md).
Atomic writes allow to skip double data writes in replicated pools, thus
reducing Write Amplification and improving write performance up to 2 times.
info_ru: |
Максимальный размер атомарной записи на диск данных, который OSD разрешено использовать.
Поддержка атомарной записи позволяет снизить мультипликатор записи (Write Amplification)
на диск с новым хранилищем ([meta_format](layout-osd.ru.md#meta_format)=3)
практически до 1 (то есть, почти до нулевого объёма лишней записи) в реплицированных
пулах и достигнуть наилучшей возможной производительности записи.
Значение по умолчанию авто-определяется во время инициализации OSD из
`/sys/block/xx/queue/atomic_write_max_bytes` либо принимается равным 4096,
так как все известные диски поддерживают атомарную запись 4 КБ блоков.
Автоопределение применяется только для NVMe-дисков, так как SAS диски требуют
использования отдельной команды WRITE ATOMIC, а для неё нужен флаг RWF_ATOMIC
(см. ниже [#use_atomic_flag]), а он в текущих версиях Linux работает некорректно.
Вы также можете проверить, поддерживают ли ваши NVMe-диски атомарную запись,
с помощью команды `nvme id-ctrl /dev/nvme0n1 | grep awupf`. Если значение awupf
плюс 1, умноженное на текущий выбранный размер блока NVMe-диска, больше 4 КБ,
то новое хранилище может использовать атомарные записи для достижения лучшей
производительности. Единственные известные диски, которые поддерживают это сейчас -
[Micron и Kioxia](../intro/quickstart.ru.md).
Атомарная запись позволяет не использовать двойную запись данных (в журнал и на
устройство данных) в реплицированных пулах и таким образом снижает амплификацию
записи (объём служебной записи на диск) и улучшает производительность записи
вплоть до 2-х кратного прироста.
- name: use_atomic_flag
type: bool
info: |
This option controls whether Vitastor OSDs use RWF_ATOMIC write flag with atomic writes.
This flag is supported since Linux 6.11 and adds some safety to atomic writes - the kernel
guarantees to not fragment write requests with it and also to check them against the actual
device atomic write capabilities.
However, the option is disabled by default because the flag is currently UNUSABLE - Linux
incorrectly requires writes with that flag to be of power-of-2 length and length-aligned.
I.e., for example, 12 KB writes and not-8-KB aligned 8 KB writes are forbidden by the kernel,
even though the NVMe specification allows them.
For NVMe disks with `scheduler=none` writes aren't fragmented anyway so it's not a big deal.
However, you can rebuild your kernel with [this patch](../../patches/linux-fix-atomic-write-checks.diff)
and turn this option on. It will make your atomic writes a bit safer.
info_ru: |
Данная опция контролирует использование Vitastor OSD флага RWF_ATOMIC при атомарной записи
блоков. Этот флаг поддерживается, начиная с версии ядра Linux 6.11 и добавляет немного корректности
атомарным записям - ядро гарантирует отсутствие фрагментации запросов записи с этим флагом и
проверяет их на соответствие реальным возможностям устройства.
Однако, данная опция по умолчанию отключена, так как флаг в текущих версиях Linux работает
абсолютно НЕКОРРЕКТНО - при нём Linux требует, чтобы запросы записи имели длину, равную
степени двойки и были выровнены на эту длину. То есть, например, 12 КБ запросы записи, а также
8 КБ запросы записи по не-кратному 8 КБ смещению запрещаются ядром, хотя спецификация NVMe их
разрешает.
Для NVMe-дисков с `scheduler=none` запросы записи и так не фрагментируются, так что это не так
уж и важно, однако вы можете пересобрать своё ядро с [этим патчем](../../patches/linux-fix-atomic-write-checks.diff)
и включить данную опцию. Это сделает вашу атомарную запись капельку безопаснее.
- name: pg_reshard_chunk_size
type: int
default: 100000
info: |
Pool PG count change is a CPU-intensive operation because OSDs store the full object database
in memory and have to move all entries between old and new PGs. Thus it's performed in chunks,
with pauses between chunks to prevent blocking OSD's event loop and other clients' operations.
This option sets the maximum number of object is a chunk. Moving 100k objects usually takes
50-100ms. Chunk size equal to 0 means unlimited.
info_ru: |
Изменение числа PG в пуле заметно загружает процессор, так как OSD хранят полную базу данных
объектов в памяти и им приходится перемещать все записи объектов между старыми и новыми PG.
Поэтому изменение применяется порциями, с паузами между порциями, чтобы не блокировать обработку
событий OSD и операции остальных клиентов. Данная опция задаёт максимальное число объектов
в порции. Перемещение 100 тысяч объектов (значение по умолчанию) обычно занимает порядка
50-100 миллисекунд. Значение опции 0 отключает лимит размера порции.
- name: pg_reshard_chunk_pause_ms
type: ms
default: 100
info: |
This option sets the interval between handling two PG count change chunks.
info_ru: |
Данная опция задаёт интервал между обработкой двух порций изменения числа PG пулов.
+27 -3
View File
@@ -26,13 +26,37 @@ at Vitastor Kubernetes operator: https://github.com/Antilles7227/vitastor-operat
The instruction is very simple.
1. Download a Docker image of the desired version: \
`docker pull vitastor:1.10.2`
`docker pull vitalif/vitastor:v3.0.8`
2. Install scripts to the host system: \
`docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitastor:1.10.2 install.sh`
`docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.8 install.sh`
3. Reload udev rules: \
`udevadm control --reload-rules`
4. Enable the vitastor-host service: \
`systemctl enable --now vitastor-host`
And you can return to [Quick Start](../intro/quickstart.en.md).
After these steps, you can return to [Quick Start](../intro/quickstart.en.md).
## Podman
If you use Podman, run the following commands as root before installing Vitastor containers:
```
ln -s podman /usr/bin/docker
mkdir -p /etc/systemd/system/systemd-udevd.service.d
cat >/etc/systemd/system/systemd-udevd.service.d/override.conf <<EOF
[Service]
CapabilityBoundingSet=~
SystemCallFilter=@mount capset
EOF
systemctl daemon-reload
systemctl restart systemd-udevd
```
Without it, udev fails to do calls into a Podman container and Vitastor disk detection doesn't work.
## Upgrading Containers
+27 -2
View File
@@ -25,14 +25,39 @@ Vitastor можно установить в Docker/Podman. При этом etcd,
Инструкция по установке максимально простая.
1. Скачайте Docker-образ желаемой версии: \
`docker pull vitastor:1.10.2`
`docker pull vitalif/vitastor:v3.0.8`
2. Установите скрипты в хост-систему командой: \
`docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitastor:1.10.2 install.sh`
`docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.8 install.sh`
3. Перезагрузите правила udev: \
`udevadm control --reload-rules`
4. Включите сервис vitastor-host: \
`systemctl enable --now vitastor-host`
После этого вы можете возвращаться к разделу [Быстрый старт](../intro/quickstart.ru.md).
## Podman
Если вы используете Podman, перед установкой контейнеров Vitastor выполните следующие
команды от имени суперпользователя:
```
ln -s podman /usr/bin/docker
mkdir -p /etc/systemd/system/systemd-udevd.service.d
cat >/etc/systemd/system/systemd-udevd.service.d/override.conf <<EOF
[Service]
CapabilityBoundingSet=~
SystemCallFilter=@mount capset
EOF
systemctl daemon-reload
systemctl restart systemd-udevd
```
Без этих настроек udev не может делать вызовы внутрь Podman-контейнеров и определение дисков Vitastor не работает.
## Обновление контейнеров
Сначала обязательно проверьте раздел [Обновление Vitastor](../usage/admin.ru.md#обновление-vitastor),
+1 -1
View File
@@ -30,7 +30,7 @@ volume_backend_name = vitastor-testcluster
image_volume_cache_enabled = True
volume_clear = none
vitastor_etcd_address = 192.168.7.2:2379
vitastor_etcd_prefix =
vitastor_etcd_prefix = /vitastor
vitastor_config_path = /etc/vitastor/vitastor.conf
vitastor_pool_id = 1
image_upload_use_cinder_backend = True
+1 -1
View File
@@ -29,7 +29,7 @@ volume_backend_name = vitastor-testcluster
image_volume_cache_enabled = True
volume_clear = none
vitastor_etcd_address = 192.168.7.2:2379
vitastor_etcd_prefix =
vitastor_etcd_prefix = /vitastor
vitastor_config_path = /etc/vitastor/vitastor.conf
vitastor_pool_id = 1
image_upload_use_cinder_backend = True
+13 -3
View File
@@ -11,12 +11,21 @@
- Trust Vitastor package signing key:
`wget https://vitastor.io/debian/pubkey.gpg -O /etc/apt/trusted.gpg.d/vitastor.gpg`
- Add Vitastor package repository to your /etc/apt/sources.list:
- Debian 12 (Bookworm/Sid): `deb https://vitastor.io/debian bookworm main`
- Debian 13 (Trixie/Sid): `deb https://vitastor.io/debian trixie main`
- Debian 12 (Bookworm): `deb https://vitastor.io/debian bookworm main`
- Debian 11 (Bullseye): `deb https://vitastor.io/debian bullseye main`
- Debian 10 (Buster): `deb https://vitastor.io/debian buster main`
- Ubuntu 22.04 (Jammy): `deb https://vitastor.io/debian jammy main`
- Ubuntu 24.04 (Noble): `deb https://vitastor.io/debian noble main`
- Ubuntu 26.04 (Resolute): `deb https://vitastor.io/debian resolute main`
- Add `-oldstable` to bookworm/bullseye/buster in this line to install the last
stable version from 0.9.x branch instead of 1.x
- To always prefer vitastor-patched QEMU and Libvirt versions, add the following to `/etc/apt/preferences`:
```
Package: *
Pin: origin "vitastor.io"
Pin-Priority: 501
```
- Install packages: `apt update; apt install vitastor lp-solve etcd linux-image-amd64 qemu-system-x86`
## CentOS
@@ -25,15 +34,17 @@
- CentOS 7: `yum install https://vitastor.io/rpms/centos/7/vitastor-release.rpm`
- CentOS 8: `dnf install https://vitastor.io/rpms/centos/8/vitastor-release.rpm`
- AlmaLinux 9 and other RHEL 9 clones (Rocky, Oracle...): `dnf install https://vitastor.io/rpms/centos/9/vitastor-release.rpm`
- AlmaLinux 10 and other RHEL 10 clones: `dnf install https://vitastor.io/rpms/centos/10/vitastor-release.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`
- RHEL 9 clones: not required
- RHEL 9/10 clones: not required
- 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`
- RHEL 9 clones: `dnf install https://www.elrepo.org/elrepo-release-9.el9.elrepo.noarch.rpm`
- RHEL 10 clones: not required
- Install packages: `yum/dnf install vitastor lpsolve etcd kernel-ml qemu-kvm`
## Installation requirements
@@ -42,7 +53,6 @@
recommended because io_uring is a relatively new technology and there is
at least one bug which reproduces with io_uring and HP SmartArray
controllers in 5.4
- liburing 0.4 or newer
- lp_solve
- etcd 3.4.15 or newer. Earlier versions won't work because of various bugs,
for example [#12402](https://github.com/etcd-io/etcd/pull/12402).
+13 -3
View File
@@ -11,12 +11,21 @@
- Добавьте ключ репозитория Vitastor:
`wget https://vitastor.io/debian/pubkey.gpg -O /etc/apt/trusted.gpg.d/vitastor.gpg`
- Добавьте репозиторий Vitastor в /etc/apt/sources.list:
- Debian 12 (Bookworm/Sid): `deb https://vitastor.io/debian bookworm main`
- Debian 13 (Trixie/Sid): `deb https://vitastor.io/debian trixie main`
- Debian 12 (Bookworm): `deb https://vitastor.io/debian bookworm main`
- Debian 11 (Bullseye): `deb https://vitastor.io/debian bullseye main`
- Debian 10 (Buster): `deb https://vitastor.io/debian buster main`
- Ubuntu 22.04 (Jammy): `deb https://vitastor.io/debian jammy main`
- Ubuntu 24.04 (Noble): `deb https://vitastor.io/debian noble main`
- Ubuntu 26.04 (Resolute): `deb https://vitastor.io/debian resolute main`
- Добавьте `-oldstable` к слову bookworm/bullseye/buster в этой строке, чтобы
установить последнюю стабильную версию из ветки 0.9.x вместо 1.x
- Чтобы всегда предпочитались версии пакетов QEMU и Libvirt с патчами Vitastor, добавьте в `/etc/apt/preferences`:
```
Package: *
Pin: origin "vitastor.io"
Pin-Priority: 501
```
- Установите пакеты: `apt update; apt install vitastor lp-solve etcd linux-image-amd64 qemu-system-x86`
## CentOS
@@ -25,15 +34,17 @@
- CentOS 7: `yum install https://vitastor.io/rpms/centos/7/vitastor-release.rpm`
- CentOS 8: `dnf install https://vitastor.io/rpms/centos/8/vitastor-release.rpm`
- AlmaLinux 9 и другие клоны RHEL 9 (Rocky, Oracle...): `dnf install https://vitastor.io/rpms/centos/9/vitastor-release.rpm`
- AlmaLinux 10 и другие клоны RHEL 10: `dnf install https://vitastor.io/rpms/centos/10/vitastor-release.rpm`
- Включите EPEL: `yum/dnf install epel-release`
- Включите дополнительные репозитории CentOS:
- CentOS 7: `yum install centos-release-scl`
- CentOS 8: `dnf install centos-release-advanced-virtualization`
- Клоны RHEL 9: не нужно
- Клоны RHEL 9/10: не нужно
- Включите 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`
- Клоны RHEL 9: `dnf install https://www.elrepo.org/elrepo-release-9.el9.elrepo.noarch.rpm`
- Клоны RHEL 10: не нужно
- Установите пакеты: `yum/dnf install vitastor lpsolve etcd kernel-ml qemu-kvm`
## Установочные требования
@@ -41,7 +52,6 @@
- Ядро Linux 5.4 или новее, для поддержки io_uring. Рекомендуется даже 5.8,
так как io_uring - относительно новый интерфейс и в версиях до 5.8 встречались
некоторые баги, например, зависание с io_uring и контроллером HP SmartArray
- liburing 0.4 или новее
- lp_solve
- etcd 3.4.15 или новее. Более старые версии не будут работать из-за разных багов,
например, [#12402](https://github.com/etcd-io/etcd/pull/12402).
+2 -2
View File
@@ -6,10 +6,10 @@
# Proxmox VE
To enable Vitastor support in Proxmox Virtual Environment (6.4-8.1 are supported):
To enable Vitastor support in Proxmox Virtual Environment (6.4-9.x are supported):
- Add the corresponding Vitastor Debian repository into sources.list on Proxmox hosts:
bookworm for 8.1, pve8.0 for 8.0, bullseye for 7.4, pve7.3 for 7.3, pve7.2 for 7.2, pve7.1 for 7.1, buster for 6.4
trixie for 9.0+, bookworm for 8.1+, pve8.0 for 8.0, bullseye for 7.4, pve7.3 for 7.3, pve7.2 for 7.2, pve7.1 for 7.1, buster for 6.4
- Install vitastor-client, pve-qemu-kvm, pve-storage-vitastor (* or see note) packages from Vitastor repository
- Define storage in `/etc/pve/storage.cfg` (see below)
- Block network access from VMs to Vitastor network (to OSDs and etcd),
+2 -2
View File
@@ -6,10 +6,10 @@
# Proxmox VE
Чтобы подключить Vitastor к Proxmox Virtual Environment (поддерживаются версии 6.4-8.1):
Чтобы подключить Vitastor к Proxmox Virtual Environment (поддерживаются версии 6.4-9.x):
- Добавьте соответствующий Debian-репозиторий Vitastor в sources.list на хостах Proxmox:
bookworm для 8.1, pve8.0 для 8.0, bullseye для 7.4, pve7.3 для 7.3, pve7.2 для 7.2, pve7.1 для 7.1, buster для 6.4
trixie для 9.0+, bookworm для 8.1+, pve8.0 для 8.0, bullseye для 7.4, pve7.3 для 7.3, pve7.2 для 7.2, pve7.1 для 7.1, buster для 6.4
- Установите пакеты vitastor-client, pve-qemu-kvm, pve-storage-vitastor (* или см. сноску) из репозитория Vitastor
- Определите тип хранилища в `/etc/pve/storage.cfg` (см. ниже)
- Обязательно заблокируйте доступ от виртуальных машин к сети Vitastor (OSD и etcd), т.к. Vitastor (пока) не поддерживает аутентификацию
+191
View File
@@ -0,0 +1,191 @@
[Documentation](../../README.md#documentation) → Installation → S3 for Vitastor
-----
[Читать на русском](s3.ru.md)
# S3 for Vitastor
The moment has come - Vitastor S3 implementation based on Zenko CloudServer is released.
## Highlights
- Zenko CloudServer is implemented in node.js.
- Object metadata is stored in MongoDB.
- Modified Zenko CloudServer version is used for Vitastor. It is slightly different from
the original, has an optimised build and unneeded dependencies are stripped off.
- Object data is stored in Vitastor block volumes, but the volume metadata is stored in
the same MongoDB, not in Vitastor etcd.
- Objects are written to volumes sequentially one after another. The space is allocated
with rounding to the sector size (4 KB), so each object takes at least 4 KB.
- An important property of such storage scheme is that small objects aren't chunked into
parts in Vitastor EC N+K pools and thus don't require reads from all N disks when
downloading.
- Deleted objects are marked as deleted, but the space is only actually freed during
asynchronously executed "defragmentation" process. Defragmentation runs automatically
in the background when a volume reaches configured amount of "garbage" (20% by default).
Defragmentation copies actual objects to new volume(s) and then removes the old volume.
Defragmentation can be configured in locationConfig.json.
## Plans for future development
- User account storage in the DB instead of a static file. Original Zenko uses
a separate closed-source "Scality Vault" service for it, that's why we use
a static file for now.
- More detailed documentation.
- Support for other (and faster) key-value DBMS for object metadata storage.
- Other performance optimisations, for example, related to the used hash function -
MD5 used for Amazon compatibility purposes is relatively slow.
- Object Lifecycle support. There is a Lifecycle implementation for Zenko called
[Backbeat](https://github.com/scality/backbeat) but it's not adapted for Vitastor yet.
- Quota support. Original Zenko uses a separate "SCUBA" service for quotas, but
it's also proprietary and not available publicly.
## Installation
In a few words:
- Install MongoDB, create a user for S3 metadata DB.
- Create a Vitastor pool for S3 data.
- Download and setup the Docker container `vitalif/vitastor-zenko`.
### Setup MongoDB
You can setup MongoDB yourself, following the [MongoDB manual](https://www.mongodb.com/docs/manual/installation/).
Or you can follow the instructions below - it describes a simple example of MongoDB setup
in Docker (through docker-compose) with 3 replicas.
1. On each host, create a file `docker-compose.yml` with the content listed below.
Replace `<YOUR_PASSWORD>` with your future mongodb administrator password, and optionally
replace `0.0.0.0` with `localhost,<server_IP>`. It's recommended to either use a private IP
or [setup TLS](https://www.mongodb.com/docs/manual/tutorial/configure-ssl/) afterwards.
```
version: '3.1'
services:
mongo:
container_name: mongo
image: mongo:7-jammy
restart: always
environment:
MONGO_INITDB_ROOT_USERNAME: root
MONGO_INITDB_ROOT_PASSWORD: <YOUR_PASSWORD>
network_mode: host
volumes:
- ./keyfile:/opt/keyfile
- ./mongo-data/db:/data/db
- ./mongo-data/configdb:/data/configdb
entrypoint: /bin/bash -c
command: [ "chown mongodb /opt/keyfile && chmod 600 /opt/keyfile && . /usr/local/bin/docker-entrypoint.sh mongod --replSet rs0 --keyFile /opt/keyfile --bind_ip 0.0.0.0" ]
```
2. Generate a shared cluster key using `openssl rand -base64 756 > ./keyfile` and copy
that `keyfile` to all hosts.
3. Start MongoDB on all hosts with `docker compose up -d mongo`.
4. Enter Mongo Shell with `docker exec -it mongo mongosh -u root -p <YOUR_PASSWORD> localhost/admin`
and execute the following command (replace IP addresses `10.10.10.{1,2,3}` with your host IPs):
`rs.initiate({ _id: 'rs0', members: [
{ _id: 1, host: '10.10.10.1:27017' },
{ _id: 2, host: '10.10.10.2:27017' },
{ _id: 3, host: '10.10.10.3:27017' }
] })`
5. Stay in Mongo Shell and create a user for the future S3 database:
`db.createUser({ user: 's3', pwd: '<YOUR_S3_PASSWORD>', roles: [
{ role: 'readWrite', db: 's3' },
{ role: 'dbAdmin', db: 's3' },
{ role: 'readWrite', db: 'vitastor' },
{ role: 'dbAdmin', db: 'vitastor' }
] })`
### Setup Vitastor
Create a pool in Vitastor for S3 object data, for example:
`vitastor-cli create-pool --ec 2+1 -n 512 s3-data --used_for_app s3:standard`
The `--used_for_app` options works as fool-proofing and prevents you from
accidentally creating a regular block volume in the S3 pool and overwriting some S3 data.
Also it hides inode space statistics from Vitastor etcd.
Retrieve the ID of your pool with `vitastor-cli ls-pools s3-data --detail`.
### Setup Vitastor S3
1. Add the following lines to `docker-compose.yml` (instead of `network_mode: host`,
you can use `ports: [ "8000:8000", "8002:8002" ]`):
```
zenko:
container_name: zenko
image: vitalif/vitastor-zenko
restart: always
security_opt:
- seccomp:unconfined
ulimits:
memlock: -1
network_mode: host
volumes:
- /etc/vitastor:/etc/vitastor
- /etc/vitastor/s3:/conf
```
2. Download Docker image: `docker pull vitalif/vitastor-zenko`
3. Extract configuration file examples from the Docker image:
```
docker run --rm -it -v /etc/vitastor:/etc/vitastor -v /etc/vitastor/s3:/conf vitalif/vitastor-zenko configure.sh
```
4. Edit configuration files in `/etc/vitastor/s3/`:
- `config.json` - common settings.
- `authdata.json` - user accounts and access keys.
- `locationConfig.json` - S3 storage class list with placement settings.
Note: it actually contains storage classes (like STANDARD, COLD, etc)
instead of "locations" (zones like us-east-1) as in the original Zenko CloudServer.
- Put your MongoDB connection data into `config.json` and `locationConfig.json`.
- Put your Vitastor pool ID into `locationConfig.json`.
- For now, the complete list of Vitastor backend settings is only available [in the code](https://git.yourcmc.ru/vitalif/zenko-arsenal/src/branch/master/lib/storage/data/vitastor/VitastorBackend.ts#L94).
### Start Zenko
Start the S3 server with:
```
docker run --restart always --security-opt seccomp:unconfined --ulimit memlock=-1 --network=host \
-v /etc/vitastor:/etc/vitastor -v /etc/vitastor/s3:/conf --name zenko vitalif/vitastor-zenko
```
If you use default settings, Zenko CloudServer starts on port 8000.
The default access key is `accessKey1` with a secret key of `verySecretKey1`.
Now you can access your S3 with, for example, [s3cmd](https://s3tools.org/s3cmd):
```
s3cmd --access_key=accessKey1 --secret_key=verySecretKey1 --host=http://localhost:8000 mb s3://testbucket
```
Or even mount it with [GeeseFS](https://github.com/yandex-cloud/geesefs):
```
AWS_ACCESS_KEY_ID=accessKey1 \
AWS_SECRET_ACCESS_KEY=verySecretKey1 \
geesefs --endpoint http://localhost:8000 testbucket mountdir
```
## Author & License
- [Zenko CloudServer](https://s3-server.readthedocs.io/en/latest/) author is Scality,
licensed under [Apache License, version 2.0](https://www.apache.org/licenses/LICENSE-2.0)
- [Vitastor](https://git.yourcmc.ru/vitalif/vitastor/) and Zenko Vitastor backend author is
Vitaliy Filippov, licensed under [VNPL-1.1](https://git.yourcmc.ru/vitalif/vitastor/src/branch/master/VNPL-1.1.txt)
(a "network copyleft" license based on AGPL/SSPL, but worded in a better way)
- Vitastor S3 repository: https://git.yourcmc.ru/vitalif/zenko-cloudserver-vitastor
- Vitastor S3 backend code: https://git.yourcmc.ru/vitalif/zenko-arsenal/src/branch/master/lib/storage/data/vitastor/VitastorBackend.ts
+171
View File
@@ -0,0 +1,171 @@
[Документация](../../README-ru.md#документация) → Установка → S3 на базе Vitastor
-----
[Read in English](s3.en.md)
# S3 на базе Vitastor
Итак, свершилось - реализация Vitastor S3 на базе Zenko CloudServer достигла
состояния готовности к публикации и использованию.
## Ключевые особенности
- Zenko CloudServer реализован на node.js.
- Метаданные объектов хранятся в MongoDB.
- Поставляется модифицированная версия Zenko CloudServer, отвязанная от лишних зависимостей,
с оптимизированной сборкой и немного отличающаяся от оригинала.
- Данные объектов хранятся в блочных томах Vitastor, однако информация о самих томах
сохраняется не в etcd Vitastor, а тоже в БД на основе MongoDB.
- Объекты записываются в тома последовательно друг за другом. Место выделяется с округлением
до размера сектора (до 4 килобайт), поэтому каждый объект занимает как минимум 4 КБ.
- Благодаря такой схеме записи объектов мелкие объекты не нарезаются на части и поэтому не
требуют чтения с N дисков данных в EC N+K пулах Vitastor.
- При удалении объекты помечаются удалёнными, но место освобождается не сразу, а при
запускаемой асинхронно "дефрагментации". Дефрагментация запускается автоматически в фоне
при достижении заданного объёма "мусора" в томе (по умолчанию 20%), копирует актуальные
объекты в новые тома, после чего очищает старый том полностью. Дефрагментацию можно
настраивать в locationConfig.json.
## Планы развития
- Хранение учётных записей в БД, а не в статическом файле (в оригинальном Zenko для
этого используется отдельный закрытый сервис "Scality Vault").
- Более подробная документация.
- Поддержка других (и более производительных) key-value СУБД для хранения метаданных.
- Другие оптимизации производительности, например, в области используемой хеш-функции
(хеш MD5, используемый в целях совместимости, относительно медленный).
- Поддержка Object Lifecycle. Реализация Lifecycle для Zenko существует и называется
[Backbeat](https://github.com/scality/backbeat), но она ещё не адаптирована для Vitastor.
- Квоты. В оригинальном Zenko для этого используется отдельный сервис "SCUBA", однако
он тоже является закрытым и недоступен для публичного использования.
## Установка
Кратко:
- Установите MongoDB, создайте пользователя для БД метаданных S3.
- Создайте в Vitastor пул для хранения данных объектов.
- Скачайте и настройте Docker-контейнер `vitalif/vitastor-zenko`.
### Установка MongoDB
Вы можете установить MongoDB сами, следуя [официальному руководству MongoDB](https://www.mongodb.com/docs/manual/installation/).
Либо вы можете последовать инструкции, приведённой ниже - здесь описан простейший пример
установки MongoDB в Docker (docker-compose) в конфигурации с 3 репликами.
1. На всех 3 серверах создайте файл `docker-compose.yml`, заменив `<ВАШ_ПАРОЛЬ>`
на собственный будущий пароль администратора mongodb, а `0.0.0.0` по желанию
заменив на на `localhost,<IP_сервера>` - желательно либо использовать публично не доступный IP,
либо потом [настроить TLS](https://www.mongodb.com/docs/manual/tutorial/configure-ssl/).
```
version: '3.1'
services:
mongo:
container_name: mongo
image: mongo:7-jammy
restart: always
environment:
MONGO_INITDB_ROOT_USERNAME: root
MONGO_INITDB_ROOT_PASSWORD: <ВАШ_ПАРОЛЬ>
network_mode: host
volumes:
- ./keyfile:/opt/keyfile
- ./mongo-data/db:/data/db
- ./mongo-data/configdb:/data/configdb
entrypoint: /bin/bash -c
command: [ "chown mongodb /opt/keyfile && chmod 600 /opt/keyfile && . /usr/local/bin/docker-entrypoint.sh mongod --replSet rs0 --keyFile /opt/keyfile --bind_ip 0.0.0.0" ]
```
2. В той же директории сгенерируйте общий ключ кластера командой `openssl rand -base64 756 > ./keyfile`
и скопируйте этот файл на все 3 сервера.
3. На всех 3 серверах запустите MongoDB командой `docker compose up -d mongo`.
4. Зайдите в Mongo Shell с помощью команды `docker exec -it mongo mongosh -u root -p <ВАШ_ПАРОЛЬ> localhost/admin`
и там выполните команду (заменив IP-адреса `10.10.10.{1,2,3}` на адреса своих серверов):
`rs.initiate({ _id: 'rs0', members: [
{ _id: 1, host: '10.10.10.1:27017' },
{ _id: 2, host: '10.10.10.2:27017' },
{ _id: 3, host: '10.10.10.3:27017' }
] })`
5. Находясь там же, в Mongo Shell, создайте пользователя с доступом к будущей базе данных S3:
`db.createUser({ user: 's3', pwd: '<ВАШ_ПАРОЛЬ_S3>', roles: [
{ role: 'readWrite', db: 's3' },
{ role: 'dbAdmin', db: 's3' },
{ role: 'readWrite', db: 'vitastor' },
{ role: 'dbAdmin', db: 'vitastor' }
] })`
### Настройка Vitastor
Создайте в Vitastor отдельный пул для данных объектов S3, например:
`vitastor-cli create-pool --ec 2+1 -n 512 s3-data --used_for_app s3:standard`
Опция `--used_for_app` работает как "защита от дурака" и не даёт вам случайно создать
в этом пуле обычный блочный том и перезаписать им какие-то данные S3, а также скрывает
статистику занятого места по томам S3 из etcd.
Получите ID своего пула с помощью команды `vitastor-cli ls-pools --detail`.
### Установка Vitastor S3
1. Добавьте в `docker-compose.yml` строки (альтернативно вместо `network_mode: host`
можно использовать `ports: [ "8000:8000", "8002:8002" ]`):
```
zenko:
container_name: zenko
image: vitalif/vitastor-zenko
restart: always
security_opt:
- seccomp:unconfined
ulimits:
memlock: -1
network_mode: host
volumes:
- /etc/vitastor:/etc/vitastor
- /etc/vitastor/s3:/conf
```
2. Извлеките из Docker-образа Vitastor примеры файлов конфигурации:
`docker run --rm -it -v /etc/vitastor:/etc/vitastor -v /etc/vitastor/s3:/conf vitalif/vitastor-zenko configure.sh`
3. Отредактируйте файлы конфигурации в `/etc/vitastor/s3/`:
- `config.json` - общие настройки.
- `authdata.json` - учётные записи и ключи доступа.
- `locationConfig.json` - список классов хранения S3 с настройками расположения.
Внимание: в данной версии это именно список S3 storage class-ов (STANDARD, COLD и т.п.),
а не зон (подобных us-east-1), как в оригинальном Zenko CloudServer.
- В `config.json` и в `locationConfig.json` пропишите свои данные подключения к MongoDB.
- В `locationConfig.json` укажите ID пула Vitastor для хранения данных.
- Полный перечень настроек Vitastor-бэкенда пока можно посмотреть [в коде](https://git.yourcmc.ru/vitalif/zenko-arsenal/src/branch/master/lib/storage/data/vitastor/VitastorBackend.ts#L94).
### Запуск
Запустите S3-сервер: `docker-compose up -d zenko`
Готово! Вы получили S3-сервер, работающий на порту 8000.
Можете попробовать обратиться к нему с помощью, например, [s3cmd](https://s3tools.org/s3cmd):
`s3cmd --host-bucket= --no-ssl --access_key=accessKey1 --secret_key=verySecretKey1 --host=http://localhost:8000 mb s3://testbucket`
Или смонтировать его с помощью [GeeseFS](https://github.com/yandex-cloud/geesefs):
`AWS_ACCESS_KEY_ID=accessKey1 AWS_SECRET_ACCESS_KEY=verySecretKey1 geesefs --endpoint http://localhost:8000 testbucket /mnt/geesefs`
## Лицензия
- Автор [Zenko CloudServer](https://s3-server.readthedocs.io/en/latest/) - Scality, лицензия [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0)
- Vitastor-бэкенд для S3, как и сам Vitastor, лицензируется на условиях [VNPL 1.1](https://git.yourcmc.ru/vitalif/vitastor/src/branch/master/VNPL-1.1.txt)
- Репозиторий сборки: https://git.yourcmc.ru/vitalif/zenko-cloudserver-vitastor
- Бэкенд хранения данных: https://git.yourcmc.ru/vitalif/zenko-arsenal/src/branch/master/lib/storage/data/vitastor/VitastorBackend.ts
+2 -2
View File
@@ -15,8 +15,8 @@
- gcc and g++ 8 or newer, clang 10 or newer, or other compiler with C++11 plus
designated initializers support from C++20
- CMake
- liburing, jerasure headers and libraries
- ISA-L, libibverbs headers and libraries (optional)
- jerasure headers and libraries
- ISA-L, libibverbs and librdmacm headers and libraries (optional)
- tcmalloc (google-perftools-dev)
## Basic instructions
+2 -2
View File
@@ -15,8 +15,8 @@
- gcc и g++ >= 8, либо clang >= 10, либо другой компилятор с поддержкой C++11 плюс
назначенных инициализаторов (designated initializers) из C++20
- CMake
- Заголовки и библиотеки liburing, jerasure
- Опционально - заголовки и библиотеки ISA-L, libibverbs
- Заголовки и библиотеки jerasure
- Опционально - заголовки и библиотеки ISA-L, libibverbs, librdmacm
- tcmalloc (google-perftools-dev)
## Базовая инструкция
+7
View File
@@ -125,6 +125,13 @@ All other client-side components are based on the client library:
all current read/write operations to it fail with EPIPE error and are retried by clients.
- After completing all secondary read/write requests, primary OSD sends the response to
the client.
- When [localized reads](../config/pool.en.md#local_reads) are enabled for a PG in a
replicated pool, and the PG is in an active and clean state (active or
active+left_on_dead), the client can send the request to one of secondary OSDs instead
of the primary. Secondary OSD checks the [PG lock](../config/osd.en.md#enable_pg_locks)
and handles the request locally without communicating to the primary. PG lock is required
for the secondary OSD to know for sure that the PG is in clean state and not switching
primary at the moment.
### Nuances of request handling
+6
View File
@@ -125,6 +125,12 @@
и если любое из этих соединений отключается, PG перезапускается, а все текущие запросы чтения
и записи в неё завершаются с ошибкой EPIPE, после чего повторяются клиентами.
- После завершения всех вторичных операций чтения/записи первичный OSD отправляет ответ клиенту.
- Если в реплицированном пуле включены [локализованные чтения](../config/pool.ru.md#local_reads),
а PG находится в чистом активном состоянии (active или active+left_on_dead), клиент может
послать запрос к одному из вторичных OSD вместо первичного. Вторичный OSD проверяет
[блокировку PG](../config/osd.ru.md#enable_pg_locks) и обрабатывает запрос локально, не
обращаясь к первичному. Блокировка PG здесь нужна, чтобы вторичный OSD мог точно знать,
что PG находится в чистом состоянии и не переключается на другой первичный OSD.
### Особенности обработки запросов
+84 -10
View File
@@ -10,8 +10,17 @@ 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
License: VNPL 1.1 for server-side code and dual VNPL 1.1 + GPL 2.0+ for client tools.
Server-side code is licensed only under the terms of VNPL.
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.
## VNPL
Vitastor Network Public License 1.1 (VNPL 1.1) is 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
@@ -20,18 +29,83 @@ 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.
The idea of VNPL is, in addition to modules linked to Vitastor code in a single
binary file, to extend copyleft action to micro-service modules only interacting
with it over the network.
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.
At the same time, VNPL doesn't impose any restrictions on software *not specially designed*
to be used with Vitastor, for example, on Windows running inside a VM with a Vitastor disk.
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).
## Explanation
Network copyleft is governed by the clause **13. Remote Network Interaction** of VNPL.
A program is considered to be a "Proxy Program" if it meets both conditions:
- It is specially designed to be used with Vitastor. Basically, it means that the program
has any functionality specific to Vitastor and thus "knows" that it works with Vitastor,
not with something random.
- It interacts with Vitastor directly or indirectly through any programming interface,
including API, CLI, network or any wrapper (also considered a Proxy Program itself).
If, in addition to that:
- You give any user an apportunity to interact with Vitastor directly or indirectly through
any computer interface including the network or any number of wrappers (Proxy Programs).
Then VNPL requires you to publish the code of all above Proxy Programs to all above users
under the terms of any GPL-compatible license - that is, GPL, LGPL, MIT/BSD or Apache 2,
because "GPL compatibility" is treated as an ability to legally include licensed code in
a GPL application.
So, if you have a "Proxy Program", but it's not open to the user who directly or indirectly
interacts with Vitastor - you are forbidden to use Vitastor under the terms of VNPL and you
need a commercial license which doesn't contain open-source requirements.
## Examples
- Vitastor Kubernetes CSI driver which creates PersistentVolumes by calling `vitastor-cli create`.
- Yes, it interacts with Vitastor through vitastor-cli.
- Yes, it is designed specially for use with Vitastor (it has no sense otherwise).
- So, CSI driver **definitely IS** a Proxy Program and must be published under the terms of
a free software license.
- Windows, installed in a VM with the system disk on Vitastor storage.
- Yes, it interacts with Vitastor indirectly - it reads and writes data through the block
device interface, emulated by QEMU.
- No, it definitely isn't designed specially for use with Vitastor - Windows was created long
ago before Vitastor and doesn't know anything about it.
- So, Windows **definitely IS NOT** a Proxy Program and VNPL doesn't require to open it.
- Cloud control panel which makes requests to Vitastor Kubernetes CSI driver.
- Yes, it interacts with Vitastor indirectly through the CSI driver, which is a Proxy Program.
- May or may not be designed specially for use with Vitastor. How to determine exactly?
Imagine that Vitastor is replaced with any other storage (for example, with a proprietary).
Do control panel functions change in any way? If they do (for example, if snapshots stop working),
then the panel contains specific functionality and thus is designed specially for use with Vitastor.
Otherwise, the panel is universal and isn't designed specially for Vitastor.
- So, whether you are required to open-source the panel also **depends** on whether it
contains specific functionality or not.
## Why?
Because I believe into the spirit of copyleft (Linux wouldn't become so popular without GPL!)
and, at the same time, I want to have a way to monetize the product.
Existing licenses including AGPL are useless for it with an SDS - SDS is a very deeply
internal software which is almost definitely invisible to the user and thus AGPL doesn't
require anyone to open the code even if they make a proprietary fork.
And, in fact, the current situation in the world where GPL is though to only restrict direct
linking of programs into a single executable file, isn't much correct. Nowadays, programs
are more often linked with network API calls, not with /usr/bin/ld, and a software product
may consist of dozens of microservices interacting with each other over the network.
That's why we need VNPL to keep the license sufficiently copyleft.
## License Texts
- VNPL 1.1 in English: [VNPL-1.1.txt](../../VNPL-1.1.txt)
- VNPL 1.1 in Russian: [VNPL-1.1-RU.txt](../../VNPL-1.1-RU.txt)
- GPL 2.0: [GPL-2.0.txt](../../GPL-2.0.txt)
+74 -5
View File
@@ -12,6 +12,14 @@
Лицензия: VNPL 1.1 на серверный код и двойная VNPL 1.1 + GPL 2.0+ на клиентский.
Серверные компоненты распространяются только на условиях VNPL.
Клиентские библиотеки распространяются на условиях двойной лицензии VNPL 1.0
и также на условиях GNU GPL 2.0 или более поздней версии. Так сделано в целях
совместимости с таким ПО, как QEMU и fio.
## VNPL
VNPL - "сетевой копилефт", собственная свободная копилефт-лицензия
Vitastor Network Public License 1.1, основанная на GNU GPL 3.0 с дополнительным
условием "Сетевого взаимодействия", требующим распространять все программы,
@@ -29,9 +37,70 @@ Vitastor Network Public License 1.1, основанная на GNU GPL 3.0 с д
На Windows и любое другое ПО, не разработанное *специально* для использования
вместе с Vitastor, никакие ограничения не накладываются.
Клиентские библиотеки распространяются на условиях двойной лицензии VNPL 1.0
и также на условиях GNU GPL 2.0 или более поздней версии. Так сделано в целях
совместимости с таким ПО, как QEMU и fio.
## Пояснение
Вы можете найти полный текст VNPL 1.1 на английском языке в файле [VNPL-1.1.txt](../../VNPL-1.1.txt),
VNPL 1.1 на русском языке в файле [VNPL-1.1-RU.txt](../../VNPL-1.1-RU.txt), а GPL 2.0 в файле [GPL-2.0.txt](../../GPL-2.0.txt).
Сетевой копилефт регулируется пунктом лицензии **13. Удалённое сетевое взаимодействие**.
Программа считается "прокси-программой", если верны оба условия:
- Она создана специально для работы вместе с Vitastor. По сути это означает, что программа
должна иметь специфичный для Vitastor функционал, то есть, "знать", что она взаимодействует
именно с Vitastor.
- Она прямо или косвенно взаимодействует с Vitastor через абсолютно любой программный
интерфейс, включая любые способы вызова: API, CLI, сеть или через какую-то обёртку (в
свою очередь тоже являющуюся прокси-программой).
Если в дополнение к этому также:
- Вы предоставляете любому пользователю возможность взаимодействовать с Vitastor по сети,
опять-таки, через любой интерфейс или любую серию "обёрток" (прокси-программ)
То, согласно VNPL, вы должны открыть код "прокси-программ" **таким пользователям** на условиях
любой GPL-совместимой лицензии - то есть, GPL, LGPL, MIT/BSD или Apache 2 - "совместимость с GPL"
понимается как возможность включать лицензируемый код в GPL-приложение.
Соответственно, если у вас есть "прокси-программа", но её код не открыт пользователю,
который прямо или косвенно взаимодействует с Vitastor - вам запрещено использовать Vitastor
на условиях VNPL и вам нужна коммерческая лицензия, не содержащая требований об открытии кода.
## Примеры
- Kubernetes CSI-драйвер Vitastor, создающий PersistentVolume с помощью вызова `vitastor-cli create`.
- Да, взаимодействует с Vitastor через vitastor-cli.
- Да, создавался специально для работы с Vitastor (иначе в чём же ещё его смысл).
- Значит, CSI-драйвер **точно считается** "прокси-программой" и должен быть открыт под свободной
лицензией.
- Windows, установленный в виртуальную машину на диске Vitastor.
- Да, взаимодействует с Vitastor "прямо или косвенно" - пишет и читает данные через интерфейс
блочного устройства, эмулируемый QEMU.
- Нет, точно не создан *специально для работы с Vitastor* - когда его создавали, никакого
Vitastor ещё и в помине не было.
- Значит, Windows **точно не считается** "прокси-программой" и на него требования VNPL не распространяются.
- Панель управления облака, делающая запросы к Kubernetes CSI-драйверу Vitastor.
- Да, взаимодействует с Vitastor косвенно через CSI-драйвер, являющийся "прокси-программой".
- Сходу не известно, создавалась ли конкретно для работы с Vitastor. Как понять, да или нет?
Представьте, что Vitastor заменён на любую другую систему хранения (например, на проприетарную).
Работа панели управления изменится? Если да (например, перестанут работать снапшоты) - значит,
панель содержит специфичный функционал и "создана специально для работы с Vitastor".
Если нет - значит, специфичного функционала панель не содержит и в принципе она универсальна.
- Нужно ли открывать панель - **зависит** от того, содержит она специфичный функционал или нет.
## Почему так?
Потому что я одновременно верю в дух копилефт-лицензий (Linux не стал бы так популярен,
если бы не GPL!) и хочу иметь возможность монетизации продукта.
При этом использовать даже AGPL для программной СХД бессмысленно - это глубоко внутреннее
ПО, которое пользователь почти наверняка не увидит вообще, поэтому и открывать код никому
никогда не придётся, даже при создании производного продукта.
Да и в целом сложившаяся в мире ситуация, при которой действие GPL ограничивается только
прямым связыванием в один исполняемый файл, не очень корректна. В настоящее время программы
гораздо чаще интегрируют сетевыми вызовами, а не с помощью /usr/bin/ld, и общий программный
продукт может состоять из нескольких десятков микросервисов, взаимодействующих по сети.
Поэтому для сохранения достаточной "копилефтности" и придумана VNPL.
## Тексты лицензий
- VNPL 1.1 на английском языке: [VNPL-1.1.txt](../../VNPL-1.1.txt)
- VNPL 1.1 на русском языке: [VNPL-1.1-RU.txt](../../VNPL-1.1-RU.txt)
- GPL 2.0: [GPL-2.0.txt](../../GPL-2.0.txt)
+6 -3
View File
@@ -14,6 +14,8 @@
- Basic part: highly-available block storage with symmetric clustering and no SPOF
- [Performance](../performance/bench2.en.md) ;-D
- [NVMe atomic write support](../config/osd.en.md#atomic_write_size) for reducing the amount
of "extra" disk writes to almost zero (Write Amplification = 1)
- [Multiple redundancy schemes](../config/pool.en.md#scheme): Replication, XOR n+1, Reed-Solomon erasure codes
based on jerasure and ISA-L libraries with any number of data and parity drives in a group
- Configuration via simple JSON data structures in etcd (parameters, pools and images)
@@ -25,10 +27,11 @@
- Recovery of degraded blocks
- Rebalancing (data movement between OSDs)
- [Lazy fsync support](../config/layout-cluster.en.md#immediate_commit)
- [Localized read support](../config/pool.en.md#local_reads) for cross-datacenter setup optimization
- Per-OSD and per-image I/O and space usage statistics in etcd
- Snapshots and copy-on-write image clones
- [Write throttling to smooth random write workloads in SSD+HDD configurations](../config/osd.en.md#throttle_small_writes)
- [RDMA/RoCEv2 support via libibverbs](../config/network.en.md#rdma_device)
- RDMA/RoCEv2 support [via libibverbs](../config/network.en.md#use_rdma) or [RDMA-CM](../config/network.en.md#use_rdmacm)
- [Scrubbing](../config/osd.en.md#auto_scrub) (verification of copies)
- [Checksums](../config/layout-osd.en.md#data_csum_type)
- [Client write-back cache](../config/client.en.md#client_enable_writeback)
@@ -37,6 +40,7 @@
- [Experimental internal etcd replacement - antietcd](../config/monitor.en.md#use_antietcd)
- [Built-in Prometheus metric exporter](../config/monitor.en.md#enable_prometheus)
- [NFS RDMA support](../usage/nfs.en.md#rdma) (probably also usable for GPUDirect)
- [S3](../installation/s3.en.md)
## Plugins and tools
@@ -50,7 +54,7 @@
- Generic user-space client library
- [Native QEMU driver](../usage/qemu.en.md)
- [Loadable fio engine for benchmarks](../usage/fio.en.md)
- [NBD proxy for kernel mounts](../usage/nbd.en.md)
- [UBLK](../usage/ublk.en.md) and [NBD](../usage/nbd.en.md) servers for kernel mounts
- [Simplified NFS proxy for file-based image access emulation (suitable for VMWare)](../usage/nfs.en.md#pseudo-fs)
## Roadmap
@@ -63,7 +67,6 @@ The following features are planned for the future:
- iSCSI and NVMeoF gateways
- Multi-threaded client
- Faster failover
- S3
- Tiered storage (SSD caching)
- NVDIMM support
- Compression (possibly)
+6 -3
View File
@@ -14,6 +14,8 @@
- Базовая часть - надёжное кластерное блочное хранилище без единой точки отказа
- [Производительность](../performance/bench2.ru.md) ;-D
- [Поддержка атомарной записи NVMe](../config/osd.ru.md#atomic_write_size) для снижения объёма
служебной записи практически до нуля (Write Amplification = 1)
- [Несколько схем отказоустойчивости](../config/pool.ru.md#scheme): репликация, XOR n+1 (1 диск чётности), коды коррекции ошибок
Рида-Соломона на основе библиотек jerasure и ISA-L с любым числом дисков данных и чётности в группе
- Конфигурация через простые человекочитаемые JSON-структуры в etcd
@@ -25,12 +27,13 @@
- Восстановление деградированных блоков
- Ребаланс, то есть перемещение данных между OSD (дисками)
- [Поддержка "ленивого" fsync (fsync не на каждую операцию)](../config/layout-cluster.ru.md#immediate_commit)
- [Локальные чтения](../config/pool.ru.md#local_reads) для оптимизации при нескольких датацентрах
- Сбор статистики ввода/вывода в etcd
- Статистика операций ввода/вывода и занятого места в разрезе инодов
- Именование инодов через хранение их метаданных в etcd
- Снапшоты и copy-on-write клоны
- [Сглаживание производительности случайной записи в SSD+HDD конфигурациях](../config/osd.ru.md#throttle_small_writes)
- [Поддержка RDMA/RoCEv2 через libibverbs](../config/network.ru.md#rdma_device)
- Поддержка RDMA/RoCEv2 [через libibverbs](../config/network.ru.md#use_rdma) или [RDMA-CM](../config/network.ru.md#use_rdmacm)
- [Фоновая проверка целостности](../config/osd.ru.md#auto_scrub) (сверка копий)
- [Контрольные суммы](../config/layout-osd.ru.md#data_csum_type)
- [Буферизация записи на стороне клиента](../config/client.ru.md#client_enable_writeback)
@@ -39,6 +42,7 @@
- [Экспериментальная встроенная замена etcd - antietcd](../config/monitor.ru.md#use_antietcd)
- [Встроенный Prometheus-экспортер метрик](../config/monitor.ru.md#enable_prometheus)
- [Поддержка NFS RDMA](../usage/nfs.ru.md#rdma) (вероятно, также подходящая для GPUDirect)
- [S3](../installation/s3.ru.md)
## Драйверы и инструменты
@@ -52,7 +56,7 @@
- Общая пользовательская клиентская библиотека для работы с кластером
- [Драйвер диска для QEMU](../usage/qemu.ru.md)
- [Драйвер диска для утилиты тестирования производительности fio](../usage/fio.ru.md)
- [NBD-прокси для монтирования образов ядром](../usage/nbd.ru.md) ("блочное устройство в режиме пользователя")
- [UBLK](../usage/ublk.ru.md) и [NBD](../usage/nbd.ru.md) серверы для монтирования образов ядром ("блочное устройство в режиме пользователя")
- [Упрощённая NFS-прокси для эмуляции файлового доступа к образам (подходит для VMWare)](../usage/nfs.ru.md#псевдо-фс)
## Планы развития
@@ -63,7 +67,6 @@
- iSCSI и NVMeoF прокси
- Многопоточный клиент
- Более быстрое переключение при отказах
- S3
- Поддержка SSD-кэширования (tiered storage)
- Поддержка NVDIMM
- Возможно, сжатие
+10 -7
View File
@@ -18,9 +18,10 @@
## Preparation
- Get some SATA or NVMe SSDs with capacitors (server-grade drives). You can use desktop SSDs
with lazy fsync, but prepare for inferior single-thread latency. Read more about capacitors
[here](../config/layout-cluster.en.md#immediate_commit).
- Get some SATA or NVMe SSDs with capacitors (server-grade drives). The best performance
is achieved with Micron or Kioxia NVMes with atomic write support (see below). You can use desktop
SSDs with lazy fsync, but prepare for inferior single-thread latency. Read more about
capacitors [here](../config/layout-cluster.en.md#immediate_commit).
- If you want to use HDDs, get modern HDDs with Media Cache or SSD Cache: HGST Ultrastar,
Toshiba MG, Seagate EXOS or something similar. If your drives don't have such cache then
you also need small SSDs for journal and metadata (even 2 GB per 1 TB of HDD space is enough).
@@ -30,9 +31,11 @@
## Recommended drives
- SATA SSD: Micron 5100/5200/5300/5400, Samsung PM863/PM883/PM893, Intel D3-S4510/4520/4610/4620, Kingston DC500M
- NVMe: Micron 9100/9200/9300/9400, Micron 7300/7450, Samsung PM983/PM9A3, Samsung PM1723/1735/1743,
Intel DC-P3700/P4500/P4600, Intel D5-P4320/P5530, Intel D7-P5500/P5600, Intel Optane, Kingston DC1000B/DC1500M
- NVMe with atomic write support (ideal!): Micron 7450/7500/7550, Kioxia CD6/CD7/CD8/CD9
- Other NVMe: Micron 9100/9200/9300/9400/9550, Micron 7300, Samsung PM983/PM9A3, Samsung PM1723/1735/1743,
Intel DC-P3700/P4500/P4600, Intel/Solidigm D5-P4320/P5530, Intel/Solidigm D7-P5500/P5600, Solidigm D7-PS1010/PS1030/P5810,
Intel Optane, Kingston DC1000B/DC1500M, Kioxia CD6/CD7/CD8/CD9
- SATA SSD: Micron 5100/5200/5300/5400, Samsung PM863/PM883/PM893, Intel/Solidigm D3-S4510/4520/4610/4620, Kingston DC500M
- HDD: HGST Ultrastar, Toshiba MG, Seagate EXOS
## Configure monitors
@@ -50,7 +53,7 @@ On the monitor hosts:
## Configure OSDs
- Put etcd_address and osd_network into `/etc/vitastor/vitastor.conf`. Example:
- Put etcd_address and [osd_network](../config/network.en.md#osd_network) into `/etc/vitastor/vitastor.conf`. Example:
```
{
"etcd_address": ["10.200.1.10:2379","10.200.1.11:2379","10.200.1.12:2379"],
+9 -6
View File
@@ -18,8 +18,9 @@
## Подготовка
- Возьмите серверы с SSD (SATA или NVMe), желательно с конденсаторами (серверные SSD). Можно
использовать и десктопные SSD, включив режим отложенного fsync, но производительность будет хуже.
- Возьмите серверы с SSD (SATA или NVMe), желательно с конденсаторами (серверные SSD). Наилучшая
производительность достигается на дисках Micron и Kioxia с поддержкой атомарной записи (см. ниже).
Можно использовать и десктопные SSD, включив режим отложенного fsync, но производительность будет хуже.
О конденсаторах читайте [здесь](../config/layout-cluster.ru.md#immediate_commit).
- Если хотите использовать HDD, берите современные модели с Media или SSD кэшем - HGST Ultrastar,
Toshiba MG, Seagate EXOS или что-то похожее. Если такого кэша у ваших дисков нет,
@@ -30,9 +31,11 @@
## Рекомендуемые диски
- SATA SSD: Micron 5100/5200/5300/5400, Samsung PM863/PM883/PM893, Intel D3-S4510/4520/4610/4620, Kingston DC500M
- NVMe: Micron 9100/9200/9300/9400, Micron 7300/7450, Samsung PM983/PM9A3, Samsung PM1723/1735/1743,
Intel DC-P3700/P4500/P4600, Intel D5-P4320/P5530, Intel D7-P5500/P5600, Intel Optane, Kingston DC1000B/DC1500M
- NVMe с поддержкой атомарной записи (идеально!): Micron 7450/7500/7550, Kioxia CD6/CD7/CD8/CD9
- Другие NVMe: Micron 9100/9200/9300/9400/9550, Micron 7300, Samsung PM983/PM9A3, Samsung PM1723/1735/1743,
Intel DC-P3700/P4500/P4600, Intel/Solidigm D5-P4320/P5530, Intel/Solidigm D7-P5500/P5600, Solidigm D7-PS1010/PS1030/P5810,
Intel Optane, Kingston DC1000B/DC1500M, Kioxia CD6/CD7/CD8/CD9
- SATA SSD: Micron 5100/5200/5300/5400, Samsung PM863/PM883/PM893, Intel/Solidigm D3-S4510/4520/4610/4620, Kingston DC500M
- HDD: HGST Ultrastar, Toshiba MG, Seagate EXOS
## Настройте мониторы
@@ -50,7 +53,7 @@
## Настройте OSD
- Пропишите etcd_address и osd_network в `/etc/vitastor/vitastor.conf`. Например:
- Пропишите etcd_address и [osd_network](../config/network.ru.md#osd_network) в `/etc/vitastor/vitastor.conf`. Например:
```
{
"etcd_address": ["10.200.1.10:2379","10.200.1.11:2379","10.200.1.12:2379"],
+28 -20
View File
@@ -14,7 +14,7 @@ Replicated setups:
- Linear read: `min(total network bandwidth, sum(disk read MB/s))`.
- Linear write: `min(total network bandwidth, sum(disk write MB/s / number of replicas))`.
- Saturated parallel read iops: `min(total network bandwidth, sum(disk read iops))`.
- Saturated parallel write iops: `min(total network bandwidth / number of replicas, sum(disk write iops / number of replicas / (write amplification = 4)))`.
- Saturated parallel write iops: `min(total network bandwidth / number of replicas, sum(disk write iops / number of replicas / write amplification))`.
EC/XOR setups (EC N+K):
- Single-threaded (T1Q1) read latency: 1.5 network roundtrips + 1 disk read.
@@ -26,28 +26,36 @@ EC/XOR setups (EC N+K):
- Linear read: `min(total network bandwidth, sum(disk read MB/s))`.
- Linear write: `min(total network bandwidth, sum(disk write MB/s * N/(N+K)))`.
- Saturated parallel read iops: `min(total network bandwidth, sum(disk read iops))`.
- Saturated parallel write iops: roughly `total iops / (N+K) / WA`. More exactly,
`min(total network bandwidth * N/(N+K), sum(disk randrw iops / (N*4 + K*5 + 1)))` with
random read/write mix corresponding to `(N-1)/(N*4 + K*5 + 1)*100 % reads`.
- For example, with EC 2+1 it is: `(7% randrw iops) / 14`.
- With EC 6+3 it is: `(12.5% randrw iops) / 40`.
- Saturated parallel write iops: roughly `total iops / (N+K) / WA`. More exactly:
- With the new store: `min(total network bandwidth * N/(N+K), sum(disk randrw iops / (2 + N-1 + K*2)))`,
with random read/write mix corresponding to `(N-1)/(2 + N-1 + K*2)*100 % reads`.
- For example, with EC 2+1 it is: `(20% randrw iops) / 5`.
- With EC 6+3 it is: `(38% randrw iops) / 13`.
- With the old store: `min(total network bandwidth * N/(N+K), sum(disk randrw iops / (3 + N-1 + K*3)))`,
with random read/write mix corresponding to `(N-1)/(3 + N-1 + K*3)*100 % reads`.
- For example, with EC 2+1 it is: `(14% randrw iops) / 7`.
- With EC 6+3 it is: `(30% randrw iops) / 17`.
Write amplification for 4 KB blocks is usually 3-5 in Vitastor:
1. Journal block write
2. Journal data write
3. Metadata block write
4. Another journal block write for EC/XOR setups
5. Data block write
Write Amplification factor:
- For the new store and for 4 KB writes: WA is always 1 unless you set [atomic_write_size](../config/osd.en.md#atomic_write_size) to 0 manually.
- For the new store and for 8-124 KB writes: WA is 1 if you use NVMe drives with atomic write support, or roughly 2 if you use other drives.
- For the old store, WA is roughly `(2 * write size + 4 KB) / (write size)`. So, for 4 KB writes it's 3, and for 8-124 KB writes it's closer to 2.
- For both the new and the old store and for writes of [block_size](../config/layout-cluster.en.md#block_size): WA is almost 1.
If you manage to get an SSD which handles 512 byte blocks well (Optane?) you may
lower 1, 3 and 4 to 512 bytes (1/8 of data size) and get WA as low as 2.375.
Write Amplification consists of:
- For the new store:
- Buffer block write if non-atomic
- Data block write
- Metadata write(s) (amortized)
- For the old store:
- Journal block write (amortized)
- Journal data write
- Metadata block write
- Another journal block write for EC/XOR setups (amortized)
- Data block write
Implemented NVDIMM support can basically eliminate WA at all - all extra writes will
go to DRAM memory. But this requires a test cluster with NVDIMM - please contact me
if you want to provide me with such cluster for tests.
Lazy fsync also reduces WA for parallel workloads because journal blocks are only
written when they fill up or fsync is requested.
Other possibilities to reduce WA would be to use SSDs with internal 512-byte blocks
or NVDIMM, but both options seem unavailable on the market at the moment.
## In Practice
+27 -20
View File
@@ -27,29 +27,36 @@
- Линейное чтение: сумма МБ/с чтения всех дисков, либо общая производительность сети, если в сеть упрётся раньше.
- Линейная запись: сумма МБ/с записи всех дисков * N/(N+K), либо производительность сети * N / (N+K), если в сеть упрётся раньше.
- Параллельное случайное мелкое чтение: сумма IOPS чтения всех дисков либо производительность сети, если в сеть упрётся раньше.
- Параллельная случайная мелкая запись: грубо `(сумма IOPS / (N+K) / WA)`. Если точнее, то:
сумма смешанного IOPS всех дисков при `(N-1)/(N*4 + K*5 + 1)*100 %` чтения, делённая на `(N*4 + K*5 + 1)`.
Либо, производительность сети * N/(N+K), если в сеть упрётся раньше.
- Например, при EC 2+1 это: `(сумма IOPS при 7% чтения) / 14`.
- При EC 6+3 это: `(сумма IOPS при 12.5% чтения) / 40`.
- Параллельная случайная мелкая запись: грубо `(сумма IOPS / (N+K) / WA)`.
Либо `производительность сети * N/(N+K)`, если в сеть упрётся раньше. Если точнее, то:
- С новым хранилищем: сумма смешанного IOPS всех дисков при `(N-1)/(2 + N-1 + K*2)*100 %` чтения, делённая на `(2 + N-1 + K*2)`.
- Например, при EC 2+1 это: `(сумма IOPS при 20% чтения) / 5`.
- При EC 6+3 это: `(сумма IOPS при 38% чтения) / 13`.
- Со старым хранилищем: сумма смешанного IOPS всех дисков при `(N-1)/(3 + N-1 + K*3)*100 %` чтения, делённая на `(3 + N-1 + K*3)`.
- Например, при EC 2+1 это: `(сумма IOPS при 14% чтения) / 7`.
- При EC 6+3 это: `(сумма IOPS при 30% чтения) / 17`.
WA (мультипликатор записи) для 4 КБ блоков в Vitastor обычно составляет 3-5:
1. Запись метаданных в журнал
2. Запись блока данных в журнал
3. Запись метаданных в БД
4. Ещё одна запись метаданных в журнал при использовании EC
5. Запись блока данных на диск данных
WA (Write Amplification, мультипликатор записи):
- С новым хранилищем для 4 КБ записи: WA всегда примерно 1, если только вы не установите [atomic_write_size](../config/osd.ru.md#atomic_write_size) вручную в 0.
- С новым хранилищем и большими записями (8-124 КБ): WA примерно 1, если вы используете NVMe-диски с поддержкой атомарной записи,
или примерно 2, если вы используете другие диски.
- Со старым хранилищем, WA примерно `(2 * размер записи + 4 КБ) / (размер записи)`. То есть, для 4 КБ записи WA=3, а для 8-124 КБ WA ближе к 2.
- И с новым, и со старым хранилищем и для записи размером [block_size](../config/layout-cluster.ru.md#block_size): WA примерно равен 1.
Если вы найдёте SSD, хорошо работающий с 512-байтными блоками данных (Optane?),
то 1, 3 и 4 можно снизить до 512 байт (1/8 от размера данных) и получить WA всего 2.375.
Мультипликатор записи состоит из:
- С новым хранилищем:
- Запись блока буфера, если диски без поддержки атомарной записи
- Запись блока данных
- Запись(-и) блоков метаданных (амортизированные)
- Со старым хранилищем:
- Запись блока журнала (амортизированная)
- Запись данных в журнал
- Запись блока метаданных
- Ещё одна запись блока журнала для EC/XOR пулов (амортизированная)
- Запись блока данных
Если реализовать поддержку NVDIMM, то WA можно, условно говоря, ликвидировать вообще - все
дополнительные операции записи смогут обслуживаться DRAM памятью. Но для этого необходим
тестовый кластер с NVDIMM - пишите, если готовы предоставить такой для тестов.
Кроме того, WA снижается при использовании отложенного/ленивого сброса при параллельной
нагрузке, т.к. блоки журнала записываются на диск только когда они заполняются или явным
образом запрашивается fsync.
Другими потенциальными возможностями снижения WA могли бы быть SSD с внутренним 512-байтным блоком
либо NVDIMM, но и то, и другое сейчас выглядит недоступным на рынке.
## На практике
+37 -4
View File
@@ -14,6 +14,7 @@
- [Removing a failed disk](#removing-a-failed-disk)
- [Adding a disk](#adding-a-disk)
- [Restoring from lost pool configuration](#restoring-from-lost-pool-configuration)
- [Incompatibility problems](#Incompatibility-problems)
- [Upgrading Vitastor](#upgrading-vitastor)
- [OSD memory usage](#osd-memory-usage)
@@ -35,10 +36,19 @@ PG state consists of exactly 1 base state and an arbitrary number of additional
PG state always includes exactly 1 of the following base states:
- **active** — PG is active and handles user I/O.
- **incomplete** — Not enough OSDs are available to activate this PG. That is, more disks
are lost than it's allowed by the pool's redundancy scheme. For example, if the pool has
pg_size=3 and pg_minsize=1, part of the data may be written only to 1 OSD. If that exact
OSD is lost, PG will become **incomplete**.
- **incomplete** — Not enough OSDs are available to activate this PG. More exactly, that
means one of the following:
- Less than pg_minsize current target OSDs are available for the PG. I.e. more disks
are lost than allowed by the pool's redundancy scheme.
- All OSDs of some of PG's history records are unavailable, or, for EC pools, less
than (pg_size-parity_chunks) OSDs are available in one of the history records.
In other words it means that some data in this PG was written to an OSD set such that
it's currently impossible to read it back because these OSDs are down. For example,
if the pool has pg_size=3 and pg_minsize=1, part of the data may be written only to
1 OSD. If that exact OSD is lost, PG becomes **incomplete**.
- [allow_net_split](../config/osd.en.md#allow_net_split) is disabled (default) and
primary OSD of the PG can't connect to some secondary OSDs marked as alive in etcd.
I.e. a network partition happened: OSDs can talk to etcd, but not to some other OSDs.
- **offline** — PG isn't activated by any OSD at all. Either primary OSD isn't set for
this PG at all (if the pool is just created), or an unavailable OSD is set as primary,
or the primary OSD refuses to start this PG (for example, because of wrong block_size),
@@ -157,6 +167,17 @@ done
After that all PGs should peer and find all previous data.
## Incompatibility problems
### ISA-L 2.31
⚠ It is FORBIDDEN to use Vitastor 2.1.0 and earlier versions with ISA-L 2.31 and newer if
you use EC N+K pools and K > 1 on a CPU with GF-NI instruction support, because it WILL
lead to **data loss** during EC recovery.
If you accidentally upgraded ISA-L to 2.31 but didn't upgrade Vitastor and restarted OSDs,
then stop them as soon as possible and either update Vitastor or roll back ISA-L.
## Upgrading Vitastor
Every upcoming Vitastor version is usually compatible with previous both forward
@@ -210,6 +231,18 @@ Upgrading from <= 0.5.x to >= 0.6.x is not supported.
Downgrade are also allowed freely, except the following specific instructions:
### 3.x -> 2.x
Versions 3.0.0 and newer contain two store implementations - an old one and a new
one, unsupported in 2.x and previous versions. So you should check your OSD store
versions before downgrading to 2.x with the following command:
`vitastor-disk read-sb /dev/vitastor/osdXX-data | jq -r .meta_format`
If it prints 3 then OSD uses the new store and you can't downgrade it to 2.x.
If it prints 2 or nothing then OSD uses the old store and the downgrade is allowed.
### 1.8.0 to 1.7.1
Before downgrading from version >= 1.8.0 to version <= 1.7.1
+38 -4
View File
@@ -14,6 +14,7 @@
- [Удаление неисправного диска](#удаление-неисправного-диска)
- [Добавление диска](#добавление-диска)
- [Восстановление потерянной конфигурации пулов](#восстановление-потерянной-конфигурации-пулов)
- [Проблемы несовместимости](#проблемы-несовместимости)
- [Обновление Vitastor](#обновление-vitastor)
- [Потребление памяти OSD](#потребление-памяти-osd)
@@ -35,10 +36,20 @@
Состояние PG включает в себя ровно 1 флаг из следующих:
- **active** — PG активна и обрабатывает запросы ввода-вывода от пользователей.
- **incomplete** — Недостаточно живых OSD, чтобы включить эту PG.
То есть, дисков потеряно больше, чем разрешено схемой отказоустойчивости пула и pg_minsize.
Например, если у пула pg_size=3 и pg_minsize=1, то часть данных может записаться всего на 1 OSD.
Если потом конкретно этот OSD упадёт, PG окажется **incomplete**.
- **incomplete** — Недостаточно живых OSD, чтобы включить эту PG. Если точнее, то это
означает один из следующих вариантов:
- Доступно менее, чем pg_minsize текущих целевых OSD данной PG. Иными словами, потеряно
больше дисков, чем это разрешает схема отказоустойчивости пула.
- Все OSD одной из исторических записей PG недоступны, или, для EC-пулов, в одной
из исторических записей PG доступно менее, чем (pg_size-parity_chunks) OSD. Другими
словами это означает, что часть данных этой PG была записана в такой набор OSD, из
которого их сейчас невозможно прочитать обратно, так как OSD не включены. Например,
если у пула pg_size=3 и pg_minsize=1, то часть данных может записаться всего на 1 OSD.
Если потом конкретно этот OSD упадёт, PG окажется **incomplete**.
- [allow_net_split](../config/osd.ru.md#allow_net_split) отключено (по умолчанию) и
первичный OSD данной PG не может соединиться с частью вторичных OSD этой PG, помеченных
как живых в etcd. Это означает, что произошло разделение сети: OSD могут общаться с etcd,
но не могут общаться с частью других OSD.
- **offline** — PG вообще не активирована ни одним OSD. Либо первичный OSD не назначен вообще
(если пул только создан), либо в качестве первичного назначен недоступный OSD, либо
назначенный OSD отказывается запускать эту PG (например, из-за несовпадения block_size),
@@ -153,6 +164,17 @@ done
После этого все PG должны пройти peering и найти все предыдущие данные.
## Проблемы несовместимости
### ISA-L 2.31
⚠ ЗАПРЕЩЕНО использовать Vitastor 2.1.0 и более ранних версий с библиотекой ISA-L версии 2.31
или более новой, если вы используете EC-пулы N+K и K > 1 на CPU с поддержкой инструкций GF-NI,
так как это приведёт к **потере данных** при восстановлении из EC.
Если вы случайно обновили ISA-L до 2.31, но не обновили Vitastor, и успели перезапустить OSD,
то как можно скорее остановите их все и либо обновите Vitastor, либо откатите ISA-L.
## Обновление Vitastor
Обычно каждая следующая версия Vitastor совместима с предыдущими и "вперёд", и "назад"
@@ -206,6 +228,18 @@ done
Откат (понижение версии) тоже свободно разрешён, кроме указанных ниже случаев:
### 3.x -> 2.x
Версии 3.0.0 и более новые содержат две реализации хранилища - старую и новую, не
поддерживаемую в 2.x и предыдущих версиях. Таким образом, перед откатом на 2.x вам
следует проверить, какая версия хранилища используется вашими OSD - командой:
`vitastor-disk read-sb /dev/vitastor/osdXX-data | jq -r .meta_format`
Если выводится 3, это новое хранилище и откатить такой OSD до 2.x нельзя.
Если выводится 2 или не выводится ничего, это старое хранилище и откат разрешён.
### 1.8.0 -> 1.7.1
Перед понижением версии с >= 1.8.0 до <= 1.7.1 вы должны скопировать ключ
+12 -7
View File
@@ -100,12 +100,14 @@ List images (only matching `<glob>` pattern(s) if passed).
Options:
```
--exact Do not match glob patterns as names, select only exact name matches.
-p|--pool POOL Filter images by pool ID or name
-l|--long Also report allocated size and I/O statistics
--del Also include delete operation statistics
--sort FIELD Sort by specified field (name, size, used_size, <read|write|delete>_<iops|bps|lat|queue>)
-r|--reverse Sort in descending order
-n|--count N Only list first N items
--tree Show image snapshot/clone tree
```
Example output:
@@ -355,7 +357,7 @@ Set OSD reweight, tags or noout flag. See detail description in [OSD config docu
## pg-list
`vitastor-cli pg-list|pg-ls|list-pg|ls-pg|ls-pgs [OPTIONS] [state1+state2] [^state3] [...]`
`vitastor-cli pg-list|pg-ls|list-pg|ls-pg|ls-pgs|pgs [OPTIONS] [state1+state2] [^state3] [...]`
List PGs with any of listed state filters (^ or ! in the beginning is negation). Options:
@@ -363,6 +365,7 @@ List PGs with any of listed state filters (^ or ! in the beginning is negation).
--pool <pool name or number> Only list PGs of the given pool.
--min <min pg number> Only list PGs with number >= min.
--max <max pg number> Only list PGs with number <= max.
--osd 1,2,... Only list PGs with some data on specified OSD(s).
```
Examples:
@@ -377,11 +380,11 @@ Examples:
Create a pool. Required parameters:
| <!-- --> | <!-- --> |
|--------------------------|---------------------------------------------------------------------------------------|
| `-s R` or `--pg_size R` | Number of replicas for replicated pools |
| `--ec N+K` | Number of data (N) and parity (K) chunks for erasure-coded pools |
| `-n N` or `--pg_count N` | PG count for the new pool (start with 10*<OSD count>/pg_size rounded to a power of 2) |
| <!-- --> | <!-- --> |
|--------------------------|-----------------------------------------------------------------------------------------|
| `-s R` or `--pg_size R` | Number of replicas for replicated pools |
| `--ec N+K` | Number of data (N) and parity (K) chunks for erasure-coded pools |
| `-n N` or `--pg_count N` | PG count for the new pool (start with 10*\<OSD count\>/pg_size rounded to a power of 2) |
Optional parameters:
@@ -396,9 +399,11 @@ Optional parameters:
| `--immediate_commit none` | Put pool only on OSDs with this or larger immediate_commit (none < small < all) |
| `--level_placement <rules>` | Use additional failure domain rules (example: "dc=112233") |
| `--raw_placement <rules>` | Specify raw PG generation rules ([details](../config/pool.en.md#raw_placement)) |
| `--local_reads primary` | Local read policy for replicated pools: primary, nearest or random |
| `--primary_affinity_tags tags` | Prefer to put primary copies on OSDs with all specified tags |
| `--scrub_interval <time>` | Enable regular scrubbing for this pool. Format: number + unit s/m/h/d/M/y |
| `--used_for_fs <name>` | Mark pool as used for VitastorFS with metadata in image <name> |
| `--used_for_app fs:<name>` | Mark pool as used for VitastorFS with metadata in image `<name>` |
| `--used_for_app s3:<name>` | Mark pool as used for S3 location with name `<name>` |
| `--pg_stripe_size <number>` | Increase object grouping stripe |
| `--max_osd_combinations 10000` | Maximum number of random combinations for LP solver input |
| `--wait` | Wait for the new pool to come online |
+14 -8
View File
@@ -22,6 +22,8 @@ vitastor-cli - интерфейс командной строки для адм
- [flatten](#flatten)
- [rm-data](#rm-data)
- [merge-data](#merge-data)
- [describe](#describe)
- [fix](#fix)
- [alloc-osd](#alloc-osd)
- [rm-osd](#rm-osd)
- [osd-tree](#osd-tree)
@@ -100,12 +102,14 @@ kaveri 2/1 32 0 B 10 G 0 B 100% 0%
Опции:
```
--exact Не применять ФС-шаблоны к именам, выводить только точные совпадения
-p|--pool POOL Фильтровать образы по пулу (ID или имени)
-l|--long Также выводить статистику занятого места и ввода-вывода
--del Также выводить статистику операций удаления
--sort FIELD Сортировать по заданному полю (name, size, used_size, <read|write|delete>_<iops|bps|lat|queue>)
-r|--reverse Сортировать в обратном порядке
-n|--count N Показывать только первые N записей
--tree Вывести снапшоты и клоны в виде дерева
```
Пример вывода:
@@ -375,9 +379,10 @@ OSD PARENT UP SIZE USED% TAGS WEIGHT BLOCK BITMAP
в начале фильтра означает отрицание). Опции:
```
--pool <pool name or number> Only list PGs of the given pool.
--min <min pg number> Only list PGs with number >= min.
--max <max pg number> Only list PGs with number <= max.
--pool <pool name or number> Вывести только PG в заданном пуле.
--min <min pg number> Вывести только PG с номерами >= min.
--max <max pg number> Вывести только PG с номерами <= max.
--osd 1,2,... Вывести только PG с данными на заданных OSD.
```
Примеры:
@@ -392,11 +397,11 @@ OSD PARENT UP SIZE USED% TAGS WEIGHT BLOCK BITMAP
Создать пул. Обязательные параметры:
| <!-- --> | <!-- --> |
|---------------------------|---------------------------------------------------------------------------------------------|
| `-s R` или `--pg_size R` | Число копий данных для реплицированных пулов |
| `--ec N+K` | Число частей данных (N) и чётности (K) для пулов с кодами коррекции ошибок |
| `-n N` или `--pg_count N` | Число PG для нового пула (начните с 10*<число OSD>/pg_size, округлённого до степени двойки) |
| <!-- --> | <!-- --> |
|---------------------------|-----------------------------------------------------------------------------------------------|
| `-s R` или `--pg_size R` | Число копий данных для реплицированных пулов |
| `--ec N+K` | Число частей данных (N) и чётности (K) для пулов с кодами коррекции ошибок |
| `-n N` или `--pg_count N` | Число PG для нового пула (начните с 10*\<число OSD\>/pg_size, округлённого до степени двойки) |
Необязательные параметры:
@@ -411,6 +416,7 @@ OSD PARENT UP SIZE USED% TAGS WEIGHT BLOCK BITMAP
| `--immediate_commit none` | ...только OSD с этим или большим immediate_commit (none < small < all) |
| `--level_placement <rules>` | Задать правила дополнительных доменов отказа (пример: "dc=112233") |
| `--raw_placement <rules>` | Задать низкоуровневые правила генерации PG ([детали](../config/pool.ru.md#raw_placement)) |
| `--local_reads primary` | Политика локальных чтений для реплик: primary, nearest или random |
| `--primary_affinity_tags tags` | Предпочитать OSD со всеми данными тегами для роли первичных |
| `--scrub_interval <time>` | Включить скрабы с заданным интервалом времени (число + единица s/m/h/d/M/y) |
| `--pg_stripe_size <number>` | Увеличить блок группировки объектов по PG |
+24
View File
@@ -14,6 +14,7 @@ It supports the following commands:
- [upgrade-simple](#upgrade-simple)
- [resize](#resize)
- [raw-resize](#raw-resize)
- [trim](#trim)
- [start/stop/restart/enable/disable](#start/stop/restart/enable/disable)
- [purge](#purge)
- [read-sb](#read-sb)
@@ -50,6 +51,9 @@ Options (automatic mode):
```
--osd_per_disk <N>
Create <N> OSDs on each disk (default 1)
--meta_format 3
Metadata store version. 3 is the new log-structured store, 2 is the stable store
from Vitastor 0.9-2.x, 1 is the legacy store from Vitastor 0.6-0.8.
--hybrid
Prepare hybrid (HDD+SSD, NVMe+SATA or etc) OSDs using provided devices. By default,
any passed SSDs will be used for journals and metadata, HDDs will be used for data,
@@ -72,6 +76,8 @@ Options (automatic mode):
--max_other 10%
Use disks for OSD data even if they already have non-Vitastor partitions,
but only if these take up no more than this percent of disk space.
--dry-run
Check and print new OSD count for each disk but do not actually create them.
```
Options (single-device mode):
@@ -89,6 +95,8 @@ Options (single-device mode):
Options (both modes):
```
--tags tag1,tag2 Set new OSD tag(s)
--weight <number> Set new OSD weight (between 0 to 1)
--journal_size 1G/32M Set journal size (area or partition size)
--block_size 1M/128k Set blockstore object size
--bitmap_granularity 4k Set bitmap granularity
@@ -97,6 +105,9 @@ Options (both modes):
--data_device_block 4k Override data device block size
--meta_device_block 4k Override metadata device block size
--journal_device_block 4k Override journal device block size
--discard_on_start 0 TRIM unused data device blocks every OSD start (default off)
--min_discard_size 1M Minimum TRIM block size
--json Enable JSON output
```
[immediate_commit](../config/layout-cluster.en.md#immediate_commit) setting is
@@ -179,6 +190,19 @@ parameters from OSD command line (i.e. from systemd unit or superblock).
SIZE may include k/m/g/t suffixes. If any of the new layout parameter
options are not specified, old values will be used.
## trim
`vitastor-disk trim <osd_num>|<osd_device> [<osd_num>|<osd_device>...]`
Try to discard unused blocks (SSD TRIM) on the data device of each of the OSD(s).
May only be used on stopped OSDs. Options:
```
--min_discard_size 1M Minimum TRIM block size
--discard_granularity 0 Override device's discard granularity
```
## start/stop/restart/enable/disable
`vitastor-disk start|stop|restart|enable|disable [--now] <device> [device2 device3 ...]`
+24
View File
@@ -50,6 +50,9 @@ vitastor-disk - инструмент командной строки для уп
```
--osd_per_disk <N>
Создавать по несколько (<N>) OSD на каждом диске (по умолчанию 1)
--meta_format 3
Версия хранилища метаданных. 3 - новое лог-структурированное хранилище,
2 - стабильное хранилище из Vitastor 0.9-2.x, 1 - старое хранилище из Vitastor 0.6-0.8.
--hybrid
Инициализировать гибридные (HDD+SSD, NVMe+SATA и т.п.) OSD на указанных дисках.
По умолчанию, SSD будут использованы для журналов и метаданных, а HDD - для данных,
@@ -74,6 +77,8 @@ vitastor-disk - инструмент командной строки для уп
--max_other 10%
Использовать диски под данные OSD, даже если на них уже есть не-Vitastor-овые
разделы, но только в случае, если они занимают не более данного процента диска.
--dry-run
Проверить и вывести число новых OSD для каждого диска, но не создавать их.
```
Опции для режима одного OSD:
@@ -91,6 +96,8 @@ vitastor-disk - инструмент командной строки для уп
Опции для обоих режимов:
```
--tags tag1,tag2 Задать теги для новых OSD
--weight <number> Задать вес для новых OSD (от 0 до 1)
--journal_size 1G/32M Задать размер журнала (области или раздела журнала)
--block_size 1M/128k Задать размер объекта хранилища
--bitmap_granularity 4k Задать гранулярность битовых карт
@@ -99,6 +106,9 @@ vitastor-disk - инструмент командной строки для уп
--data_device_block 4k Задать размер блока устройства данных
--meta_device_block 4k Задать размер блока метаданных
--journal_device_block 4k Задать размер блока журнала
--discard_on_start 0 Выполнять TRIM пустых блоков данных при запуске OSD (по умолчанию нет)
--min_discard_size 1M Минимальный размер блока для TRIM
--json Включить JSON-вывод
```
Настройка [immediate_commit](../config/layout-cluster.ru.md#immediate_commit)
@@ -182,6 +192,20 @@ throttle_target_mbs, throttle_target_parallelism, throttle_threshold_us.
`РАЗМЕР` может быть указан с суффиксами k/m/g/t. Если любой из новых параметров
расположения не указан, он принимается равным старому значению.
## trim
`vitastor-disk trim <osd_num>|<osd_device> [<osd_num>|<osd_device>...]`
Попробовать пометить пустые блоки дисков данных всех указанных OSD неиспользуемыми
(выполнить команду SSD TRIM).
Можно использовать только с остановленными OSD. Опции:
```
--min_discard_size 1M Минимальный размер блока для TRIM
--discard_granularity 0 Кратность размера блока для TRIM
```
## start/stop/restart/enable/disable
`vitastor-disk start|stop|restart|enable|disable [--now] <device> [device2 device3 ...]`
+63 -1
View File
@@ -14,6 +14,9 @@ Commands:
- [upgrade](#upgrade)
- [defrag](#defrag)
⚠️ Important: follow the instructions from [Linux NFS write size](#linux-nfs-write-size)
for optimal Vitastor NFS performance if you use EC and HDD and mount your NFS from Linux.
## Pseudo-FS
Simplified pseudo-FS proxy is used for file-based image access emulation. It's not
@@ -58,7 +61,7 @@ To use VitastorFS:
2. Create an image for FS metadata, preferably in a faster (SSD or replica-HDD) pool,
but you can create it in the data pool too if you want (image size doesn't matter):
`vitastor-cli create -s 10G -p fastpool testfs`
3. Mark data pool as an FS pool: `vitastor-cli modify-pool --used-for-fs testfs data-pool`
3. Mark data pool as an FS pool: `vitastor-cli modify-pool --used-for-app fs:testfs data-pool`
4. Either mount the FS: `vitastor-nfs mount --fs testfs --pool data-pool /mnt/vita`
5. Or start the NFS server: `vitastor-nfs start --fs testfs --pool data-pool`
@@ -86,6 +89,8 @@ POSIX features currently not implemented in VitastorFS:
instead of actually allocated space
- Access times (`atime`) are not tracked (like `-o noatime`)
- Modification time (`mtime`) is updated lazily every second (like `-o lazytime`)
- Permission enforcement is disabled by default (and Linux NFS client doesn't
enforce them too). Use `--enforce 1` to enable it.
Other notable missing features which should be addressed in the future:
- Inode ID reuse. Currently inode IDs always grow, the limit is 2^48 inodes, so
@@ -100,6 +105,62 @@ Other notable missing features which should be addressed in the future:
in the DB. The FS is implemented is such way that this garbage doesn't affect its
function, but having a tool to clean it up still seems a right thing to do.
## Linux NFS write size
Linux NFS client (nfs/nfsv3/nfsv4 kernel modules) has a hard-coded maximum I/O size,
currently set to 1 MB - see `rsize` and `wsize` in [man 5 nfs](https://linux.die.net/man/5/nfs).
This means that when you write to a file in an FS mounted over NFS, the maximum write
request size is 1 MB, even in the O_DIRECT mode and even if the original write request
is larger.
However, for optimal linear write performance in Vitastor EC (erasure-coded) pools,
the size of write requests should be a multiple of [block_size](../config/layout-cluster.en.md#block_size),
multiplied by the data chunk count of the pool ([pg_size](../config/pool.en.md#pg_size)-[parity_chunks](../config/pool.en.md#parity_chunks)).
When write requests are smaller or not a multiple of this number, Vitastor has to first
read paired data blocks from disks, calculate new parity blocks and only then write them
back. Obviously this is 2-3 times slower than a simple disk write.
Vitastor HDD setups use 1 MB block_size by default. So, for optimal performance, if
you use EC 2+1 and HDD, you need your NFS client to send 2 MB write requests, if you
use EC 4+1 - 4 MB and so on.
But Linux NFS client only writes in 1 MB chunks. 😢
The good news is that you can fix it by rebuilding Linux NFS kernel modules 😉 🤩!
You need to change NFS_MAX_FILE_IO_SIZE in nfs_xdr.h and then rebuild and reload modules.
The instruction, using Debian as an example (should be ran under root):
```
# download current Linux kernel headers required to build modules
apt-get install linux-headers-`uname -r`
# replace NFS_MAX_FILE_IO_SIZE with a desired number (here it's 4194304 - 4 MB)
sed -i 's/NFS_MAX_FILE_IO_SIZE\s*.*/NFS_MAX_FILE_IO_SIZE\t(4194304U)/' /lib/modules/`uname -r`/source/include/linux/nfs_xdr.h
# download current Linux kernel source
mkdir linux_src
cd linux_src
apt-get source linux-image-`uname -r`-unsigned
# build NFS modules
cd linux-*/fs/nfs
make -C /lib/modules/`uname -r`/build M=$PWD -j8 modules
make -C /lib/modules/`uname -r`/build M=$PWD modules_install
# move default NFS modules away
mv /lib/modules/`uname -r`/kernel/fs/nfs ~/nfs_orig_`uname -r`
depmod -a
# unload old modules and load the new ones
rmmod nfsv3 nfs
modprobe nfsv3
```
After these (not much complicated 🙂) manipulations NFS begins to be mounted
with new wsize and rsize by default and it fixes Vitastor-NFS linear write performance.
## Horizontal scaling
Linux NFS 3.0 client doesn't support built-in scaling or failover, i.e. you can't
@@ -199,4 +260,5 @@ Options:
| `--nfspath <PATH>` | set NFS export path to \<PATH> (default is /) |
| `--pidfile <FILE>` | write process ID to the specified file |
| `--logfile <FILE>` | log to the specified file |
| `--enforce 1` | enforce permissions at the server side (no by default) |
| `--foreground 1` | stay in foreground, do not daemonize |
+67 -1
View File
@@ -14,6 +14,9 @@
- [upgrade](#upgrade)
- [defrag](#defrag)
⚠️ Важно: для оптимальной производительности Vitastor NFS в Linux при использовании
HDD и EC (erasure кодов) выполните инструкции из раздела [Размер записи Linux NFS](#размер-записи-linux-nfs).
## Псевдо-ФС
Упрощённая реализация псевдо-ФС используется для эмуляции файлового доступа к блочным
@@ -60,7 +63,7 @@ JSON-формате :-). Для инспекции содержимого БД
или по крайней мере на HDD, но без EC), но можно и в том же пуле, что данные
(размер образа значения не имеет):
`vitastor-cli create -s 10G -p fastpool testfs`
3. Пометьте пул данных как ФС-пул: `vitastor-cli modify-pool --used-for-fs testfs data-pool`
3. Пометьте пул данных как ФС-пул: `vitastor-cli modify-pool --used-for-app fs:testfs data-pool`
4. Либо примонтируйте ФС: `vitastor-nfs mount --fs testfs --pool data-pool /mnt/vita`
5. Либо запустите сетевой NFS-сервер: `vitastor-nfs start --fs testfs --pool data-pool`
@@ -88,6 +91,8 @@ JSON-формате :-). Для инспекции содержимого БД
stat(2), так что `du` всегда показывает сумму размеров файлов, а не фактически занятое место
- Времена доступа (`atime`) не отслеживаются (как будто ФС смонтирована с `-o noatime`)
- Времена модификации (`mtime`) отслеживаются асинхронно (как будто ФС смонтирована с `-o lazytime`)
- Привилегии доступа по умолчанию не проверяются сервером (клиент NFS Linux их также не проверяет).
Чтобы включить проверки, используйте опцию `--enforce 1`.
Другие недостающие функции, которые нужно добавить в будущем:
- Переиспользование номеров инодов. В текущей реализации номера инодов всё время
@@ -104,6 +109,66 @@ JSON-формате :-). Для инспекции содержимого БД
записи. ФС устроена так, что на работу они не влияют, но для порядка и их стоит
уметь подчищать.
## Размер записи Linux NFS
Клиент Linux NFS (модули ядра nfs/nfsv3/nfsv4) имеет фиксированный в коде максимальный
размер запроса ввода-вывода, равный 1 МБ - см. `rsize` и `wsize` в [man 5 nfs](https://linux.die.net/man/5/nfs).
Это означает, что когда вы записываете в файл в примонтированной по NFS файловой системе,
максимальный размер запроса записи составляет 1 МБ, даже в режиме O_DIRECT и даже если
исходный запрос записи был больше.
Однако для оптимальной скорости линейной записи в Vitastor при использовании EC-пулов
(пулов с кодами коррекции ошибок) запросы записи должны быть по размеру кратны
[block_size](../config/layout-cluster.ru.md#block_size), умноженному на число частей
данных пула ([pg_size](../config/pool.ru.md#pg_size)-[parity_chunks](../config/pool.ru.md#parity_chunks)).
Если запросы записи меньше или не кратны, то Vitastor приходится сначала прочитать
с дисков старые версии парных блоков данных, рассчитать новые блоки чётности и только
после этого записать их на диски. Естественно, это в 2-3 раза медленнее простой записи
на диск.
При этом block_size на жёстких дисках по умолчанию устанавливается равным 1 МБ.
Таким образом, если вы используете EC 2+1 и HDD, для оптимальной скорости записи вам
нужно, чтобы NFS-клиент писал по 2 МБ, если EC 4+1 и HDD - то по 4 МБ, и т.п.
А Linux NFS-клиент пишет только по 1 МБ. 😢
Но это можно исправить, пересобрав модули ядра Linux NFS 😉 🤩! Для этого нужно
поменять значение переменной NFS_MAX_FILE_IO_SIZE в заголовочном файле nfs_xdr.h,
после чего пересобрать модули NFS.
Инструкция по пересборке на примере Debian (выполнять под root):
```
# скачиваем заголовки для сборки модулей для текущего ядра Linux
apt-get install linux-headers-`uname -r`
# заменяем в заголовках NFS_MAX_FILE_IO_SIZE на желаемый (здесь 4194304 - 4 МБ)
sed -i 's/NFS_MAX_FILE_IO_SIZE\s*.*/NFS_MAX_FILE_IO_SIZE\t(4194304U)/' /lib/modules/`uname -r`/source/include/linux/nfs_xdr.h
# скачиваем исходный код текущего ядра
mkdir linux_src
cd linux_src
apt-get source linux-image-`uname -r`-unsigned
# собираем модули NFS
cd linux-*/fs/nfs
make -C /lib/modules/`uname -r`/build M=$PWD -j8 modules
make -C /lib/modules/`uname -r`/build M=$PWD modules_install
# убираем в сторону штатные модули NFS
mv /lib/modules/`uname -r`/kernel/fs/nfs ~/nfs_orig_`uname -r`
depmod -a
# выгружаем старые модули и загружаем новые
rmmod nfsv3 nfs
modprobe nfsv3
```
После такой (относительно нехитрой 🙂) манипуляции NFS начинает по умолчанию
монтироваться с новыми wsize и rsize, и производительность линейной записи в Vitastor-NFS
исправляется.
## Горизонтальное масштабирование
Клиент Linux NFS 3.0 не поддерживает встроенное масштабирование или отказоустойчивость.
@@ -207,4 +272,5 @@ VitastorFS из GPUDirect.
| `--nfspath <PATH>` | установить путь NFS-экспорта в \<PATH> (по умолчанию /) |
| `--pidfile <FILE>` | записать ID процесса в заданный файл |
| `--logfile <FILE>` | записывать логи в заданный файл |
| `--enforce 1` | проверять права доступа на стороне сервера (по умолчанию нет) |
| `--foreground 1` | не уходить в фон после запуска |
+21 -17
View File
@@ -130,23 +130,16 @@ Linux kernel, starting with version 5.15, supports a new interface for attaching
to the host - VDUSE (vDPA Device in Userspace). QEMU, starting with 7.2, has support for
exporting QEMU block devices over this protocol using qemu-storage-daemon.
VDUSE is currently the best interface to attach Vitastor disks as kernel devices because:
- It avoids data copies and thus achieves much better performance than [NBD](nbd.en.md)
- It doesn't have NBD timeout problem - the device doesn't die if an operation executes for too long
VDUSE advantages:
- VDUSE copies memory 1 time instead of 2, and is thus faster than [NBD](nbd.en.md) for linear read/write.
- It doesn't have NBD timeout problem - the device doesn't die if an operation executes for too long.
- It doesn't have hung device problem - if the userspace process dies it can be restarted (!)
and block device will continue operation
- It doesn't seem to have the device number limit
and block device will continue operation (UBLK can do it too).
- It doesn't seem to have the device number limit (UBLK also doesn't).
Example performance comparison:
| | direct fio | NBD | VDUSE |
|----------------------|-------------|-------------|-------------|
| linear write | 3.85 GB/s | 1.12 GB/s | 3.85 GB/s |
| 4k random write Q128 | 240000 iops | 120000 iops | 178000 iops |
| 4k random write Q1 | 9500 iops | 7620 iops | 7640 iops |
| linear read | 4.3 GB/s | 1.8 GB/s | 2.85 GB/s |
| 4k random read Q128 | 287000 iops | 140000 iops | 189000 iops |
| 4k random read Q1 | 9600 iops | 7640 iops | 7780 iops |
At the same time, VDUSE may be slower or faster than [UBLK](ublk.en.md) for linear read/write,
and iops-wise it's sometimes even slower than NBD. See performance comparison examples at the page [UBLK](ublk.en.md).
To try VDUSE you need at least Linux 5.15, built with VDUSE support
(CONFIG_VDPA=m, CONFIG_VDPA_USER=m, CONFIG_VIRTIO_VDPA=m).
@@ -162,10 +155,12 @@ apt-get install linux-headers-`uname -r`
apt-get build-dep linux-image-`uname -r`-unsigned
apt-get source linux-image-`uname -r`-unsigned
cd linux*/drivers/vdpa
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules modules_install
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m modules_install
cat Module.symvers >> /lib/modules/`uname -r`/build/Module.symvers
cd ../virtio
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules modules_install
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m modules_install
depmod -a
```
@@ -191,3 +186,12 @@ To remove the device:
vdpa dev del test1
kill <qemu-storage-daemon_process_PID>
```
## Veeam
Vitastor QEMU driver has a feature that allows to trick third-party systems like Veeam not able to parse qemu-img
vitastor URIs: [qemu_file_mirror_path](../config/client.en.md#qemu_file_mirror_path).
To make such systems work, you should set this option to an FS directory path (for example, `/mnt/vitastor/`) and
mount this directory using [`vitastor-nfs mount --block`](../usage/nfs.en.md). It will make them access
your images using files and, hopefully, succeed in doing their normal job :).
+21 -18
View File
@@ -132,24 +132,16 @@ qemu-system-x86_64 -enable-kvm -m 2048 -M accel=kvm,memory-backend=mem \
к системе - VDUSE (vDPA Device in Userspace), а в QEMU, начиная с версии 7.2, есть поддержка
экспорта блочных устройств QEMU по этому протоколу через qemu-storage-daemon.
VDUSE - на данный момент лучший интерфейс для подключения дисков Vitastor в виде блочных
устройств на уровне ядра, ибо:
- VDUSE не копирует данные и поэтому достигает значительно лучшей производительности, чем [NBD](nbd.ru.md)
- Также оно не имеет проблемы NBD-таймаута - устройство не умирает, если операция выполняется слишком долго
- Также оно не имеет проблемы подвисающих устройств - если процесс-обработчик умирает, его можно
перезапустить (!) и блочное устройство продолжит работать
- По-видимому, у него нет предела числа подключаемых в систему устройств
Преимущества VDUSE:
Пример сравнения производительности:
- VDUSE копирует данные 1 раз, а не 2, и поэтому он быстрее, чем [NBD](nbd.ru.md) при линейном доступе.
- VDUSE не имеет проблемы NBD-таймаута - устройство не умирает, если операция выполняется слишком долго.
- VDUSE не имеет проблемы подвисающих устройств - если процесс-обработчик умирает, его можно
перезапустить (!) и блочное устройство продолжит работать (в UBLK это тоже поддерживается).
- По-видимому, у него нет предела числа подключаемых в систему устройств (в UBLK лимита тоже нет).
| | Прямой fio | NBD | VDUSE |
|--------------------------|-------------|-------------|-------------|
| линейная запись | 3.85 GB/s | 1.12 GB/s | 3.85 GB/s |
| 4k случайная запись Q128 | 240000 iops | 120000 iops | 178000 iops |
| 4k случайная запись Q1 | 9500 iops | 7620 iops | 7640 iops |
| линейное чтение | 4.3 GB/s | 1.8 GB/s | 2.85 GB/s |
| 4k случайное чтение Q128 | 287000 iops | 140000 iops | 189000 iops |
| 4k случайное чтение Q1 | 9600 iops | 7640 iops | 7780 iops |
Однако, при линейном доступе VDUSE может быть медленнее UBLK (а может быть и быстрее), а по iops
VDUSE иногда даже медленнее NBD. Пример сравнения производительности смотрите на странице [UBLK](ublk.ru.md).
Чтобы попробовать VDUSE, вам нужно ядро Linux как минимум версии 5.15, собранное с поддержкой
VDUSE (CONFIG_VDPA=m, CONFIG_VDPA_USER=m, CONFIG_VIRTIO_VDPA=m).
@@ -165,10 +157,12 @@ apt-get install linux-headers-`uname -r`
apt-get build-dep linux-image-`uname -r`-unsigned
apt-get source linux-image-`uname -r`-unsigned
cd linux*/drivers/vdpa
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules modules_install
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m modules_install
cat Module.symvers >> /lib/modules/`uname -r`/build/Module.symvers
cd ../virtio
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules modules_install
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m -j8 modules
make -C /lib/modules/`uname -r`/build M=$PWD CONFIG_VDPA=m CONFIG_VDPA_USER=m CONFIG_VIRTIO_VDPA=m modules_install
depmod -a
```
@@ -194,3 +188,12 @@ vdpa dev add name test1 mgmtdev vduse
vdpa dev del test1
kill <PID_процесса_qemu-storage-daemon>
```
## Veeam
Драйвер Vitastor QEMU имеет функцию, которая позволяет обманывать сторонние системы типа Veeam, которые
не могут сами по себе разобрать адреса дисков в vitastor: [qemu_file_mirror_path](../config/client.ru.md#qemu_file_mirror_path).
Чтобы заставить такие системы работать, вам нужно установить эту опцию равной пути к некоторому каталогу
в ФС (например, `/mnt/vitastor/`) и примонтировать этот каталог с помощью [`vitastor-nfs mount --block`](../usage/nfs.ru.md).
Они начнут обращаться к образам как к файлам и, вероятно, смогут заработать корректно :).
+116
View File
@@ -0,0 +1,116 @@
[Documentation](../../README.md#documentation) → Usage → UBLK
-----
[Читать на русском](ublk.ru.md)
# UBLK
[ublk](https://docs.kernel.org/block/ublk.html) is a new io_uring-based Linux interface
for user-space block device drivers, available since Linux 6.0.
It's not zero-copy, but it's still a fast implementation, outperforming both [NBD](nbd.en.md)
and [VDUSE](qemu.en.md#vduse) iops-wise and may or may not outperform VDUSE in linear I/O MB/s.
ublk also allows to recover devices even if the server (vitastor-ublk process) dies.
## Example performance comparison
TCP (100G), 3 hosts each with 6 NVMe OSDs, 3 replicas, single client
| | direct fio | NBD | VDUSE | UBLK |
|----------------------|-------------|-------------|------------|-------------|
| linear write | 3807 MB/s | 1832 MB/s | 3226 MB/s | 3027 MB/s |
| linear read | 3067 MB/s | 1885 MB/s | 1800 MB/s | 2076 MB/s |
| 4k random write Q128 | 128624 iops | 91060 iops | 94621 iops | 149450 iops |
| 4k random read Q128 | 117769 iops | 153408 iops | 93157 iops | 171987 iops |
| 4k random write Q1 | 8090 iops | 6442 iops | 6316 iops | 7272 iops |
| 4k random read Q1 | 9474 iops | 7200 iops | 6840 iops | 8038 iops |
RDMA (100G), 3 hosts each with 6 NVMe OSDs, 3 replicas, single client
| | direct fio | NBD | VDUSE | UBLK |
|----------------------|-------------|-------------|-------------|-------------|
| linear write | 6998 MB/s | 1878 MB/s | 4249 MB/s | 3140 MB/s |
| linear read | 8628 MB/s | 3389 MB/s | 5062 MB/s | 3674 MB/s |
| 4k random write Q128 | 222541 iops | 181589 iops | 138281 iops | 218222 iops |
| 4k random read Q128 | 412647 iops | 239987 iops | 151663 iops | 269583 iops |
| 4k random write Q1 | 11601 iops | 8592 iops | 9111 iops | 10000 iops |
| 4k random read Q1 | 10102 iops | 7788 iops | 8111 iops | 8965 iops |
## Commands
vitastor-ublk supports the following commands:
- [map](#map)
- [unmap](#unmap)
- [ls](#ls)
## map
To create a local block device for a Vitastor image run:
```
vitastor-ublk map [/dev/ublkbN] --image testimg
```
It will output a block device name like /dev/ublkb0 which you can then use as a normal disk.
You can also use `--pool <POOL> --inode <INODE> --size <SIZE>` instead of `--image <IMAGE>` if you want.
vitastor-ublk supports all usual Vitastor configuration options like `--config_path <path_to_config>` plus ublk-specific:
* `--recover` \
Recover a mapped device if the previous ublk server is dead.
* `--queue_depth 256` \
Maximum queue size for the device.
* `--max_io_size 1M` \
Maximum single I/O size for the device. Default: `max(1 MB, pool block size * EC part count)`.
* `--readonly` \
Make the device read-only.
* `--hdd` \
Mark the device as rotational.
* `--logfile /path/to/log/file.txt` \
Write log messages to the specified file instead of dropping them (in background mode)
or printing them to the standard output (in foreground mode).
* `--dev_num N` \
Use the specified device /dev/ublkbN instead of automatic selection (alternative syntax
to /dev/ublkbN positional parameter).
* `--foreground 1` \
Stay in foreground, do not daemonize.
Note that `ublk_queue_depth` and `ublk_max_io_size` may also be specified
in `/etc/vitastor/vitastor.conf` or in other configuration file specified with `--config_path`.
## unmap
To unmap the device run:
```
vitastor-ublk unmap /dev/ublkb0
```
## ls
```
vitastor-ublk ls [--json]
```
List mapped images.
Example output (normal format):
```
/dev/ublkb0
image: bench
pid: 584536
/dev/ublkb1
image: bench1
pid: 584546
```
Example output (JSON format):
```
{"/dev/ublkb0": {"image": "bench", "pid": 584536}, "/dev/ublkb1": {"image": "bench1", "pid": 584546}}
```
+121
View File
@@ -0,0 +1,121 @@
[Документация](../../README-ru.md#документация) → Использование → UBLK
-----
[Read in English](ublk.en.md)
# UBLK
[ublk](https://docs.kernel.org/block/ublk.html) - это новый Linux-интерфейс на основе io_uring
для реализации блочных устройств в пространстве пользователя, доступный, начиная с Linux 6.0.
ublk тоже копирует память (т.е. не является zero-copy), но по IOPS всё равно обгоняет и
[NBD](nbd.ru.md), и [VDUSE](qemu.ru.md#vduse), и иногда может даже обгонять VDUSE по
скорости линейного доступа. Также ublk позволяет оживлять устройства, у которых умер
сервер (процесс-обработчик vitastor-ublk).
## Пример сравнения производительности
TCP (100G), 3 сервера с 6 NVMe OSD каждый, 3 реплики, один клиент
| | Прямой fio | NBD | VDUSE | UBLK |
|--------------------------|-------------|-------------|------------|-------------|
| линейная запись | 3807 MB/s | 1832 MB/s | 3226 MB/s | 3027 MB/s |
| линейное чтение | 3067 MB/s | 1885 MB/s | 1800 MB/s | 2076 MB/s |
| 4k случайная запись Q128 | 128624 iops | 91060 iops | 94621 iops | 149450 iops |
| 4k случайное чтение Q128 | 117769 iops | 153408 iops | 93157 iops | 171987 iops |
| 4k случайная запись Q1 | 8090 iops | 6442 iops | 6316 iops | 7272 iops |
| 4k случайное чтение Q1 | 9474 iops | 7200 iops | 6840 iops | 8038 iops |
RDMA (100G), 3 сервера с 6 NVMe OSD каждый, 3 реплики, один клиент
| | Прямой fio | NBD | VDUSE | UBLK |
|--------------------------|-------------|-------------|-------------|-------------|
| линейная запись | 6998 MB/s | 1878 MB/s | 4249 MB/s | 3140 MB/s |
| линейное чтение | 8628 MB/s | 3389 MB/s | 5062 MB/s | 3674 MB/s |
| 4k случайная запись Q128 | 222541 iops | 181589 iops | 138281 iops | 218222 iops |
| 4k случайное чтение Q128 | 412647 iops | 239987 iops | 151663 iops | 269583 iops |
| 4k случайная запись Q1 | 11601 iops | 8592 iops | 9111 iops | 10000 iops |
| 4k случайное чтение Q1 | 10102 iops | 7788 iops | 8111 iops | 8965 iops |
## Команды
vitastor-ublk поддерживает следующие команды:
- [map](#map)
- [unmap](#unmap)
- [ls](#ls)
## map
Чтобы создать локальное блочное устройство для образа, выполните команду:
```
vitastor-ublk map [/dev/ublkbN] --image testimg
```
Команда напечатает название блочного устройства вида /dev/ublkb0, которое потом можно
будет использовать как обычный диск.
Для обращения по номеру инода, аналогично другим командам, можно использовать опции
`--pool <POOL> --inode <INODE> --size <SIZE>` вместо `--image testimg`.
vitastor-ublk поддерживает все обычные опции Vitastor, например, `--config_path <path_to_config>`,
плюс специфичные для ublk:
* `--recover` \
Восстановить ранее подключённое устройство, у которого умер обработчик.
* `--queue_depth 256` \
Максимальная глубина очереди устройства.
* `--max_io_size 1M` \
Максимальный размер запроса ввода-вывода для устройства. По умолчанию: `max(1 MB, блок данных пула * число частей данных EC)`.
* `--readonly` \
Подключить устройство в режиме только для чтения.
* `--hdd` \
Пометить устройство как вращающийся жёсткий диск (флаг rotational).
* `--logfile /path/to/log/file.txt` \
Писать сообщения о процессе работы в заданный файл, вместо пропуска их
при фоновом режиме запуска или печати на стандартный вывод при запуске
в консоли с `--foreground 1`.
* `--dev_num N` \
Использовать заданное устройство `/dev/ublkbN` вместо автоматического подбора.
* `--foreground 1` \
Не уводить процесс в фоновый режим.
Обратите внимание, что опции `ublk_queue_depth` и `ublk_max_io_size` можно
также задавать в `/etc/vitastor/vitastor.conf` или в другом файле конфигурации,
заданном опцией `--config_path`.
## unmap
Для отключения устройства выполните:
```
vitastor-ublk unmap /dev/ublkb0
```
## ls
```
vitastor-ublk ls [--json]
```
Вывести подключённые устройства.
Пример вывода в обычном формате:
```
/dev/ublkb0
image: bench
pid: 584536
/dev/ublkb1
image: bench1
pid: 584546
```
Пример вывода в JSON-формате:
```
{"/dev/ublkb0": {"image": "bench", "pid": 584536}, "/dev/ublkb1": {"image": "bench1", "pid": 584546}}
```
+1 -1
Submodule json11 updated: fd37016cf8...edcd85b8bd
+1 -1
View File
@@ -16,7 +16,7 @@ async function create_http_server(cfg, handler)
};
if (cfg.mon_https_ca)
{
tls.mon_https_ca = await fsp.readFile(cfg.mon_https_ca);
tls.ca = await fsp.readFile(cfg.mon_https_ca);
}
if (cfg.mon_https_client_auth)
{
+5 -5
View File
@@ -253,7 +253,7 @@ function random_custom_combinations(osd_tree, rules, count, ordered)
for (let i = 1; i < rules.length; i++)
{
const filtered = filter_tree_by_rules(osd_tree, rules[i], selected);
const idx = select_murmur3(filtered.length, i => 'p:'+f.id+':'+filtered[i].id);
const idx = select_murmur3(filtered.length, i => 'p:'+f.id+':'+(filtered[i].name || filtered[i].id));
selected.push(idx == null ? { levels: {}, id: null } : filtered[idx]);
}
const size = selected.filter(s => s.id !== null).length;
@@ -270,7 +270,7 @@ function random_custom_combinations(osd_tree, rules, count, ordered)
for (const item_rules of rules)
{
const filtered = selected.length ? filter_tree_by_rules(osd_tree, item_rules, selected) : first;
const idx = select_murmur3(filtered.length, i => n+':'+filtered[i].id);
const idx = select_murmur3(filtered.length, i => n+':'+(filtered[i].name || filtered[i].id));
selected.push(idx == null ? { levels: {}, id: null } : filtered[idx]);
}
const size = selected.filter(s => s.id !== null).length;
@@ -340,9 +340,9 @@ function filter_tree_by_rules(osd_tree, rules, selected)
}
// Convert from
// node_list = { id: string|number, level: string, size?: number, parent?: string|number }[]
// node_list = { id: string|number, name?: string, level: string, size?: number, parent?: string|number }[]
// to
// node_tree = { [node_id]: { id, level, size?, parent?, children?: child_node_id[], levels: { [level]: id, ... } } }
// node_tree = { [node_id]: { id, name?, level, size?, parent?, children?: child_node[], levels: { [level]: id, ... } } }
function index_tree(node_list)
{
const tree = { '': { children: [], levels: {} } };
@@ -357,7 +357,7 @@ function index_tree(node_list)
tree[parent_id].children = tree[parent_id].children || [];
tree[parent_id].children.push(tree[node.id]);
}
const cur = tree[''].children;
const cur = [ ...tree[''].children ];
for (let i = 0; i < cur.length; i++)
{
cur[i].levels[cur[i].level] = cur[i].id;
+244
View File
@@ -0,0 +1,244 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
// Extract OSDs from the lowest affected tree level into a separate (flat) map
// to run PG optimisation on failure domains instead of individual OSDs
//
// node_list = same input as for index_tree()
// rules = [ level, operator, value ][][]
// returns { nodes: new_node_list, leaves: { new_folded_node_id: [ extracted_leaf_nodes... ] } }
function fold_failure_domains(node_list, rules)
{
const interest = {};
for (const level_rules of rules)
{
for (const rule of level_rules)
interest[rule[0]] = true;
}
const max_numeric_id = node_list.reduce((a, c) => a < (0|c.id) ? (0|c.id) : a, 0);
let next_id = max_numeric_id;
const node_map = node_list.reduce((a, c) => { a[c.id||''] = c; return a; }, {});
const old_ids_by_new = {};
const extracted_nodes = {};
let folded = true;
while (folded)
{
const per_parent = {};
for (const node_id in node_map)
{
const node = node_map[node_id];
const p = node.parent || '';
per_parent[p] = per_parent[p]||[];
per_parent[p].push(node);
}
folded = false;
for (const node_id in per_parent)
{
const fold_node = node_id !== '' && per_parent[node_id].length > 0 && per_parent[node_id].filter(child => per_parent[child.id||''] || interest[child.level]).length == 0;
if (fold_node)
{
const old_node = node_map[node_id];
const new_id = ++next_id;
node_map[new_id] = {
...old_node,
id: new_id,
name: node_id, // for use in murmur3 hashes
size: per_parent[node_id].reduce((a, c) => a + (Number(c.size)||0), 0),
};
delete node_map[node_id];
old_ids_by_new[new_id] = node_id;
extracted_nodes[new_id] = [];
for (const child of per_parent[node_id])
{
if (old_ids_by_new[child.id])
{
extracted_nodes[new_id].push(...extracted_nodes[child.id]);
delete extracted_nodes[child.id];
}
else
extracted_nodes[new_id].push(child);
delete node_map[child.id];
}
folded = true;
}
}
}
return { nodes: Object.values(node_map), leaves: extracted_nodes };
}
// Distribute PGs mapped to "folded" nodes to individual OSDs according to their weights
// folded_pgs = optimize_result.int_pgs before folding
// prev_pgs = optional previous PGs from optimize_change() input
// extracted_nodes = output from fold_failure_domains
function unfold_failure_domains(folded_pgs, prev_pgs, extracted_nodes)
{
const maps = {};
let found = false;
for (const new_id in extracted_nodes)
{
const weights = {};
for (const sub_node of extracted_nodes[new_id])
{
weights[sub_node.id] = sub_node.size;
}
maps[new_id] = { weights, prev: [], next: [], pos: 0 };
found = true;
}
if (!found)
{
return folded_pgs;
}
for (let i = 0; i < folded_pgs.length; i++)
{
for (let j = 0; j < folded_pgs[i].length; j++)
{
if (maps[folded_pgs[i][j]])
{
maps[folded_pgs[i][j]].prev.push(prev_pgs && prev_pgs[i] && prev_pgs[i][j] || 0);
}
}
}
for (const new_id in maps)
{
maps[new_id].next = adjust_distribution(maps[new_id].weights, maps[new_id].prev);
}
const mapped_pgs = [];
for (let i = 0; i < folded_pgs.length; i++)
{
mapped_pgs.push(folded_pgs[i].map(osd => (maps[osd] ? maps[osd].next[maps[osd].pos++] : osd)));
}
return mapped_pgs;
}
// Return the new array of items re-distributed as close as possible to weights in wanted_weights
// wanted_weights = { [key]: weight }
// cur_items = key[]
function adjust_distribution(wanted_weights, cur_items)
{
const item_map = {};
for (let i = 0; i < cur_items.length; i++)
{
const item = cur_items[i];
item_map[item] = (item_map[item] || { target: 0, cur: [] });
item_map[item].cur.push(i);
}
let total_weight = 0;
for (const item in wanted_weights)
{
total_weight += Number(wanted_weights[item]) || 0;
}
for (const item in wanted_weights)
{
const weight = wanted_weights[item] / total_weight * cur_items.length;
if (weight > 0)
{
item_map[item] = (item_map[item] || { target: 0, cur: [] });
item_map[item].target = weight;
}
}
const diff = (item) => (item_map[item].cur.length - item_map[item].target);
const most_underweighted = Object.keys(item_map)
.filter(item => item_map[item].target > 0)
.sort((a, b) => diff(a) - diff(b));
// Items with zero target weight MUST never be selected - remove them
// and remap each of them to a most underweighted item
for (const item in item_map)
{
if (!item_map[item].target)
{
const prev = item_map[item];
delete item_map[item];
for (const idx of prev.cur)
{
const move_to = most_underweighted[0];
item_map[move_to].cur.push(idx);
move_leftmost(most_underweighted, diff);
}
}
}
// Other over-weighted items are only moved if it improves the distribution
while (most_underweighted.length > 1)
{
const first = most_underweighted[0];
const last = most_underweighted[most_underweighted.length-1];
const first_diff = diff(first);
const last_diff = diff(last);
if (Math.abs(first_diff+1)+Math.abs(last_diff-1) < Math.abs(first_diff)+Math.abs(last_diff))
{
item_map[first].cur.push(item_map[last].cur.pop());
move_leftmost(most_underweighted, diff);
move_rightmost(most_underweighted, diff);
}
else
{
break;
}
}
const new_items = new Array(cur_items.length);
for (const item in item_map)
{
for (const idx of item_map[item].cur)
{
new_items[idx] = item;
}
}
return new_items;
}
function move_leftmost(sorted_array, diff)
{
// Re-sort by moving the leftmost item to the right if it changes position
const first = sorted_array[0];
const new_diff = diff(first);
let r = 0;
while (r < sorted_array.length-1 && diff(sorted_array[r+1]) <= new_diff)
r++;
if (r > 0)
{
for (let i = 0; i < r; i++)
sorted_array[i] = sorted_array[i+1];
sorted_array[r] = first;
}
}
function move_rightmost(sorted_array, diff)
{
// Re-sort by moving the rightmost item to the left if it changes position
const last = sorted_array[sorted_array.length-1];
const new_diff = diff(last);
let r = sorted_array.length-1;
while (r > 0 && diff(sorted_array[r-1]) > new_diff)
r--;
if (r < sorted_array.length-1)
{
for (let i = sorted_array.length-1; i > r; i--)
sorted_array[i] = sorted_array[i-1];
sorted_array[r] = last;
}
}
// map previous PGs to folded nodes
function fold_prev_pgs(pgs, extracted_nodes)
{
const unmap = {};
for (const new_id in extracted_nodes)
{
for (const sub_node of extracted_nodes[new_id])
{
unmap[sub_node.id] = new_id;
}
}
const mapped_pgs = [];
for (let i = 0; i < pgs.length; i++)
{
mapped_pgs.push(pgs[i].map(osd => (unmap[osd] || osd)));
}
return mapped_pgs;
}
module.exports = {
fold_failure_domains,
unfold_failure_domains,
adjust_distribution,
fold_prev_pgs,
};
+10 -5
View File
@@ -10,16 +10,19 @@ const NO_OSD = 'Z';
async function lp_solve(text)
{
const cp = child_process.spawn('lp_solve');
let stdout = '', stderr = '', finish_cb;
let stdout = '', stderr = '', finish_cb, finished = 0;
cp.stdout.on('data', buf => stdout += buf.toString());
cp.stderr.on('data', buf => stderr += buf.toString());
cp.on('exit', () => finish_cb && finish_cb());
cp.stdout.on('end', () => finish_cb());
cp.stderr.on('end', () => finish_cb());
cp.stdin.write(text);
cp.stdin.end();
if (cp.exitCode == null)
await new Promise(ok => (finish_cb = () =>
{
await new Promise(ok => finish_cb = ok);
}
finished++;
if (finished == 2)
ok();
}));
if (!stdout.trim())
{
return null;
@@ -98,6 +101,7 @@ async function optimize_initial({ osd_weights, combinator, pg_count, pg_size = 3
score: lp_result.score,
weights: lp_result.vars,
int_pgs,
pg_effsize,
space: eff * pg_effsize,
total_space: total_weight,
};
@@ -409,6 +413,7 @@ async function optimize_change({ prev_pgs: prev_int_pgs, osd_weights, combinator
int_pgs: new_pgs,
differs,
osd_differs,
pg_effsize,
space: pg_effsize * pg_list_space_efficiency(new_pgs, osd_weights, pg_minsize, parity_space),
total_space: total_weight,
};
+108
View File
@@ -0,0 +1,108 @@
// Copyright (c) Vitaliy Filippov, 2019+
// License: VNPL-1.1 (see README.md for details)
const assert = require('assert');
const { fold_failure_domains, unfold_failure_domains, adjust_distribution } = require('./fold.js');
const DSL = require('./dsl_pgs.js');
const LPOptimizer = require('./lp_optimizer.js');
const stableStringify = require('../stable-stringify.js');
async function run()
{
// Test run adjust_distribution
console.log('adjust_distribution');
const rand = [];
for (let i = 0; i < 100; i++)
{
rand.push(1 + Math.floor(10*Math.random()));
// or rand.push(0);
}
const adj = adjust_distribution({ 1: 1, 2: 1, 3: 1, 4: 1, 5: 1, 6: 1, 7: 1, 8: 1, 9: 1, 10: 1 }, rand);
//console.log(rand.join(' '));
console.log(rand.reduce((a, c) => { a[c] = (a[c]||0)+1; return a; }, {}));
//console.log(adj.join(' '));
console.log(adj.reduce((a, c) => { a[c] = (a[c]||0)+1; return a; }, {}));
console.log('Movement: '+rand.reduce((a, c, i) => a+(rand[i] != adj[i] ? 1 : 0), 0)+'/'+rand.length);
console.log('\nfold_failure_domains');
console.log(JSON.stringify(fold_failure_domains(
[
{ id: 1, level: 'osd', size: 1, parent: 'disk1' },
{ id: 2, level: 'osd', size: 2, parent: 'disk1' },
{ id: 'disk1', level: 'disk', parent: 'host1' },
{ id: 'host1', level: 'host', parent: 'dc1' },
{ id: 'dc1', level: 'dc' },
],
[ [ [ 'dc' ], [ 'host' ] ] ]
), 0, 2));
console.log('\nfold_failure_domains empty rules');
console.log(JSON.stringify(fold_failure_domains(
[
{ id: 1, level: 'osd', size: 1, parent: 'disk1' },
{ id: 2, level: 'osd', size: 2, parent: 'disk1' },
{ id: 'disk1', level: 'disk', parent: 'host1' },
{ id: 'host1', level: 'host', parent: 'dc1' },
{ id: 'dc1', level: 'dc' },
],
[]
), 0, 2));
console.log('\noptimize_folded');
// 5 DCs, 2 hosts per DC, 10 OSD per host
const nodes = [];
for (let i = 1; i <= 100; i++)
{
nodes.push({ id: i, level: 'osd', size: 1, parent: 'host'+(1+(0|((i-1)/10))) });
}
for (let i = 1; i <= 10; i++)
{
nodes.push({ id: 'host'+i, level: 'host', parent: 'dc'+(1+(0|((i-1)/2))) });
}
for (let i = 1; i <= 5; i++)
{
nodes.push({ id: 'dc'+i, level: 'dc' });
}
// Check rules
const rules = DSL.parse_level_indexes({ dc: '112233', host: '123456' }, [ 'dc', 'host', 'osd' ]);
assert.deepEqual(rules, [[],[["dc","=",1],["host","!=",[1]]],[["dc","!=",[1]]],[["dc","=",3],["host","!=",[3]]],[["dc","!=",[1,3]]],[["dc","=",5],["host","!=",[5]]]]);
// Check tree folding
const { nodes: folded_nodes, leaves: folded_leaves } = fold_failure_domains(nodes, rules);
const expected_folded = [];
const expected_leaves = {};
for (let i = 1; i <= 10; i++)
{
expected_folded.push({ id: 100+i, name: 'host'+i, level: 'host', size: 10, parent: 'dc'+(1+(0|((i-1)/2))) });
expected_leaves[100+i] = [ ...new Array(10).keys() ].map(k => ({ id: 10*(i-1)+k+1, level: 'osd', size: 1, parent: 'host'+i }));
}
for (let i = 1; i <= 5; i++)
{
expected_folded.push({ id: 'dc'+i, level: 'dc' });
}
assert.equal(stableStringify(folded_nodes), stableStringify(expected_folded));
assert.equal(stableStringify(folded_leaves), stableStringify(expected_leaves));
// Now optimise it
console.log('1000 PGs, EC 112233');
const leaf_weights = folded_nodes.reduce((a, c) => { if (Number(c.id)) { a[c.id] = c.size; } return a; }, {});
let res = await LPOptimizer.optimize_initial({
osd_weights: leaf_weights,
combinator: new DSL.RuleCombinator(folded_nodes, rules, 10000, false),
pg_size: 6,
pg_count: 1000,
ordered: false,
});
LPOptimizer.print_change_stats(res, false);
assert.equal(res.space, 100, 'Initial distribution');
const unfolded_res = { ...res };
unfolded_res.int_pgs = unfold_failure_domains(res.int_pgs, null, folded_leaves);
const osd_weights = nodes.reduce((a, c) => { if (Number(c.id)) { a[c.id] = c.size; } return a; }, {});
unfolded_res.space = unfolded_res.pg_effsize * LPOptimizer.pg_list_space_efficiency(unfolded_res.int_pgs, osd_weights, 0, 1);
LPOptimizer.print_change_stats(unfolded_res, false);
assert.equal(res.space, 100, 'Initial distribution');
}
run().catch(console.error);
+1
View File
@@ -96,6 +96,7 @@ class Mon
}
else
{
res.setHeader('Content-Type', 'text/plain; version=0.0.4; charset=utf-8');
res.write(export_prometheus_metrics(this.state));
}
}
+5 -5
View File
@@ -15,7 +15,7 @@ function get_osd_tree(global_config, state)
const stat = state.osd.stats[osd_num];
const osd_cfg = state.config.osd[osd_num];
let reweight = osd_cfg == null ? 1 : Number(osd_cfg.reweight);
if (reweight < 0 || isNaN(reweight))
if (isNaN(reweight) || reweight < 0 || reweight > 1)
reweight = 1;
if (stat && stat.size && reweight && (state.osd.state[osd_num] || Number(stat.time) >= down_time ||
osd_cfg && osd_cfg.noout))
@@ -87,7 +87,7 @@ function make_hier_tree(global_config, tree)
tree[''] = { children: [] };
for (const node_id in tree)
{
if (node_id === '' || tree[node_id].level === 'osd' && (!tree[node_id].size || tree[node_id].size <= 0))
if (node_id === '')
{
continue;
}
@@ -107,10 +107,10 @@ function make_hier_tree(global_config, tree)
deleted = 0;
for (const node_id in tree)
{
if (tree[node_id].level !== 'osd' && (!tree[node_id].children || !tree[node_id].children.length))
if (!(tree[node_id].children||[]).length && (tree[node_id].size||0) <= 0)
{
const parent = tree[node_id].parent;
if (parent)
if (parent && tree[parent])
{
tree[parent].children = tree[parent].children.filter(c => c != tree[node_id]);
}
@@ -179,7 +179,7 @@ function filter_osds_by_block_layout(orig_tree, osd_stats, block_size, bitmap_gr
if (orig_tree[osd].level === 'osd')
{
const osd_stat = osd_stats[osd];
if (osd_stat && (osd_stat.bs_block_size && osd_stat.bs_block_size != block_size ||
if (osd_stat && (osd_stat.data_block_size && osd_stat.data_block_size != block_size ||
osd_stat.bitmap_granularity && osd_stat.bitmap_granularity != bitmap_granularity ||
osd_stat.immediate_commit == 'small' && immediate_commit == 'all' ||
osd_stat.immediate_commit == 'none' && immediate_commit != 'none'))

Some files were not shown because too many files have changed in this diff Show More