issue-5293: fixed caching of stale GetNodeAttr responses in case of concurrent WriteData/SetNodeAttr/AllocateData requests by qkrorlqr · Pull Request #5308 · ydb-platform/nbs

qkrorlqr · 2026-02-24T14:07:54Z

Notes

introducing a per-fs atomic GlobalAttrVersion counter - incrementing it every time upon attr-modifying operation completion
invalidating node attr in TNodeCache upon each attr-modifying operation completion and setting the just-incremented GlobalAttrVersion as this node's version
loading current node attr version from TNodeCache upon attr-reading operation start
resetting attr_timeout in the attrs returned to the guest if upon attr-reading operation completion we find out that the version of the node in TNodeCache has changed since the start of the operation

extra:

moved TDirectoryBuilder to a separate .h/.cpp file pair and added unit tests for it
moved NodeCacheLock into TNodeCache so TNodeCache is now thread-safe
TNodeCache benchmark
implemented TNodeCache sharding by node id to reduce contention (the results can be checked via the added benchmark)

// 16 client threads, 1 shard (equivalent to non-sharded version)
----------- UpdateNode16 ---------------
 samples:       4364
 iterations:    11758825
 iterations hr:    11.8M
 run time:      5.000951105
 per iteration: 1235.532165 cycles

// 16 client threads, 16 shards
----------- UpdateNode16_16 ---------------
 samples:       10080
 iterations:    62736801
 iterations hr:    62.7M
 run time:      5.001321319
 per iteration: 156.7914719 cycles

Issue

#5293

github-actions · 2026-02-24T16:12:25Z

Note

This is an automated comment that will be appended during run.

🟢 linux-x86_64-relwithdebinfo target: cloud/filestore/ (test time: 6939s): all tests PASSED for commit 6a2ca31.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
3439	3439	0	0	0	0	0

github-actions · 2026-02-25T20:44:01Z

Note

This is an automated comment that will be appended during run.

🟢 linux-x86_64-release-asan target: cloud/filestore/ (test time: 5176s): all tests PASSED for commit 3b9e79d.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
2911	2911	0	0	0	0	0

github-actions · 2026-02-25T21:12:21Z

Note

This is an automated comment that will be appended during run.

🟢 linux-x86_64-release-msan target: cloud/filestore/ (test time: 5050s): all tests PASSED for commit 3b9e79d.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
2911	2911	0	0	0	0	0

github-actions · 2026-02-25T21:16:16Z

Note

This is an automated comment that will be appended during run.

🔴 linux-x86_64-relwithdebinfo target: cloud/filestore/ (test time: 7008s): some tests FAILED for commit 3b9e79d.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
3441	3440	0	1	0	0	0

🔴 linux-x86_64-relwithdebinfo target: cloud/filestore/ (test time: 184s): some tests FAILED for commit 3b9e79d.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
2	1	0	1	0	0	0

🔴 linux-x86_64-relwithdebinfo target: cloud/filestore/ (test time: 180s): some tests FAILED for commit 3b9e79d.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
2	1	0	1	0	0	0

github-actions · 2026-02-25T21:17:31Z

Note

This is an automated comment that will be appended during run.

🟢 linux-x86_64-release-tsan target: cloud/filestore/ (test time: 5460s): all tests PASSED for commit 3b9e79d.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
2901	2901	0	0	0	0	0

github-actions · 2026-02-25T21:18:52Z

Note

This is an automated comment that will be appended during run.

🟢 linux-x86_64-release-ubsan target: cloud/filestore/ (test time: 4967s): all tests PASSED for commit 3b9e79d.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
2942	2942	0	0	0	0	0

github-actions · 2026-02-27T01:04:37Z

Note

This is an automated comment that will be appended during run.

🔴 linux-x86_64-release-asan target: cloud/filestore/ (test time: 16962s): some tests FAILED for commit d9dfe9e.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
2796	2190	0	191	4	347	64

github-actions · 2026-02-27T01:36:42Z

Note

This is an automated comment that will be appended during run.

🔴 linux-x86_64-release-msan target: cloud/filestore/ (test time: 17153s): some tests FAILED for commit d9dfe9e.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
2796	2190	0	191	4	347	64

github-actions · 2026-02-27T01:38:08Z

Note

This is an automated comment that will be appended during run.

🔴 linux-x86_64-release-tsan target: cloud/filestore/ (test time: 17333s): some tests FAILED for commit d9dfe9e.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
2786	2179	0	192	4	347	64

github-actions · 2026-02-27T02:04:20Z

Note

This is an automated comment that will be appended during run.

🔴 linux-x86_64-release-ubsan target: cloud/filestore/ (test time: 18194s): some tests FAILED for commit d9dfe9e.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
2825	2202	0	209	4	346	64

github-actions · 2026-02-27T03:56:44Z

Note

This is an automated comment that will be appended during run.

🔴 linux-x86_64-relwithdebinfo target: cloud/filestore/ (test time: 27185s): some tests FAILED for commit d9dfe9e.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
3324	2483	0	298	4	475	64

🔴 linux-x86_64-relwithdebinfo target: cloud/filestore/ (test time: 23869s): some tests FAILED for commit d9dfe9e.

TESTS	PASSED	ERRORS	FAILED	FAILED BUILD	SKIPPED	MUTED^?
949	109	0	300	0	476	64

…oncurrent WriteData/SetNodeAttr/AllocateData requests

…ate only size + small TNode cleanup

…pair + added uts for it

… content chunks

…ationRace finally fixed

…ationRace test cleanup

…TNodeCache

…TNodeCache - fix

neihar · 2026-02-27T11:58:14Z

cloud/filestore/libs/vfs_fuse/fs_directory_content_format.cpp

+
+NProto::TError ResetAttrTimeout(
+    char* data,
+    ui64 len,


remainingLen

neihar · 2026-02-27T11:58:44Z

cloud/filestore/libs/vfs_fuse/fs_directory_content_format.cpp

+    const TNodeIdVisitor& visitor)
+{
+#if defined(FUSE_VIRTIO)
+    while (len > sizeof(fuse_direntplus)) {


probably add commit that expect no name with 0 lens

neihar · 2026-02-27T12:24:29Z

cloud/filestore/libs/vfs_fuse/fs_impl_list.cpp

-            content.GetSize());
+        TBuffer c(content.GetData(), content.GetSize());
+
+        ResetAttrTimeout(c.Data(), c.Size(), [&] (ui64 ino) {


will it work as expected for '.' and '..'

neihar · 2026-02-27T12:31:47Z

cloud/filestore/libs/vfs_fuse/fs_impl_data.cpp


        if (self->CheckResponse(self, *callContext, req, response)) {
+            //
+            // Disallow result caching for all concurrently running operations


should we take care about WriteBufLocal as well?

neihar · 2026-02-27T12:34:41Z

cloud/filestore/libs/vfs_fuse/fs_directory_handle.h

+    struct TContent
+    {
+        TBufferPtr Buffer;
+        ui64 AttrVersion = 0;


we do not serialize it, that means we will reset it, meaning, that after restart we will have stale caches. probably discussion is needed

neihar · 2026-02-27T12:48:01Z

cloud/filestore/libs/vfs_fuse/fs_impl.cpp

        entry.entry_timeout = GetEntryCacheTimeout(attrs).SecondsFloat();
-
-        ConvertAttr(Config->GetPreferredBlockSize(), node->Attrs, entry.attr);
+    } else {


not clear why, probably some comment here can help

qkrorlqr added large-tests Launch large tests for PR filestore Add this label to run only cloud/filestore build and tests on PR labels Feb 24, 2026

qkrorlqr added 13 commits February 27, 2026 11:36

issue-5293: fixed caching of stale GetNodeAttr responses in case of c…

874c84a

…oncurrent WriteData/SetNodeAttr/AllocateData requests

issue-5293: invalidating node attrs in cache instead of trying to upd…

eca9da6

…ate only size + small TNode cleanup

issue-5293: TNodeCache comment

6d6f048

issue-5293: ListDir vs attr modification race ut

410186f

issue-5293: added AttrVersion to TDirectoryContent

25023d1

issue-5293: extracted TDirectoryBuilder into a separate .h/.cpp file …

21d5e7e

…pair + added uts for it

issue-5293: a func for resetting attr_timeout in serialized directory…

dbe3c41

… content chunks

issue-5293: ResetAttrTimeout -> VisitEntries

2aafc01

issue-5293: back to ResetAttrTimeout - but with an ino filter

7213cab

issue-5293: ShouldNotAllowGuestAttrCachingUponListNodesAndAttrModific…

7fb83d2

…ationRace finally fixed

issue-5293: ShouldNotAllowGuestAttrCachingUponListNodesAndAttrModific…

8d1d314

…ationRace test cleanup

issue-5293: refactored TNodeCache iface and moved NodeCacheLock into …

40544b4

…TNodeCache

issue-5293: refactored TNodeCache iface and moved NodeCacheLock into …

b430327

…TNodeCache - fix

qkrorlqr force-pushed the users/qkrorlqr/fix-set-get-attr-race branch from d9dfe9e to b430327 Compare February 27, 2026 10:43

qkrorlqr added 3 commits February 27, 2026 13:09

issue-5293: TNodeCache bench

30776dd

issue-5293: TNodeCache sharding

639fc09

issue-5293: using 16 TNodeCache shards in fs_impl.cpp

59fc686

qkrorlqr marked this pull request as ready for review February 27, 2026 12:44

neihar reviewed Feb 27, 2026

View reviewed changes

Conversation

qkrorlqr commented Feb 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Notes

Issue

Uh oh!

github-actions bot commented Feb 24, 2026

Uh oh!

github-actions bot commented Feb 25, 2026

Uh oh!

github-actions bot commented Feb 25, 2026

Uh oh!

github-actions bot commented Feb 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Feb 25, 2026

Uh oh!

github-actions bot commented Feb 25, 2026

Uh oh!

github-actions bot commented Feb 27, 2026

Uh oh!

github-actions bot commented Feb 27, 2026

Uh oh!

github-actions bot commented Feb 27, 2026

Uh oh!

github-actions bot commented Feb 27, 2026

Uh oh!

github-actions bot commented Feb 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

neihar Feb 27, 2026

Choose a reason for hiding this comment

Uh oh!

neihar Feb 27, 2026

Choose a reason for hiding this comment

Uh oh!

neihar Feb 27, 2026

Choose a reason for hiding this comment

Uh oh!

neihar Feb 27, 2026

Choose a reason for hiding this comment

Uh oh!

neihar Feb 27, 2026

Choose a reason for hiding this comment

Uh oh!

neihar Feb 27, 2026

Choose a reason for hiding this comment

Uh oh!

neihar Feb 27, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

qkrorlqr commented Feb 24, 2026 •

edited

Loading

github-actions bot commented Feb 25, 2026 •

edited

Loading

github-actions bot commented Feb 27, 2026 •

edited

Loading