admission: lock work queue before reading waiting length #131109
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
kvoli
added
backport-23.2.x
![blathers[crl]](https://avatars.githubusercontent.com/in/59700?s=40&u=373d9101a12edc3847d2fb6ae78e3adcfb4ca41e&v=4){kind=avatar}
|
It looks like your PR touches production code but doesn't add or edit any test code. Did you consider adding tests to your PR? 🦉 Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf. |
|
This change is |
kvoli
force-pushed
the
240920.kvadmission-fix-vlog-race
branch
from
d5f5c7c to
68e39b4
Compare
sumeerbhola
approved these changes
Sep 24, 2024
There was a problem hiding this comment.
Reviewed 1 of 1 files at r1.
Reviewable status:complete! 1 of 0 LGTMs obtained (waiting on @kvoli and @pav-kv)
pkg/util/admission/work_queue.go line 926 at r1 (raw file):
if log.V(1) { log.Infof(q.ambientCtx, "async-path: len(waiting-work)=%d dequeued t%d pri=%s r%s origin=n%s log-position=%s ingested=%t", tenant.waitingWorkHeap.Len(),
can you fix this one too, since q.mu.Unlock() has already been called.
When replicated work is submitted for admission, it returns early and admission proceeds asynchronously to the caller. When `V(1)` is enabled, we also log the current queue length in this code path, which is protected by a mutex that wasn't acquired previously. Acquire the queue mutex when `V(1)` is enabled to prevent a race. Part of: cockroachdb#130187 Release note: None
kvoli
force-pushed
the
240920.kvadmission-fix-vlog-race
branch
from
68e39b4 to
2eca0bc
Compare
kvoli
commented
Sep 24, 2024
There was a problem hiding this comment.
TYFTR!
Reviewable status:
pkg/util/admission/work_queue.go line 926 at r1 (raw file):
Previously, sumeerbhola wrote…
can you fix this one too, since
q.mu.Unlock()has already been called.
Done.
|
bors r=sumeerbhola |
craig bot
pushed a commit
that referenced
this pull request
Sep 25, 2024
130728: kvserver: add rac2 v1 integration tests r=sumeerbhola a=kvoli 1st commit from #130619. 2nd-3rd commits from #131106. 4th-5th commits from #131107. 6th-7th commits from #131108. 8th commit from #131109. --- Introduce several tests in `flow_control_integration_test.go`, mirroring the existing tests but applied to the replication flow control v2 machinery. The tests largely follow an identical pattern to the existing v1 tests, swapping in rac2 metrics and vtables. The following tests are added: ``` TestFlowControlBasicV2 TestFlowControlRangeSplitMergeV2 TestFlowControlBlockedAdmissionV2 TestFlowControlAdmissionPostSplitMergeV2 TestFlowControlCrashedNodeV2 TestFlowControlRaftSnapshotV2 TestFlowControlRaftMembershipV2 TestFlowControlRaftMembershipRemoveSelfV2 TestFlowControlClassPrioritizationV2 TestFlowControlQuiescedRangeV2 TestFlowControlUnquiescedRangeV2 TestFlowControlTransferLeaseV2 TestFlowControlLeaderNotLeaseholderV2 TestFlowControlGranterAdmitOneByOneV2 ``` These tests all have at least two variants: ``` V2EnabledWhenLeaderV1Encoding V2EnabledWhenLeaderV2Encoding ``` When `V2EnabledWhenLeaderV1Encoding` is run, the tests use a different testdata file, which has a `_v1_encoding` suffix. A separate file is necessary because when the protocol enablement level is `V2EnabledWhenLeaderV1Encoding`, all entries which are subject to admission control are encoded as `raftpb.LowPri`, regardless of their original priority, as we don't want to pay the cost to deserialize the raft admission meta. The v1 encoding variants retain the same comments as the v2 encoding, however any comments referring to regular tokens should be interpreted as elastic tokens instead, due to the above. Two v1 tests are not ported over to v2: ``` TestFlowControlRaftTransportBreak TestFlowControlRaftTransportCulled ``` These omitted tests behave identically to `TestFlowControlCrashedNodeV2` as rac2 is less tightly coupled to the raft transport, instead operating on replication states (e.g., `StateProbe`, `StateReplicate`). --- Add `TestFlowControlV1ToV2Transition`, which ratchets up the enabled version of replication flow control v2: ``` v1 protocol with v1 encoding => v2 protocol with v1 encoding => v2 protocol with v2 encoding ``` The test is structured to issue writes and wait for returned tokens whenever the protocol transitions from v1 to v2, or a leader changes. More specifically, the test takes the following steps: ``` (1) Start n1, n2, n3 with v1 protocol and v1 encoding. (2) Upgrade n1 to v2 protocol with v1 encoding. (3) Transfer the range lease to n2. (4) Upgrade n2 to v2 protocol with v1 encoding. (5) Upgrade n3 to v2 protocol with v1 encoding. (6) Upgrade n1 to v2 protocol with v2 encoding. (7) Transfer the range lease to n1. (8) Upgrade n2,n3 to v2 protocol with v2 encoding. (9) Transfer the range lease to n3. ``` Between each step, we issue writes, (un)block admission and observe the flow control metrics and vtables. Resolves: #130431 Resolves: #129276 Release note: None 131252: roachtest: port decommission/mixed-versions r=srosenberg,DarrylWong a=renatolabs This commit ports the `decommission/mixed-versions` roachtest to use the `mixedversion` framework (instead of the old `newUpgradeTest` API). It also updates `acceptance/decommission-self` since both tests used shared functionality that needed to be updated. Prior to this commit, the acceptance test used the old upgrade test API even though it was not an upgrade test. Fixes: #110531 Fixes: #110530 Release note: None 131364: upgrades: give test an additional core under remote exec r=rail a=rickystewart This has been timing out. Epic: none Release note: None Co-authored-by: Austen McClernon <[email protected]> Co-authored-by: Renato Costa <[email protected]> Co-authored-by: Ricky Stewart <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When replicated work is submitted for admission, it returns early and
admission proceeds asynchronously to the caller. When
V(1)is enabled,we also log the current queue length in this code path, which is
protected by a mutex that wasn't acquired previously.
Acquire the queue mutex when
V(1)is enabled to prevent a race.Part of: #130187
Release note: None