New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
release-24.1: sql: fix leak in memory accounting around TxnFingerprintIDCache #121873
Conversation
This commit is partial revert of 88ebd70. Until that change, we attempted to perform memory accounting for txn fingeprint IDs stored in the cache for each session, but we never initialized the bytes monitor, so it didn't actually count towards the root SQL memory budget. In that change, we derived an account from the "session" monitor to fix that. However, this exposed another problem with how accounting was done: namely, on `Cache.Add` call we always grow the account and on `Cache.OnEvicted` we shrink the account. The problem is that if the txn fingerprint ID already exists in the cache, we still call `Cache.Add` (growing the account) but we will never shrink it because we didn't add a new entry. As a result, until the session is closed, we'll keep on accumulating the leak. The fix in this commit is simple - just remove any attempt for memory accounting for this txn fingerprint ID cache. Effectively, this brings us back to the state of how we were before the change mentioned above (no accounting done) without the overhead of creating redundant BytesMonitor / BoundAccount objects (since they served no real purpose). Not having any accounting done for this cache seems acceptable given that the cache stores up to 100 txns (by default), and each txn results in about 56B of usage, so we'll have about 5KiB of unaccounted for (per session) memory usage. We have much larger omissions elsewhere, so for now I left a TODO to add memory accounting in the future. Release note (bug fix): CockroachDB could previously "leak" reported memory usage (as accounted by the internal memory accounting system, the limit for which is configured via --max-sql-memory flag) on long-running sessions that issue many (hundreds of thousands or more) transactions. This, in turn, could result in "root: memory budget exceeded" errors for other queries. The bug is present in versions 23.1.17 and 23.2.3 and is now fixed.
d1e953a
to
86d4d9f
Compare
Thanks for opening a backport. Please check the backport criteria before merging:
If your backport adds new functionality, please ensure that the following additional criteria are satisfied:
Also, please add a brief release justification to the body of your PR to justify this |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Reviewed 4 of 4 files at r1, all commit messages.
Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @DrewKimball)
Backport 1/1 commits from #121847 on behalf of @yuzefovich.
/cc @cockroachdb/release
This commit is partial revert of 88ebd70. Until that change, we attempted to perform memory accounting for txn fingeprint IDs stored in the cache for each session, but we never initialized the bytes monitor, so it didn't actually count towards the root SQL memory budget. In that change, we derived an account from the "session" monitor to fix that.
However, this exposed another problem with how accounting was done: namely, on
Cache.Add
call we always grow the account and onCache.OnEvicted
we shrink the account. The problem is that if the txn fingerprint ID already exists in the cache, we still callCache.Add
(growing the account) but we will never shrink it because we didn't add a new entry. As a result, until the session is closed, we'll keep on accumulating the leak.The fix in this commit is simple - just remove any attempt for memory accounting for this txn fingerprint ID cache. Effectively, this brings us back to the state of how we were before the change mentioned above (no accounting done) without the overhead of creating redundant BytesMonitor / BoundAccount objects (since they served no real purpose). Not having any accounting done for this cache seems acceptable given that the cache stores up to 100 txns (by default), and each txn results in about 56B of usage, so we'll have about 5KiB of unaccounted for (per session) memory usage. We have much larger omissions elsewhere, so for now I left a TODO to add memory accounting in the future.
Addresses: #121844.
Epic: None
Release note (bug fix): CockroachDB could previously "leak" reported memory usage (as accounted by the internal memory accounting system, the limit for which is configured via --max-sql-memory flag) on long-running sessions that issue many (hundreds of thousands or more) transactions. This, in turn, could result in "root: memory budget exceeded" errors for other queries. The bug is present in versions 23.1.17 and 23.2.3 and is now fixed.
Release justification: bug fix.