kv: declare write access to AbortSpan on all aborting EndTxn reqs#48245
Merged
craig[bot] merged 1 commit intocockroachdb:masterfrom Apr 30, 2020
Merged
kv: declare write access to AbortSpan on all aborting EndTxn reqs#48245craig[bot] merged 1 commit intocockroachdb:masterfrom
craig[bot] merged 1 commit intocockroachdb:masterfrom
Conversation
Fixes cockroachdb#43707. Fixes cockroachdb#48046. Fixes cockroachdb#48189. Part of the change made by cockroachdb#42765 was to clear AbortSpan entries on non-poisoning, aborting EndTxn requests. Specifically, this change was made in 1328787. The change forgot to update the corresponding span declaration logic to reflect the fact that we were now writing to the AbortSpan in cases where we previously weren't. This was triggering an assertion in race builds that tried to catch this kind of undeclared span access. The assertion failure was very rare because it required the following conditions to all be met: 1. running a test with the race detector enabled 2. a txn (A) must have been aborted by another txn (B) 3. txn B must have cleared an intent on txn A's transaction record range 4. txn A must have noticed and issued a non-poisoning EndTxn(ABORT) We should backport this when we get a change (once v20.1.0 has stabilized), but I don't expect that this could actually cause any issues. The AbortSpan update was strictly a matter of performance and we should never be racing with another request that is trying to read the same AbortSpan entry.
Member
|
This change is |
andreimatei
reviewed
Apr 30, 2020
complete! 0 of 0 LGTMs obtained (waiting on
Contributor
Author
|
bors r+ |
Contributor
Build succeeded |
andreimatei
added a commit
to andreimatei/cockroach
that referenced
this pull request
May 1, 2020
We had an assertion that only a TransactionAbortedError can carry an ABORTED transaction. This assertion seems pretty dubious though - we've just seen failed rollbacks trigger it. While that was a bug - and in fact we caught it due to this assertion (see cockroachdb#48245), it seems hard to ensure that rollbacks will not fail in the future. Release note: None
andreimatei
added a commit
to andreimatei/cockroach
that referenced
this pull request
May 4, 2020
We had an assertion that only a TransactionAbortedError can carry an ABORTED transaction. This assertion seems pretty dubious though - we've just seen failed rollbacks trigger it. While that was a bug - and in fact we caught it due to this assertion (see cockroachdb#48245), it seems hard to ensure that rollbacks will not fail in the future. Release note: None
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Fixes #43707.
Fixes #48046.
Fixes #48189.
Part of the change made by #42765 was to clear AbortSpan entries on
non-poisoning, aborting EndTxn requests. Specifically, this change was
made in 1328787. The change forgot to update the corresponding span
declaration logic to reflect the fact that we were now writing to the
AbortSpan in cases where we previously weren't.
This was triggering an assertion in race builds that tried to catch this
kind of undeclared span access. The assertion failure was very rare
because it required the following conditions to all be met:
We should backport this when we get a change (once v20.1.0 has
stabilized), but I don't expect that this could actually cause any
issues. The AbortSpan update was strictly a matter of performance and we
should never be racing with another request that is trying to read the
same AbortSpan entry.