Actions
Bug #48363
openrgw: concurrent object delete can leak bucket index instance entries
% Done:
0%
Source:
Tags:
Backport:
Regression:
No
Severity:
3 - minor
Reviewed:
Affected Versions:
ceph-qa-suite:
Pull request ID:
Crash signature (v1):
Crash signature (v2):
Description
This happens when there are concurrent operations, where at least one succeeds, whereas at least another one is cancelled. Likely to also affect non-versioned objects, though cancellation might not happen as frequently there.
The bucket index entry will not be removed when the bucket index complete arrives, because of the pending entries. If the last pending entry remaining is cancelled (vs. completed), the entry will not be removed. With versioned objects it will only happen with instance entries that aren't part of the bucket listing. Non-versioned object entries won't show in listing either as they are marked with exists=false.
Actions