Details
-
Bug
-
Resolution: Fixed
-
Major
-
6.5.0
-
Triaged
-
No
-
KV-Engine Mad-Hatter Beta
Description
Deduplication in a disk snapshot can lead to replicas receiving an abort, but finds the item in the hashtable is not pending.
Consider the active receives the following ops for a given key
PRE1 CMT1 PRE2 ABRT2
The replica connects, backfills from disk, and will see
SET ABRT2
When processing the abort in VBucket::abort, ht.findForWrite(key) will find the stored value, but it will not be pending - it will be the value stored by the SET.
The exception will be thrown from around kv_engine/engines/ep/src/vbucket.cc:912
This should only be possible when receiving a disk snapshot, so it could be special cased.
Attachments
Issue Links
- relates to
-
MB-34849 Not seeing any error messages in memcached logs after rebalance fails with mover crashed error.
- Closed