Details
-
Bug
-
Resolution: Fixed
-
Test Blocker
-
6.5.0
-
Triaged
-
No
-
KV-Engine Mad-Hatter Beta
Description
If we disconnect we may receive two commits/aborts (or some combination of the two) in a row. Currently for EP Buckets this would cause an exception to throw and the DCP connection to be torn down as we would not find the existing prepare in the hash table to set the prepared seqno of the commit/abort. We need to be resilient to this and allow multiple commits/aborts. We only wish to allow this for the specific disconnect/reconnect case though. Believe it's enough to make a check such that if we receive a commit/abort with current highSeqno < seqno < streamStart then we should allow the "duplicate".
Attachments
Issue Links
- has to be done after
-
MB-34685 Allow prepares and commits in the same checkpoint
- Closed
- is duplicated by
-
MB-34450 Rebalance exited with reason mover_crashed as memcached throws "Ckpt::queueDirty() status:failure:duplicate item when vbstate is non-active:2"
- Closed
-
MB-34537 [Durability] Rebalance failed with "mover crashed" in bucket_replica update after sync_writes
- Closed
-
MB-36319 [Volume Test]: Rebalance In failed during durable doc-ops while there are ~90M durable docs already present in bucket which is at 90% DGM
- Closed
- relates to
-
MB-34412 Replica is not resilient to receiving multiple Prepares for the same key
- Closed
-
MB-34474 Replica is not resilient to receiving multiple Aborts for the same key
- Closed