Details
-
Bug
-
Resolution: Unresolved
-
Major
-
7.1.3
-
None
-
Untriaged
-
0
-
Unknown
Description
In a customer setup, the following issue has been observed:
a. KV node has failed over
b. This KV node hosted ephemeral bucket. Fail over has triggered rollback on indexer
c. Timekeeper::handleStats is trying to get seqnos. from memcached but the get seqnos. call is timing out because of KV node fail over
d. During this 2 minute window, indexer has changed the rollback timestamp but the change is not propagated to client as timekeeper is stuck
All scans in this 2 minute window have failed with rollback time mismatch error. Since KV node failover leads to rollback on all index replicas in this call, non of the replicas could serve the scans leading to service un-availability