Details
-
Bug
-
Resolution: Fixed
-
Major
-
master, 4.5.1, 4.6.5, 5.0.1, 5.1.1, 5.5.3, 6.0.1
-
None
-
Untriaged
-
No
Description
Summary
A customer has seen some very slow GET_REPLICA operations - in some cases completing after one hour.
Details
The reason is that the GET_REPLICA operation is received for a vbucket when it is in a pending state. This means that the operation is put into pending operations. We wake-up to a task to process the pending operations when moving to an active state - but not when moving to a replica state - see here . So when we move from pending to replica we do nothing. Later on we go back to a pending state and then to active at which point we process the pending operations, however significant time has past.
Need to follow-up with ns_server team to understand why we move from a pending to replica state. Also why the vbucket is being published in the vbucket map as being able to accept replica operations.
It is also good to understand why we staying in a pending state longer than would be expected (i.e. on the customer environment we are seeing ~3 minutes.)