Details
-
Bug
-
Resolution: Fixed
-
Critical
-
2.0-beta
-
Security Level: Public
-
None
Description
1. Some data is loaded to master
2. This data is replicated
4. Now master crashes
5. It's disk has older version of data than replica.
6. Master is restarted. It recovers data up to persisted checkpoint. It's open checkpoint can now be 1 behind open checkpoint of replica.
7. Master receives new mutations and closes it's open checkpoint and opens next one.
8. Replication stream is reopened. And both master and replica are not aware that master's closed checkpoint is completely different than replica's.
Note that replica has some mutations master does not have and vice versa.
Similar situation may happen when we swap 2 replicas with same open checkpoint in replication chain and do failover. It's possible that replica will have more data then new master.