Details
-
Bug
-
Resolution: Fixed
-
Major
-
3.1.2
-
Security Level: Public
-
None
-
CBG Sprint 145
-
1
Description
There have been multiple reports of the import DCP feed failing to start when data has been migrated between clusters/environments, which is only fixed when Sync Gateway's DCP checkpoints are manually deleted.
This should be handled by the DCP rollback code. Even in the case of a vb UUID in our checkpoints not being part of the new cluster's history, we should roll back to zero and start fresh with empty metadata (i.e. the equivalent of manually deleting the checkpoints).
Need to attempt to repro this locally to understand the root cause. There are a few possibilities:
- We're not removing the invalid vb uuid from our metadata on rollback to zero
- The rollback process is deadlocking somewhere in this particular case
- The rollback process is deadlocking in all cases
Based on previous fixes I don't think #3 is likely, but needs investigation.