Details
-
Improvement
-
Status: Closed
-
Blocker
-
Resolution: Fixed
-
None
-
None
-
None
-
Reviewed
Description
I was doing maintenance on clusters and rebooting a replication slave was giving me a lot of issues because the master wasn't able to read .META. while it was bombarded with getClosestRowBefore() coming from all the ReplicationSinks. Since no regions are assigned, and it doesn't retry, it goes straight back to the source which retries again.
Also it seems that in some cases with retries=1 we don't even recover from a stale .META. cache.