Consider the following scenario:
- Secondary schedules a refresh of the chunks on the primary
- Primary successfully writes the refresh locally
- Before this operation gets replicated and we advance the stable timestamp we stepDown the primary
As a result of this, since the new primary will not be inclusive of the refresh the returned opTime is completely meaninglesss since it will be associated to arolled back entry.
- is related to
-
SERVER-62987 Wrong replication logic on refreshes on secondary nodes
- Closed