Details
-
Sub-task
-
Status: Resolved
-
Minor
-
Resolution: Duplicate
-
None
-
None
-
None
-
None
Description
copy the discussion (https://github.com/apache/kafka/pull/16614#discussion_r1683340380) to here
Consider the following sequence: we take a snapshot of the epoch entries here; a new epoch entry is added and is flushed to disk; the scheduler then writes the snapshot to disk. This can lead to the case where the leader epoch file doesn't contain all entries up to the recovery point.