[KAFKA-9654] ReplicaAlterLogDirsThread can't be created again if the previous ReplicaAlterLogDirsThreadmeet encounters leader epoch error

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 2.4.2, 2.5.1
• Assignee: Chia-Ping Tsai
• Reporter: Chia-Ping Tsai
• Created: 04/Mar/20 21:10
• Updated: 20/Mar/20 22:02
• Resolved: 20/Mar/20 22:02

Description

ReplicaManager does create ReplicaAlterLogDirsThread only if an new future log is created. If the previous ReplicaAlterLogDirsThread encounters error when moving data, the target partition is moved to "failedPartitions" and ReplicaAlterLogDirsThread get idle due to empty partitions. The future log is still existent so we CAN'T either create another ReplicaAlterLogDirsThread to handle the parition or update the paritions of the idler ReplicaAlterLogDirsThread.

ReplicaManager should call ReplicaAlterLogDirsManager#addFetcherForPartitions even if there is already a future log since we can create an new ReplicaAlterLogDirsThread to handle the new partitions or update the partitions of existent ReplicaAlterLogDirsThread to make it busy again.

Issue Links

• is duplicated by: KAFKA-8001 AlterLogDirs: Fetch from future replica stalls when local replica becomes a leader
• links to: GitHub Pull Request #8223