[KAFKA-10144] Corrupted standby tasks are not always cleaned up

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Fix Version/s: 2.6.0
• Component/s: streams
• Assignee: A. Sophie Blee-Goldman
• Reporter: A. Sophie Blee-Goldman
• Created: 11/Jun/20 02:21
• Updated: 13/Jun/20 01:07
• Resolved: 13/Jun/20 01:07

Description

Thread death on the 2.6-eos-beta soak was due to re-registration of a standby task changelog that was already registered. The root cause was that the task had been marked corrupted, but `commit` threw a TaskMigratedException before we could get to calling TaskManager#handleCorruption and properly clean up the task.

For corrupted active tasks this is not a problem, since #handleLostAll will then finish the cleanup. But we intentionally don't clear standbys tasks on TaskMigratedException, leaving the task corrupted and partially registered

Issue Links

• links to: GitHub Pull Request #8849