[FLINK-23300] Job fails very slow because of no notifyAllocationFailure for DeclarativeSlotManager

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 1.13.1
• Component/s: Runtime / Task
• Assignee: Unassigned
• Reporter: Liu
• Created: 07/Jul/21 09:52
• Updated: 08/Jul/21 03:19
• Resolved: 08/Jul/21 03:19

Description

When container is killed, flink on yarn can detect the problem very quickly. But when using default DeclarativeSlotManager, notifyAllocationFailure is not called and the task is not failed until heartbeat is timeout. So the failover will be very slow.