[FLINK-17672] OperatorCoordinators receive failure notifications on task failure instead of on task restarts

• Type: Sub-task
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.11.0
• Component/s: Runtime / Coordination
• Assignee: Stephan Ewen
• Reporter: Stephan Ewen
• Created: 13/May/20 14:34
• Updated: 16/May/20 16:36
• Resolved: 16/May/20 16:35

Description

Currently, the OperatorCoordinators receive failure notifications on task restart. That follows the same approach as the InputSplit assigners from the legacy sources (after which the integration of the Coordinators with the Scheduler was modeled).

However, propagating the failure notifications during the actual failure is more intuitive, and also improve situations where tasks fail but don't get restarted for a while (this can happen for batch tasks when a TM dies and no spare resources are available). In those cases, the coordinator can react much earlier to the failure.