[FLINK-19806] Job may try to leave SUSPENDED state in ExecutionGraph#failJob()

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.10.0
• Fix Version/s: 1.11.3, 1.12.0
• Component/s: Runtime / Coordination
• Labels: pull-request-available
• Assignee: Zhu Zhu
• Reporter: Zhu Zhu
• Created: 26/Oct/20 06:44
• Updated: 17/Nov/20 10:05
• Resolved: 17/Nov/20 10:05

Description

SUSPENDED is a terminal state which a job is not supposed to leave this state once entering. However, ExecutionGraph#failJob() did not check it and may try to transition a job out from SUSPENDED state. This will cause unexpected errors and may lead to JM crash.
The problem can be visible if we rework ExecutionGraphSuspendTest to be based on DefaultScheduler.
We should harden the check in ExecutionGraph#failJob().

Issue Links

• links to: GitHub Pull Request #13790