[FLINK-12219] Yarn application can't stop when flink job failed in per-job yarn cluster mode

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.6.3, 1.8.0
• Fix Version/s: 1.7.3, 1.8.1, 1.9.0
• Component/s: Deployment / YARN, Runtime / REST
• Labels: pull-request-available
• Assignee: lamber-ken
• Reporter: lamber-ken
• Due: 17/Apr/19
• Created: 17/Apr/19 07:03
• Updated: 07/May/19 15:27
• Resolved: 07/May/19 15:27

Description

Issue detail info

In our flink(1.6.3) product env, I often encounter a scene that yarn application can't stop when flink job failed in per-job yarn cluste mode, so I deeply analyzed the reason why it happened.

When a flink job fail, system will write an archive file to a FileSystem through MiniDispatcher#archiveExecutionGraph method, then notify YarnJobClusterEntrypoint to shutDown. But, if MiniDispatcher#archiveExecutionGraph throw exceptions during execution, it affect the following calls.

So I open FLINK-12247 to solve NEP bug when system write archive to FileSystem. But We still need to consider other exceptions, so we should catch Exception / Throwable not just IOExcetion.

Flink yarn job fail flow

Flink yarn job fail on yarn

 

Flink yarn application can't stop

 

 

Issue Links

• links to: GitHub Pull Request #8254
• links to: GitHub Pull Request #8334