[FLINK-11400] JobManagerRunner does not wait for suspension of JobMaster

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.6.3, 1.7.1, 1.8.0
• Fix Version/s: 1.8.0
• Component/s: Runtime / Coordination
• Labels: pull-request-available
• Assignee: Till Rohrmann
• Reporter: Till Rohrmann
• Created: 21/Jan/19 15:53
• Updated: 30/Jan/19 11:13
• Resolved: 30/Jan/19 11:12

Description

The JobManagerRunner does not wait for the suspension of the JobMaster to finish before granting leadership again. This can lead to a state where the JobMaster tries to start the ExecutionGraph but the SlotPool is still stopped.

I suggest to linearize the leadership operations (granting and revoking leadership) similarly to the Dispatcher.

Issue Links

• links to: GitHub Pull Request #7565