[FLINK-11719] Remove JobMaster#start(JobMasterId) and #suspend

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.8.0
• Fix Version/s: 1.13.0
• Component/s: Runtime / Coordination
• Labels: pull-request-available
• Assignee: Till Rohrmann
• Reporter: Till Rohrmann
• Created: 21/Feb/19 16:00
• Updated: 25/Dec/20 01:31
• Resolved: 22/Dec/20 13:55

Description

Currently, the JobMaster contains a lot of mutable state which is necessary because it is used across different leadership sessions by the JobManagerRunner. For this purpose, we have the methods JobMaster#start(JobMasterId) and #suspend. The mutable state management makes things on the JobMaster side more complicated than they need to be. In order to improve the JobMaster's maintainability I suggest to remove this logic and instead terminate the JobMaster if the JobManagerRunner loses leadership. This entails that for every leadership we will create a new JobMaster instance.

Issue Links

• blocks: FLINK-11749 Returns the stored leadership info on ZooKeeperLeaderRetrievalService started
• fixes: FLINK-20648 Unable to restore job from savepoint when using Kubernetes based HA services
• links to: GitHub Pull Request #14431