[FLINK-20249] Rethink the necessity of the k8s internal Service even in non-HA mode

• Type: Improvement
• Status: Closed
• Priority: Minor
• Resolution: Won't Do
• Affects Version/s: 1.11.0
• Fix Version/s: 1.13.0
• Component/s: Deployment / Kubernetes
• Labels: pull-request-available
• Assignee: Unassigned
• Reporter: Ruguo Yu
• Created: 19/Nov/20 13:55
• Updated: 15/Jan/21 10:54
• Resolved: 15/Jan/21 10:54

Description

In non-HA mode, k8s will create internal service that directs the communication from TaskManagers Pod to JobManager Pod, and TM Pods could re-register to the new JM Pod once a JM Pod failover occurs.

However recently I do an experiment and find a problem that k8s will first create new TM pods and then destory old TM pods after a period of time once JM Pod failover (note: new JM podIP has changed), then job will be reschedule by JM on new TM pods, it means new TM has been registered to JM. 

During this process, internal service is active all the time, but I think it is not necessary that keep this internal service, In other words, wo can weed out internal service and use JM podIP for TM pods communication with JM pod, In this case, it be consistent with HA mode.

Finally，related experiments is in attached (k8s internal service.pdf).

Issue Links

• is related to: FLINK-20332 Add workers recovered from previous attempt to pending resources