[FLINK-34947] Only scale down JM in Foreground deletion propagation and reduce timeout

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: kubernetes-operator-1.8.0
• Fix Version/s: kubernetes-operator-1.9.0
• Component/s: Kubernetes Operator
• Labels: pull-request-available
• Assignee: Gyula Fora
• Reporter: Gyula Fora
• Created: 27/Mar/24 06:45
• Updated: 15/Apr/24 11:50
• Resolved: 15/Apr/24 11:50

Description

We introduced a logic to scale down the JobManager before the task managers are killed to have a more graceful shutdown sequence.

Currently this is always done in native mode, but it makes more sense to restrict this functionality to foreground deletion. Foreground deletion is the default and recommended propagation but in some cases users may want to use background deletion to speed up the process. In these cases we should not force a synchronous, potentially slow step.

Furthermore instead of using the entire deletion timeout we should reduce this to only use a portion of it as it's an optional step and we should always have keep enough time for the TM shutdown.

Issue Links

• links to: GitHub Pull Request #806