gracefulEvictionTasks can not be removed from rb when cluster returns to normal

[spiritNO1]

Please provide an in-depth description of the question you have:

My member cluster sometimes returns error when deal with readyz, so taint_manager patch gracefulEvictionTasks on all related rbs. But after a while the member cluster returns to ready, gracefulEvictionTasks shoule be removed from rb?

What do you think about this question?:

Environment:

• Karmada version:
• Kubernetes version:
• Others:

[spiritNO1]

@RainbowMango I think taint_manager should recover rbs after the cluster returns to ready. How do you think?

[spiritNO1]

Or I could turn off the Failover fature gate

[RainbowMango]

so taint_manager patch gracefulEvictionTasks on all related rbs.

What happened after the taint manager built the gracefulEvictionTask? Does the failover perform as expected?

[spiritNO1]

so taint_manager patch gracefulEvictionTasks on all related rbs.

What happened after the taint manager built the gracefulEvictionTask? Does the failover perform as expected?

The scheduler reschedule the rb, and since the only affinity cluster is evicted, the schedule result is empty. And the update is stuck.

[RainbowMango]

Well, I would say that's exactly the case we will improve. We hope to introduce pre-checks before performing eviction, for cases like this won't trigger failover.

Anyway, If you don't need the failover feature, you can always disable it.