When pod readiness is changing states in very short succession we are missing the ready state and are loosing all active endpoints and they are not reconciled anymore. Raising the duration of controller-manager flag --endpoint-updates-batch-period should mitigate this issue.
When pod readiness is changing states in very short succession we are missing the ready state and are loosing all active endpoints and they are not reconciled anymore. Raising the duration of controller-manager flag
--endpoint-updates-batch-periodshould mitigate this issue.Also see discussion at https://github.com/kubernetes/kubernetes/issues/117193