Hi team, we are running our flink job with Kubernetes HA services with 2 replicas for JobManager. If we kill a JM pod, we are seeing a full job restart with 20-30 seconds downtime. Is that an expected behavior? I was expecting when one JM is killed second JM will take control immediately. Is there any config that you can suggest for a faster recovery? We are using S3 for HA storage. Thanks!

The JM is a coordinator, so when that becomes unavailable, there's a need to restart from the last snapshot to make sure that the defined guarantees can be achieved.