Rolling update of flink cluster in kubernetes

[Suraj Puvvada]

Hello

Wanted to understand the best practices around running Flink in Kubernetes especially from a continuous deployment perspective.

Is it possible to do a rolling update ? 

Thanks

Suraj

[Yang Wang]

Hi Suraj,

AFAIK, it is not a good practice to rolling update the JobManager and TaskManagers. Since

every restarting of Pod will cause a failover of Flink job. Instead, i prefer to enable the HA configuration(e.g. zookeeper),

then you could completely delete the current deployment and start a new one, including JobManager and TaskManager.

All the jobs could recover from the latest checkpoint if the cluster id do not change.

You could also restart the Flink application from a savepoint without HA configuration.

Some K8s operators could make the upgrading of continuous Flink application easier[1][2].

[1]. https://github.com/lyft/flinkk8soperator

[2]. https://github.com/GoogleCloudPlatform/flink-on-k8s-operator

Best,

Yang

Suraj Puvvada <[hidden email]> 于2020年7月23日周四 上午1:30写道：

Hello

Wanted to understand the best practices around running Flink in Kubernetes especially from a continuous deployment perspective.

Is it possible to do a rolling update ? 

Thanks

Suraj