liyinan926 edited a comment on issue #26687: [SPARK-30055][k8s] Allow 
configuration of restart policy for Kubernetes pods
URL: https://github.com/apache/spark/pull/26687#issuecomment-566844904
 
 
   My biggest concern is around changing the default `restartPolicy` of 
executor pods to `Always`, which doesn't sound appropriate for batch 
run-to-completion workloads like Spark jobs. By changing it to `Always`, you 
introduce two competing controls of executor pods: the driver that manages the 
lifecycle of executor pods, and kubelets that manage the restarts of them. This 
is particularly a problem if a user chooses to keep the executor pods around by 
setting `spark.kubernetes.executor.deleteOnTermination=false`. In this 
particular case, the executor pods will be restarted indefinitely because they 
will not be deleted upon completion. A default `restartPolicy` of `OnFailure` 
makes much more sense. With that being said, however, I do think that executor 
restart in this context is a concern of the Spark core and driver, instead of 
the kubelets.
   
   Also regarding the startup delay, the only thing you really save 
significantly is the latency introduced by the k8s scheduler by doing in-place 
restarts. We no longer use an init-container, so there's no latency of waiting 
for init-containers to complete.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to