[
https://issues.apache.org/jira/browse/SPARK-26386?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Nihar Sheth resolved SPARK-26386.
---------------------------------
Resolution: Not A Problem
Attempted reinvention of "spark.kubernetes.driver.pod.name", which handles this
use case.
> [K8s] Add configuration for executor OwnerReference when running in client
> mode
> -------------------------------------------------------------------------------
>
> Key: SPARK-26386
> URL: https://issues.apache.org/jira/browse/SPARK-26386
> Project: Spark
> Issue Type: Improvement
> Components: Kubernetes
> Affects Versions: 3.0.0
> Reporter: Nihar Sheth
> Priority: Major
>
> When running in cluster mode, the executor pods have their owner references
> set to the driver pod, which should handle cleanup when the driver pod dies.
> In client mode, though, the driver is not always a pod, so there is no owner
> reference set at all. It would be useful to offer a configuration to support
> setting an executor's owner reference in client mode, for situations where we
> want the executors' lifecycles tied to another pod (for example, if there's a
> pod on the cluster that wants to run a Spark job in client mode on the same
> cluster, and we want to clean up the executors if the driver pod here dies).
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
