[
https://issues.apache.org/jira/browse/SPARK-29771?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jackey Lee updated SPARK-29771:
-------------------------------
Description:
ExecutorPodsAllocator does not limit the number of executor errors or
deletions, which may cause executor restart continuously without application
failure.
A simple example for this, add {{--conf spark.executor.extraJavaOptions=-Xmse}}
after spark-submit, which can make executor restart thousands of times without
application failure.
was:
At present, K8S scheduling does not limit the number of failures of the
executors, which may cause executor retried continuously without failing.
A simple example, we add a resource limit on default namespace. After the
driver is started, if the quota is full, the executor will retry the creation
continuously, resulting in a large amount of pod information accumulation. When
many applications encounter such situations, they will affect the K8S cluster.
> Limit executor max failures before failing the application
> ----------------------------------------------------------
>
> Key: SPARK-29771
> URL: https://issues.apache.org/jira/browse/SPARK-29771
> Project: Spark
> Issue Type: Improvement
> Components: Kubernetes
> Affects Versions: 3.0.0
> Reporter: Jackey Lee
> Priority: Major
>
> ExecutorPodsAllocator does not limit the number of executor errors or
> deletions, which may cause executor restart continuously without application
> failure.
> A simple example for this, add {{--conf
> spark.executor.extraJavaOptions=-Xmse}} after spark-submit, which can make
> executor restart thousands of times without application failure.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
