[
https://issues.apache.org/jira/browse/SPARK-55555?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dongjoon Hyun resolved SPARK-55555.
-----------------------------------
Fix Version/s: 4.2.0
Resolution: Fixed
> Support heterogeneous K8s executor management
> ---------------------------------------------
>
> Key: SPARK-55555
> URL: https://issues.apache.org/jira/browse/SPARK-55555
> Project: Spark
> Issue Type: Umbrella
> Components: Kubernetes
> Affects Versions: 4.2.0
> Reporter: Dongjoon Hyun
> Assignee: Dongjoon Hyun
> Priority: Critical
> Fix For: 4.2.0
>
>
> Historically, the Apache Spark manages executors in the well-known ways via
> (1) Dynamic Allocation, (2) Resource profiles, (3) Executor Rolling. However,
> many code patch still assumes homogeneous executors.
> This issue aims to cover more advanced resource managements by taking
> advantage of new K8s features. For example,
> - We can resize CPU and Memory requests and limits of executors via *In-Place
> Pod Update* dynamically and systematically in advance.
> - We can scale-up the PVC volumes via *AllowVolumeExpansion*.
> - We can introduce a new kind of executor type, *Recovery-mode Executor*, to
> allow only a single task per executor JVM in case of OOM.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
