[
https://issues.apache.org/jira/browse/SPARK-54553?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wenjun Ruan updated SPARK-54553:
--------------------------------
Description:
I submit Spark jobs to Kubernetes using Volcano as the scheduler.
According to the documentation:
[https://spark.apache.org/docs/latest/running-on-kubernetes.html#volcano-feature-step]
I found that we should create a {{podGroupTemplate file first, and then using
it in spark job.}}
I hope we ca does not overwrite the PodGroup name when it is already defined in
the {{{}podGroupTemplateFile{}}}.
I can submit a PR to make this behavior configurable or to preserve the
PodGroup name if it is already provided in {{{}podGroupTemplateFile{}}}.
was:
I submit Spark jobs to Kubernetes using Volcano as the scheduler.
According to the documentation:
[https://spark.apache.org/docs/latest/running-on-kubernetes.html#volcano-feature-step]
I found that the actual PodGroup name is not the one I specified in the
{{{}podGroupTemplateFile{}}}. Instead, it is overwritten by
{{{}VolcanoFeatureStep{}}}.
I hope Spark does not overwrite the PodGroup name when it is already defined in
the {{{}podGroupTemplateFile{}}}.
Here is the relevant code:
```scala
override def getAdditionalPreKubernetesResources(): Seq[HasMetadata] = {
if (kubernetesConf.isInstanceOf[KubernetesExecutorConf]) {
logWarning(
"VolcanoFeatureStep#getAdditionalPreKubernetesResources() is not
supported for executor."
)
return Seq.empty
}
lazy val client = new DefaultVolcanoClient
val template = kubernetesConf.getOption(POD_GROUP_TEMPLATE_FILE_KEY)
val pg = template.map(client.podGroups.load(_).item).getOrElse(new PodGroup())
var metadata = pg.getMetadata
if (metadata == null) metadata = new ObjectMeta
metadata.setName(podGroupName)
metadata.setNamespace(namespace)
pg.setMetadata(metadata)
var spec = pg.getSpec
if (spec == null) spec = new PodGroupSpec
pg.setSpec(spec)
Seq(pg)
}
```
I can submit a PR to make this behavior configurable or to preserve the
PodGroup name if it is already provided in {{{}podGroupTemplateFile{}}}.
> Supports receiving podgroup JSON format configurations when using Volcano.
> --------------------------------------------------------------------------
>
> Key: SPARK-54553
> URL: https://issues.apache.org/jira/browse/SPARK-54553
> Project: Spark
> Issue Type: Improvement
> Components: Kubernetes
> Affects Versions: 4.0.1
> Reporter: Wenjun Ruan
> Priority: Major
>
> I submit Spark jobs to Kubernetes using Volcano as the scheduler.
> According to the documentation:
> [https://spark.apache.org/docs/latest/running-on-kubernetes.html#volcano-feature-step]
> I found that we should create a {{podGroupTemplate file first, and then using
> it in spark job.}}
> I hope we ca does not overwrite the PodGroup name when it is already defined
> in the {{{}podGroupTemplateFile{}}}.
>
> I can submit a PR to make this behavior configurable or to preserve the
> PodGroup name if it is already provided in {{{}podGroupTemplateFile{}}}.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
