[
https://issues.apache.org/jira/browse/SPARK-39965?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
pralabhkumar updated SPARK-39965:
---------------------------------
Description:
>From Spark32 . as a part of [https://github.com/apache/spark/pull/32288] ,
>functionality is added to delete PVC if the Spark driver died.
[https://github.com/apache/spark/blob/786a70e710369b195d7c117b33fe9983044014d6/resource-managers/kubernetes/core/src/main/scala/org/apache/spark/scheduler/cluster/k8s/KubernetesClusterSchedulerBackend.scala#L144]
However there are cases , where spark on K8s doesn't use PVC and use host path
for storage.
[https://spark.apache.org/docs/latest/running-on-kubernetes.html#using-kubernetes-volumes]
Now even in those cases , it request to delete PVC (which is not required) .
Moreover in the cluster , where Spark user doesn't have access to list or
delete PVC , it throws exception .
io.fabric8.kubernetes.client.KubernetesClientException: Failure executing: GET
at:
[https://kubernetes.default.svc/api/v1/namespaces/<>/persistentvolumeclaims?labelSelector=spark-app-selector%3Dspark-332bd09284b3442f8a6a214fabcd6ab1|https://kubernetes.default.svc/api/v1/namespaces/dpi-dev/persistentvolumeclaims?labelSelector=spark-app-selector%3Dspark-332bd09284b3442f8a6a214fabcd6ab1].
Message: Forbidden!Configured service account doesn't have access. Service
account may have been revoked. persistentvolumeclaims is forbidden: User
"system:serviceaccount:dpi-dev:spark" cannot list resource
"persistentvolumeclaims" in API group "" in the namespace "<>".
Ideally there should be configuration
spark.kubernetes.driver.pvc.deleteOnTermination which should be checked before
calling to delete PVC. If user have not set up PV then there is no need to call
the api and delete PVC .
was:
In org.apache.spark.util getConfiguredLocalDirs
{code:java}
if (isRunningInYarnContainer(conf)) {
// If we are in yarn mode, systems can have different disk layouts so we must
set it
// to what Yarn on this system said was available. Note this assumes that
Yarn has
// created the directories already, and that they are secured so that only the
// user has access to them.
randomizeInPlace(getYarnLocalDirs(conf).split(","))
} else if (conf.getenv("SPARK_EXECUTOR_DIRS") != null) {
conf.getenv("SPARK_EXECUTOR_DIRS").split(File.pathSeparator)
} else if (conf.getenv("SPARK_LOCAL_DIRS") != null) {
conf.getenv("SPARK_LOCAL_DIRS").split(",")
}{code}
randomizedInplace is not called conf.getenv("SPARK_LOCAL_DIRS").split(",") .
This is what used in case of K8s and the shuffle locations are not randomized.
IMHO , this should be randomized , so that all the directories have equal
changes of pushing the data as was done on yarn side
> Spark on K8s delete pvc even though it's not being used.
> --------------------------------------------------------
>
> Key: SPARK-39965
> URL: https://issues.apache.org/jira/browse/SPARK-39965
> Project: Spark
> Issue Type: Bug
> Components: Kubernetes
> Affects Versions: 3.3.0
> Reporter: pralabhkumar
> Priority: Minor
>
> From Spark32 . as a part of [https://github.com/apache/spark/pull/32288] ,
> functionality is added to delete PVC if the Spark driver died.
> [https://github.com/apache/spark/blob/786a70e710369b195d7c117b33fe9983044014d6/resource-managers/kubernetes/core/src/main/scala/org/apache/spark/scheduler/cluster/k8s/KubernetesClusterSchedulerBackend.scala#L144]
>
> However there are cases , where spark on K8s doesn't use PVC and use host
> path for storage.
> [https://spark.apache.org/docs/latest/running-on-kubernetes.html#using-kubernetes-volumes]
>
> Now even in those cases , it request to delete PVC (which is not required) .
> Moreover in the cluster , where Spark user doesn't have access to list or
> delete PVC , it throws exception .
>
> io.fabric8.kubernetes.client.KubernetesClientException: Failure executing:
> GET at:
> [https://kubernetes.default.svc/api/v1/namespaces/<>/persistentvolumeclaims?labelSelector=spark-app-selector%3Dspark-332bd09284b3442f8a6a214fabcd6ab1|https://kubernetes.default.svc/api/v1/namespaces/dpi-dev/persistentvolumeclaims?labelSelector=spark-app-selector%3Dspark-332bd09284b3442f8a6a214fabcd6ab1].
> Message: Forbidden!Configured service account doesn't have access. Service
> account may have been revoked. persistentvolumeclaims is forbidden: User
> "system:serviceaccount:dpi-dev:spark" cannot list resource
> "persistentvolumeclaims" in API group "" in the namespace "<>".
>
> Ideally there should be configuration
> spark.kubernetes.driver.pvc.deleteOnTermination which should be checked
> before calling to delete PVC. If user have not set up PV then there is no
> need to call the api and delete PVC .
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
