Skip to content

[SPARK-56238][K8S] Fix app ID propagation in KubernetesClusterSchedulerBackend for client mode submission #55355

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
xiaoxuandev wants to merge 2 commits into from
+37 −4
Closed

[SPARK-56238][K8S] Fix app ID propagation in KubernetesClusterSchedulerBackend for client mode submission #55355

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2aa37a0
[SPARK-56238][K8S] Fix app ID propagation in KubernetesClusterSchedul…
xiaoxuandev Apr 15, 2026
e2d5ad8
remove unrelated test
xiaoxuandev Apr 15, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 5 additions & 4 deletions ...main/scala/org/apache/spark/scheduler/cluster/k8s/KubernetesClusterSchedulerBackend.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,9 @@ private[spark] class KubernetesClusterSchedulerBackend(

private val minRegisteredExecutors = initialExecutors * minRegisteredRatio

private val appId: String =
conf.getOption("spark.app.id").getOrElse(KubernetesConf.getKubernetesAppId())

private val namespace = conf.get(KUBERNETES_NAMESPACE)

private val PATCH_CONTEXT = PatchContext.of(PatchType.STRATEGIC_MERGE)
Expand Down Expand Up @@ -98,13 +101,11 @@ private[spark] class KubernetesClusterSchedulerBackend(
/**
* Get an application ID associated with the job.
* This returns the string value of spark.app.id if set, otherwise
* the locally-generated ID.
* a generated Kubernetes app ID.
*
* @return The application ID
*/
override def applicationId(): String = {
conf.getOption("spark.app.id").getOrElse(KubernetesConf.getKubernetesAppId())
}
override def applicationId(): String = appId

override def start(): Unit = {
super.start()
Expand Down
32 changes: 32 additions & 0 deletions ...scala/org/apache/spark/scheduler/cluster/k8s/KubernetesClusterSchedulerBackendSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -314,4 +314,36 @@ class KubernetesClusterSchedulerBackendSuite extends SparkFunSuite with BeforeAn
endpoint.receiveAndReply(context).apply(GenerateExecID("cheeseBurger"))
verify(context).reply("1")
}

test("SPARK-56238: applicationId() is stable across calls when spark.app.id is not set") {
Copy link
Copy Markdown
Member

@dongjoon-hyun dongjoon-hyun Apr 15, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This test case seems to reproduce the reported scenario.

// Use isolated mocks so we don't mutate the shared sc/rpcEnv state.
val confWithoutAppId = new SparkConf(false)
.set("spark.executor.instances", "3")
.set(KUBERNETES_EXECUTOR_DECOMMISSION_LABEL.key, "soLong")
.set(KUBERNETES_EXECUTOR_DECOMMISSION_LABEL_VALUE.key, "cruelWorld")
val localSc = mock(classOf[SparkContext])
val localEnv = mock(classOf[SparkEnv])
val localRpcEnv = mock(classOf[RpcEnv])
when(localSc.conf).thenReturn(confWithoutAppId)
when(localSc.env).thenReturn(localEnv)
when(localSc.resourceProfileManager).thenReturn(resourceProfileManager)
when(localEnv.rpcEnv).thenReturn(localRpcEnv)
when(localRpcEnv.setupEndpoint(any(), any())).thenReturn(driverEndpointRef)
val localTaskScheduler = mock(classOf[TaskSchedulerImpl])
when(localTaskScheduler.sc).thenReturn(localSc)
val backendWithoutAppId = new KubernetesClusterSchedulerBackend(
Copy link
Copy Markdown
Member

@dongjoon-hyun dongjoon-hyun Apr 15, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do you happen to know when this situation happens in the production environment, @xiaoxuandev ? I'm wondering if this is a valid case in the Apache Spark usage.

This only affects client mode.

One more question. Do you know if this is a regression or not? (as Enrico claims)

Copy link
Copy Markdown
Contributor Author

@xiaoxuandev xiaoxuandev Apr 15, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, this is a regression introduced by #54269. The original code cached the generated ID in private val appId, which was correct.

Affected versions: v4.2.0-preview3+. All 4.0.x and 4.1.x releases are clean.

Regarding production usage: this affects Kubernetes client mode (--deploy-mode client), where the driver runs outside the K8s cluster. In that path, spark.app.id is not pre-set before backend.start(), so each call to applicationId() during start() would generate a different UUID.

Copy link
Copy Markdown
Member

@dongjoon-hyun dongjoon-hyun Apr 15, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Got it. Thank you for confirming, @xiaoxuandev .

localTaskScheduler,
localSc,
kubernetesClient,
schedulerExecutorService,
eventQueue,
podAllocator,
lifecycleManager,
watchEvents,
pollEvents)
val id1 = backendWithoutAppId.applicationId()
val id2 = backendWithoutAppId.applicationId()
assert(id1 === id2, "applicationId() must return the same value on repeated calls")
assert(id1.startsWith("spark-"), "generated app ID should have the spark- prefix")
}
}