Github user mccheah commented on a diff in the pull request:
https://github.com/apache/spark/pull/21366#discussion_r193933656
--- Diff:
resource-managers/kubernetes/core/src/main/scala/org/apache/spark/scheduler/cluster/k8s/ExecutorPodsLifecycleManager.scala
---
@@ -0,0 +1,146 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.spark.scheduler.cluster.k8s
+
+import com.google.common.cache.Cache
+import io.fabric8.kubernetes.api.model.Pod
+import io.fabric8.kubernetes.client.KubernetesClient
+import scala.collection.JavaConverters._
+import scala.collection.mutable
+
+import org.apache.spark.SparkConf
+import org.apache.spark.deploy.k8s.Config._
+import org.apache.spark.scheduler.ExecutorExited
+import org.apache.spark.util.Utils
+
+private[spark] class ExecutorPodsLifecycleManager(
+ conf: SparkConf,
+ executorBuilder: KubernetesExecutorBuilder,
+ kubernetesClient: KubernetesClient,
+ snapshotsStore: ExecutorPodsSnapshotsStore,
+ // Use a best-effort to track which executors have been removed
already. It's not generally
+ // job-breaking if we remove executors more than once but it's ideal
if we make an attempt
+ // to avoid doing so. Expire cache entries so that this data structure
doesn't grow beyond
+ // bounds.
+ removedExecutorsCache: Cache[java.lang.Long, java.lang.Long]) {
+
+ import ExecutorPodsLifecycleManager._
+
+ private val eventProcessingInterval =
conf.get(KUBERNETES_EXECUTOR_EVENT_PROCESSING_INTERVAL)
+
+ def start(schedulerBackend: KubernetesClusterSchedulerBackend): Unit = {
+ snapshotsStore.addSubscriber(eventProcessingInterval) {
+ onNextSnapshot(schedulerBackend, _)
+ }
+ }
+
+ private def onNextSnapshot(
+ schedulerBackend: KubernetesClusterSchedulerBackend,
+ snapshot: ExecutorPodsSnapshot): Unit = {
+ val execIdsRemovedInThisRound = mutable.HashSet.empty[Long]
+ snapshot.executorPods.foreach { case (execId, state) =>
+ state match {
+ case PodDeleted(pod) =>
+ removeExecutorFromSpark(schedulerBackend, pod, execId)
+ execIdsRemovedInThisRound += execId
+ case errorOrSucceeded @ (PodFailed(_) | PodSucceeded(_)) =>
+ removeExecutorFromK8s(errorOrSucceeded.pod)
+ removeExecutorFromSpark(schedulerBackend, errorOrSucceeded.pod,
execId)
+ execIdsRemovedInThisRound += execId
+ case _ =>
+ }
+ }
+
+ // Reconcile the case where Spark claims to know about an executor but
the corresponding pod
+ // is missing from the cluster. This would occur if we miss a deletion
event and the pod
+ // transitions immediately from running io absent.
+ (schedulerBackend.getExecutorIds().map(_.toLong).toSet
--- End diff --
Another case where the previous approach was insufficient. If Spark gets an
executor connected to it that is not in the current snapshot, then we should
probably get rid of it.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
