Uploaded image for project: 'Spark'
  1. Spark
  2. SPARK-28488

Race in k8s scheduler shutdown can lead to misleading exceptions.

    XMLWordPrintableJSON

    Details

    • Type: Bug
    • Status: Open
    • Priority: Minor
    • Resolution: Unresolved
    • Affects Version/s: 3.0.0
    • Fix Version/s: None
    • Component/s: Kubernetes, Spark Core
    • Labels:
      None

      Description

      There's a race when shutting down the k8s scheduler backend that may cause ugly exceptions to show up in the logs:

      19/07/22 14:43:46 ERROR Utils: Uncaught exception in thread kubernetes-executor-snapshots-subscribers-0
      org.apache.spark.SparkException: Could not find CoarseGrainedScheduler.
              at org.apache.spark.rpc.netty.Dispatcher.postMessage(Dispatcher.scala:162)
              at org.apache.spark.rpc.netty.Dispatcher.postOneWayMessage(Dispatcher.scala:143)
              at org.apache.spark.rpc.netty.NettyRpcEnv.send(NettyRpcEnv.scala:193)
              at org.apache.spark.rpc.netty.NettyRpcEndpointRef.send(NettyRpcEnv.scala:537)
              at org.apache.spark.scheduler.cluster.CoarseGrainedSchedulerBackend.removeExecutor(CoarseGrainedSchedulerBackend.scala:509)
              at org.apache.spark.scheduler.cluster.k8s.KubernetesClusterSchedulerBackend.doRemoveExecutor(KubernetesClusterSchedulerBackend.scala:63)
              at org.apache.spark.scheduler.cluster.k8s.ExecutorPodsLifecycleManager.removeExecutorFromSpark(ExecutorPodsLifecycleManager.scala:143)
              at org.apache.spark.scheduler.cluster.k8s.ExecutorPodsLifecycleManager.$anonfun$onNewSnapshots$2(ExecutorPodsLifecycleManager.scala:64)
              at scala.collection.immutable.HashMap$HashMap1.foreach(HashMap.scala:234)
              at scala.collection.immutable.HashMap$HashTrieMap.foreach(HashMap.scala:465)
              at org.apache.spark.scheduler.cluster.k8s.ExecutorPodsLifecycleManager.$anonfun$onNewSnapshots$1(ExecutorPodsLifecycleManager.scala:59)
              at org.apache.spark.scheduler.cluster.k8s.ExecutorPodsLifecycleManager.$anonfun$onNewSnapshots$1$adapted(ExecutorPodsLifecycleManager.scala:58)
              at scala.collection.mutable.ResizableArray.foreach(ResizableArray.scala:62)
              at scala.collection.mutable.ResizableArray.foreach$(ResizableArray.scala:55)
              at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:49)
              at org.apache.spark.scheduler.cluster.k8s.ExecutorPodsLifecycleManager.onNewSnapshots(ExecutorPodsLifecycleManager.scala:58)
              at org.apache.spark.scheduler.cluster.k8s.ExecutorPodsLifecycleManager.$anonfun$start$1(ExecutorPodsLifecycleManager.scala:50)
              at org.apache.spark.scheduler.cluster.k8s.ExecutorPodsLifecycleManager.$anonfun$start$1$adapted(ExecutorPodsLifecycleManager.scala:50)
              at org.apache.spark.scheduler.cluster.k8s.ExecutorPodsSnapshotsStoreImpl.$anonfun$callSubscriber$1(ExecutorPodsSnapshotsStoreImpl.scala:110)
              at org.apache.spark.util.Utils$.tryLogNonFatalError(Utils.scala:1330)
              at org.apache.spark.scheduler.cluster.k8s.ExecutorPodsSnapshotsStoreImpl.org$apache$spark$scheduler$cluster$k8s$ExecutorPodsSnapshotsStoreImpl$$callSubscriber(ExecutorPodsSnapshotsStoreImpl.scala:107)
              at org.apache.spark.scheduler.cluster.k8s.ExecutorPodsSnapshotsStoreImpl$$anon$1.run(ExecutorPodsSnapshotsStoreImpl.scala:80)
              at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
              at java.util.concurrent.FutureTask.run(FutureTask.java:266)
              at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$201(ScheduledThreadPoolExecutor.java:180)
              at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:293)
              at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
              at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
              at java.lang.Thread.run(Thread.java:748)
      

      Basically, because the scheduler endpoint is shut down before the executors used internally by the spark-on-k8s code, those may send messages to an endpoint that does not exist anymore.

        Attachments

          Issue Links

            Activity

              People

              • Assignee:
                Unassigned
                Reporter:
                vanzin Marcelo Masiero Vanzin
              • Votes:
                0 Vote for this issue
                Watchers:
                2 Start watching this issue

                Dates

                • Created:
                  Updated: