[SPARK-30821][K8S] Handle executor failure with multiple containers

Handle executor failure with multiple containers

Added a spark property spark.kubernetes.executor.checkAllContainers,
with default being false. When it's true, the executor snapshot will
take all containers in the executor into consideration when deciding
whether the executor is in "Running" state, if the pod restart policy is
"Never". Also, added the new spark property to the doc.

### What changes were proposed in this pull request?

Checking of all containers in the executor pod when reporting executor status, if the `spark.kubernetes.executor.checkAllContainers` property is set to true.

### Why are the changes needed?

Currently, a pod remains "running" as long as there is at least one running container. This prevents Spark from noticing when a container has failed in an executor pod with multiple containers. With this change, user can configure the behavior to be different. Namely, if any container in the executor pod has failed, either the executor process or one of its sidecars, the pod is considered to be failed, and it will be rescheduled.

### Does this PR introduce _any_ user-facing change?

Yes, new spark property added.
User is now able to choose whether to turn on this feature using the `spark.kubernetes.executor.checkAllContainers` property.

### How was this patch tested?

Unit test was added and all passed.
I tried to run integration test by following the instruction [here](https://spark.apache.org/developer-tools.html) (section "Testing K8S") and also [here](https://github.com/apache/spark/blob/master/resource-managers/kubernetes/integration-tests/README.md), but I wasn't able to run it smoothly as it fails to talk with minikube cluster. Maybe it's because my minikube version is too new (I'm using v1.13.1)...? Since I've been trying it for two days and still can't make it work, I decided to submit this PR and hopefully the Jenkins test will pass.

Closes #29924 from huskysun/exec-sidecar-failure.

Authored-by: Shiqi Sun <s.sun@salesforce.com>
Signed-off-by: Holden Karau <hkarau@apple.com>

docs/running-on-kubernetes.md

@@ -1181,6 +1181,14 @@ See the [configuration page](configuration.html) for information on Spark config
   </td>
   <td>3.0.0</td>
 </tr>
+<tr>
+  <td><code>spark.kubernetes.executor.checkAllContainers</code></td>
+  <td>false</td>
+  <td>
+  Specify whether executor pods should be check all containers (including sidecars) or only the executor container when determining the pod status.
+  </td>
+  <td>3.1.0</td>
+</tr>
 <tr>
   <td><code>spark.kubernetes.submission.connectionTimeout</code></td>
   <td>10000</td>

resource-managers/kubernetes/core/src/main/scala/org/apache/spark/deploy/k8s/Config.scala

@@ -417,6 +417,14 @@ private[spark] object Config extends Logging {
       .stringConf
       .createOptional
 
+  val KUBERNETES_EXECUTOR_CHECK_ALL_CONTAINERS =
+    ConfigBuilder("spark.kubernetes.executor.checkAllContainers")
+      .doc("If set to true, all containers in the executor pod will be checked when reporting" +
+        "executor status.")
+      .version("3.1.0")
+      .booleanConf
+      .createWithDefault(false)
+
   val KUBERNETES_DRIVER_LABEL_PREFIX = "spark.kubernetes.driver.label."
   val KUBERNETES_DRIVER_ANNOTATION_PREFIX = "spark.kubernetes.driver.annotation."
   val KUBERNETES_DRIVER_SERVICE_ANNOTATION_PREFIX = "spark.kubernetes.driver.service.annotation."

resource-managers/kubernetes/core/src/main/scala/org/apache/spark/scheduler/cluster/k8s/ExecutorPodsSnapshot.scala

@@ -18,6 +18,7 @@ package org.apache.spark.scheduler.cluster.k8s
 
 import java.util.Locale
 
+import io.fabric8.kubernetes.api.model.ContainerStateTerminated
 import io.fabric8.kubernetes.api.model.Pod
 
 import org.apache.spark.deploy.k8s.Constants._
@@ -37,6 +38,7 @@ private[spark] case class ExecutorPodsSnapshot(executorPods: Map[Long, ExecutorP
 }
 
 object ExecutorPodsSnapshot extends Logging {
+  private var shouldCheckAllContainers: Boolean = _
 
   def apply(executorPods: Seq[Pod]): ExecutorPodsSnapshot = {
     ExecutorPodsSnapshot(toStatesByExecutorId(executorPods))
@@ -44,6 +46,10 @@ object ExecutorPodsSnapshot extends Logging {
 
   def apply(): ExecutorPodsSnapshot = ExecutorPodsSnapshot(Map.empty[Long, ExecutorPodState])
 
+  def setShouldCheckAllContainers(watchAllContainers: Boolean): Unit = {
+    shouldCheckAllContainers = watchAllContainers
+  }
+
   private def toStatesByExecutorId(executorPods: Seq[Pod]): Map[Long, ExecutorPodState] = {
     executorPods.map { pod =>
       (pod.getMetadata.getLabels.get(SPARK_EXECUTOR_ID_LABEL).toLong, toState(pod))
@@ -59,7 +65,15 @@ object ExecutorPodsSnapshot extends Logging {
         case "pending" =>
           PodPending(pod)
         case "running" =>
-          PodRunning(pod)
+          if (shouldCheckAllContainers &&
+            "Never" == pod.getSpec.getRestartPolicy &&
+            pod.getStatus.getContainerStatuses.stream
+              .map[ContainerStateTerminated](cs => cs.getState.getTerminated)
+              .anyMatch(t => t != null && t.getExitCode != 0)) {
+            PodFailed(pod)
+          } else {
+            PodRunning(pod)
+          }
         case "failed" =>
           PodFailed(pod)
         case "succeeded" =>

resource-managers/kubernetes/core/src/main/scala/org/apache/spark/scheduler/cluster/k8s/KubernetesClusterManager.scala

@@ -95,10 +95,13 @@ private[spark] class KubernetesClusterManager extends ExternalClusterManager wit
     val schedulerExecutorService = ThreadUtils.newDaemonSingleThreadScheduledExecutor(
       "kubernetes-executor-maintenance")
 
+    ExecutorPodsSnapshot.setShouldCheckAllContainers(
+      sc.conf.get(KUBERNETES_EXECUTOR_CHECK_ALL_CONTAINERS))
     val subscribersExecutor = ThreadUtils
       .newDaemonThreadPoolScheduledExecutor(
         "kubernetes-executor-snapshots-subscribers", 2)
     val snapshotsStore = new ExecutorPodsSnapshotsStoreImpl(subscribersExecutor)
+
     val removedExecutorsCache = CacheBuilder.newBuilder()
       .expireAfterWrite(3, TimeUnit.MINUTES)
       .build[java.lang.Long, java.lang.Long]()

resource-managers/kubernetes/core/src/test/scala/org/apache/spark/scheduler/cluster/k8s/DeterministicExecutorPodsSnapshotsStore.scala

@@ -21,6 +21,8 @@ import scala.collection.mutable
 
 class DeterministicExecutorPodsSnapshotsStore extends ExecutorPodsSnapshotsStore {
 
+  ExecutorPodsSnapshot.setShouldCheckAllContainers(false)
+
   private val snapshotsBuffer = mutable.Buffer.empty[ExecutorPodsSnapshot]
   private val subscribers = mutable.Buffer.empty[Seq[ExecutorPodsSnapshot] => Unit]
 

resource-managers/kubernetes/core/src/test/scala/org/apache/spark/scheduler/cluster/k8s/ExecutorLifecycleTestUtils.scala

@@ -76,6 +76,33 @@ object ExecutorLifecycleTestUtils {
       .build()
   }
 
+  /**
+   * [SPARK-30821]
+   * This creates a pod with one container in running state and one container in failed
+   * state (terminated with non-zero exit code). This pod is used for unit-testing the
+   * spark.kubernetes.executor.checkAllContainers Spark Conf.
+   */
+  def runningExecutorWithFailedContainer(executorId: Long): Pod = {
+    new PodBuilder(podWithAttachedContainerForId(executorId))
+      .editOrNewStatus()
+        .withPhase("running")
+        .addNewContainerStatus()
+          .withNewState()
+            .withNewTerminated()
+              .withExitCode(1)
+            .endTerminated()
+          .endState()
+        .endContainerStatus()
+        .addNewContainerStatus()
+          .withNewState()
+            .withNewRunning()
+            .endRunning()
+          .endState()
+        .endContainerStatus()
+      .endStatus()
+      .build()
+  }
+
   def succeededExecutor(executorId: Long): Pod = {
     new PodBuilder(podWithAttachedContainerForId(executorId))
       .editOrNewStatus()
@@ -117,7 +144,10 @@ object ExecutorLifecycleTestUtils {
         .addToLabels(SPARK_APP_ID_LABEL, TEST_SPARK_APP_ID)
         .addToLabels(SPARK_ROLE_LABEL, SPARK_POD_EXECUTOR_ROLE)
         .addToLabels(SPARK_EXECUTOR_ID_LABEL, executorId.toString)
-        .endMetadata()
+      .endMetadata()
+      .editOrNewSpec()
+        .withRestartPolicy("Never")
+      .endSpec()
       .build()
     val container = new ContainerBuilder()
       .withName("spark-executor")

resource-managers/kubernetes/core/src/test/scala/org/apache/spark/scheduler/cluster/k8s/ExecutorPodsSnapshotSuite.scala

@@ -16,31 +16,55 @@
  */
 package org.apache.spark.scheduler.cluster.k8s
 
+import io.fabric8.kubernetes.api.model.Pod
+
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.scheduler.cluster.k8s.ExecutorLifecycleTestUtils._
 
 class ExecutorPodsSnapshotSuite extends SparkFunSuite {
 
+  def testCase(pod: Pod, state: Pod => ExecutorPodState): (Pod, ExecutorPodState) =
+    (pod, state(pod))
+
+  def doTest(testCases: Seq[(Pod, ExecutorPodState)]): Unit = {
+    val snapshot = ExecutorPodsSnapshot(testCases.map(_._1))
+    for (((_, state), i) <- testCases.zipWithIndex) {
+      assertResult(state.getClass.getName, s"executor ID $i") {
+        snapshot.executorPods(i).getClass.getName
+      }
+    }
+  }
+
   test("States are interpreted correctly from pod metadata.") {
-    val pods = Seq(
-      pendingExecutor(0),
-      runningExecutor(1),
-      succeededExecutor(2),
-      failedExecutorWithoutDeletion(3),
-      deletedExecutor(4),
-      unknownExecutor(5))
-    val snapshot = ExecutorPodsSnapshot(pods)
-    assert(snapshot.executorPods ===
-      Map(
-        0L -> PodPending(pods(0)),
-        1L -> PodRunning(pods(1)),
-        2L -> PodSucceeded(pods(2)),
-        3L -> PodFailed(pods(3)),
-        4L -> PodDeleted(pods(4)),
-        5L -> PodUnknown(pods(5))))
+    ExecutorPodsSnapshot.setShouldCheckAllContainers(false)
+    val testCases = Seq(
+      testCase(pendingExecutor(0), PodPending),
+      testCase(runningExecutor(1), PodRunning),
+      testCase(succeededExecutor(2), PodSucceeded),
+      testCase(failedExecutorWithoutDeletion(3), PodFailed),
+      testCase(deletedExecutor(4), PodDeleted),
+      testCase(unknownExecutor(5), PodUnknown)
+    )
+    doTest(testCases)
+  }
+
+  test("SPARK-30821: States are interpreted correctly from pod metadata"
+    + " when configured to check all containers.") {
+    ExecutorPodsSnapshot.setShouldCheckAllContainers(true)
+    val testCases = Seq(
+      testCase(pendingExecutor(0), PodPending),
+      testCase(runningExecutor(1), PodRunning),
+      testCase(runningExecutorWithFailedContainer(2), PodFailed),
+      testCase(succeededExecutor(3), PodSucceeded),
+      testCase(failedExecutorWithoutDeletion(4), PodFailed),
+      testCase(deletedExecutor(5), PodDeleted),
+      testCase(unknownExecutor(6), PodUnknown)
+    )
+    doTest(testCases)
   }
 
   test("Updates add new pods for non-matching ids and edit existing pods for matching ids") {
+    ExecutorPodsSnapshot.setShouldCheckAllContainers(false)
     val originalPods = Seq(
       pendingExecutor(0),
       runningExecutor(1))

resource-managers/kubernetes/core/src/test/scala/org/apache/spark/scheduler/cluster/k8s/ExecutorPodsSnapshotsStoreSuite.scala

@@ -35,6 +35,7 @@ class ExecutorPodsSnapshotsStoreSuite extends SparkFunSuite with BeforeAndAfter
   before {
     eventBufferScheduler = new DeterministicScheduler()
     eventQueueUnderTest = new ExecutorPodsSnapshotsStoreImpl(eventBufferScheduler)
+    ExecutorPodsSnapshot.setShouldCheckAllContainers(false)
   }
 
   test("Subscribers get notified of events periodically.") {