YARN-4408. Fix issue that NodeManager still reports negative running containers. Contributed by Robert Kanter.

(cherry picked from commit 62e9348bc1)

hadoop-yarn-project/CHANGES.txt

@@ -1032,6 +1032,9 @@ Release 2.8.0 - UNRELEASED
     YARN-4384. updateNodeResource CLI should not accept negative values for resource.
     (Junping Du via wangda)
 
+    YARN-4408. Fix issue that NodeManager reports negative running containers. 
+    (Robert Kanter via junping_du)
+
 Release 2.7.3 - UNRELEASED
 
   INCOMPATIBLE CHANGES

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/java/org/apache/hadoop/yarn/server/nodemanager/containermanager/container/ContainerImpl.java

@@ -1041,7 +1041,12 @@ public class ContainerImpl implements Container {
       ContainerDoneTransition {
     @Override
     public void transition(ContainerImpl container, ContainerEvent event) {
+      if (container.wasLaunched) {
         container.metrics.endRunningContainer();
+      } else {
+        LOG.warn("Container exited with success despite being killed and not" +
+            "actually running");
+      }
       container.metrics.completedContainer();
       NMAuditLogger.logSuccess(container.user,
           AuditConstants.FINISH_SUCCESS_CONTAINER, "ContainerImpl",

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/test/java/org/apache/hadoop/yarn/server/nodemanager/containermanager/container/TestContainer.java

@@ -397,7 +397,8 @@ public class TestContainer {
   }
 
   @Test
-  public void testKillOnLocalizedWhenContainerNotLaunched() throws Exception {
+  public void testKillOnLocalizedWhenContainerNotLaunchedContainerKilled()
+      throws Exception {
     WrappedContainer wc = null;
     try {
       wc = new WrappedContainer(17, 314159265358979L, 4344, "yak");
@@ -426,6 +427,62 @@ public class TestContainer {
     }
   }
 
+  @Test
+  public void testKillOnLocalizedWhenContainerNotLaunchedContainerSuccess()
+      throws Exception {
+    WrappedContainer wc = null;
+    try {
+      wc = new WrappedContainer(17, 314159265358979L, 4344, "yak");
+      wc.initContainer();
+      wc.localizeResources();
+      assertEquals(ContainerState.LOCALIZED, wc.c.getContainerState());
+      wc.killContainer();
+      assertEquals(ContainerState.KILLING, wc.c.getContainerState());
+      wc.containerSuccessful();
+      wc.drainDispatcherEvents();
+      assertEquals(ContainerState.EXITED_WITH_SUCCESS,
+          wc.c.getContainerState());
+      assertNull(wc.c.getLocalizedResources());
+      verifyCleanupCall(wc);
+      wc.c.handle(new ContainerEvent(wc.c.getContainerId(),
+          ContainerEventType.CONTAINER_RESOURCES_CLEANEDUP));
+      assertEquals(ContainerState.DONE, wc.c.getContainerState());
+      assertEquals(0, metrics.getRunningContainers());
+    } finally {
+      if (wc != null) {
+        wc.finished();
+      }
+    }
+  }
+
+  @Test
+  public void testKillOnLocalizedWhenContainerNotLaunchedContainerFailure()
+      throws Exception {
+    WrappedContainer wc = null;
+    try {
+      wc = new WrappedContainer(17, 314159265358979L, 4344, "yak");
+      wc.initContainer();
+      wc.localizeResources();
+      assertEquals(ContainerState.LOCALIZED, wc.c.getContainerState());
+      wc.killContainer();
+      assertEquals(ContainerState.KILLING, wc.c.getContainerState());
+      wc.containerFailed(ExitCode.FORCE_KILLED.getExitCode());
+      wc.drainDispatcherEvents();
+      assertEquals(ContainerState.EXITED_WITH_FAILURE,
+          wc.c.getContainerState());
+      assertNull(wc.c.getLocalizedResources());
+      verifyCleanupCall(wc);
+      wc.c.handle(new ContainerEvent(wc.c.getContainerId(),
+          ContainerEventType.CONTAINER_RESOURCES_CLEANEDUP));
+      assertEquals(ContainerState.DONE, wc.c.getContainerState());
+      assertEquals(0, metrics.getRunningContainers());
+    } finally {
+      if (wc != null) {
+        wc.finished();
+      }
+    }
+  }
+
   @Test
   public void testKillOnLocalizedWhenContainerLaunched() throws Exception {
     WrappedContainer wc = null;