YARN-4005. Completed container whose app is finished is possibly not removed from NMStateStore. Contributed by Jun Gong

hadoop-yarn-project/CHANGES.txt

@@ -763,6 +763,9 @@ Release 2.8.0 - UNRELEASED
     YARN-3992. TestApplicationPriority.testApplicationPriorityAllocation fails 
     intermittently. (Contributed by Sunil G)
 
+    YARN-4005. Completed container whose app is finished is possibly not
+    removed from NMStateStore. (Jun Gong via jianhe)
+
 Release 2.7.2 - UNRELEASED
 
   INCOMPATIBLE CHANGES

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/java/org/apache/hadoop/yarn/server/nodemanager/NodeStatusUpdaterImpl.java

@@ -474,12 +474,12 @@ public class NodeStatusUpdaterImpl extends AbstractService implements
         } else {
           if (!isContainerRecentlyStopped(containerId)) {
             pendingCompletedContainers.put(containerId, containerStatus);
+          }
+        }
         // Adding to finished containers cache. Cache will keep it around at
         // least for #durationToTrackStoppedContainers duration. In the
         // subsequent call to stop container it will get removed from cache.
         addCompletedContainer(containerId);
-          }
-        }
       } else {
         containerStatuses.add(containerStatus);
       }

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/test/java/org/apache/hadoop/yarn/server/nodemanager/TestNodeStatusUpdater.java

@@ -994,6 +994,40 @@ public class TestNodeStatusUpdater {
     Assert.assertTrue(containerIdSet.contains(runningContainerId));
   }
 
+  @Test(timeout = 10000)
+  public void testCompletedContainersIsRecentlyStopped() throws Exception {
+    NodeManager nm = new NodeManager();
+    nm.init(conf);
+    NodeStatusUpdaterImpl nodeStatusUpdater =
+        (NodeStatusUpdaterImpl) nm.getNodeStatusUpdater();
+    ApplicationId appId = ApplicationId.newInstance(0, 0);
+    Application completedApp = mock(Application.class);
+    when(completedApp.getApplicationState()).thenReturn(
+        ApplicationState.FINISHED);
+    ApplicationAttemptId appAttemptId =
+        ApplicationAttemptId.newInstance(appId, 0);
+    ContainerId containerId = ContainerId.newContainerId(appAttemptId, 1);
+    Token containerToken =
+        BuilderUtils.newContainerToken(containerId, "host", 1234, "user",
+            BuilderUtils.newResource(1024, 1), 0, 123,
+            "password".getBytes(), 0);
+    Container completedContainer = new ContainerImpl(conf, null,
+        null, null, null, null,
+        BuilderUtils.newContainerTokenIdentifier(containerToken)) {
+      @Override
+      public ContainerState getCurrentState() {
+        return ContainerState.COMPLETE;
+      }
+    };
+
+    nm.getNMContext().getApplications().putIfAbsent(appId, completedApp);
+    nm.getNMContext().getContainers().put(containerId, completedContainer);
+
+    Assert.assertEquals(1, nodeStatusUpdater.getContainerStatuses().size());
+    Assert.assertTrue(nodeStatusUpdater.isContainerRecentlyStopped(
+        containerId));
+  }
+
   @Test
   public void testCleanedupApplicationContainerCleanup() throws IOException {
     NodeManager nm = new NodeManager();