YARN-5190. Registering/unregistering container metrics in ContainerMonitorImpl and ContainerImpl causing uncaught exception in ContainerMonitorImpl. Contributed by Junping Du

(cherry picked from commit 99cc439e29)

hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/metrics2/impl/MetricsSystemImpl.java

@@ -255,6 +255,7 @@ void unregisterSource(String name) {
     if (namedCallbacks.containsKey(name)) {
       namedCallbacks.remove(name);
     }
+    DefaultMetricsSystem.removeSourceName(name);
   }
 
   synchronized

hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/metrics2/lib/DefaultMetricsSystem.java

@@ -107,6 +107,11 @@ public static void removeMBeanName(ObjectName name) {
     INSTANCE.removeObjectName(name.toString());
   }
 
+  @InterfaceAudience.Private
+  public static void removeSourceName(String name) {
+    INSTANCE.removeSource(name);
+  }
+
   @InterfaceAudience.Private
   public static String sourceName(String name, boolean dupOK) {
     return INSTANCE.newSourceName(name, dupOK);
@@ -127,6 +132,10 @@ synchronized void removeObjectName(String name) {
     mBeanNames.map.remove(name);
   }
 
+  synchronized void removeSource(String name) {
+    sourceNames.map.remove(name);
+  }
+
   synchronized String newSourceName(String name, boolean dupOK) {
     if (sourceNames.map.containsKey(name)) {
       if (dupOK) {

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/java/org/apache/hadoop/yarn/server/nodemanager/containermanager/monitor/ContainerMetrics.java

@@ -160,6 +160,12 @@ public static ContainerMetrics forContainer(
         DefaultMetricsSystem.instance(), containerId, flushPeriodMs, delayMs);
   }
 
+  public synchronized static ContainerMetrics getContainerMetrics(
+      ContainerId containerId) {
+    // could be null
+    return usageMetrics.get(containerId);
+  }
+
   synchronized static ContainerMetrics forContainer(
       MetricsSystem ms, ContainerId containerId, long flushPeriodMs,
       long delayMs) {
@@ -205,12 +211,14 @@ public synchronized void getMetrics(MetricsCollector collector, boolean all) {
   }
 
   public synchronized void finished() {
-    this.finished = true;
-    if (timer != null) {
-      timer.cancel();
-      timer = null;
+    if (!finished) {
+      this.finished = true;
+      if (timer != null) {
+        timer.cancel();
+        timer = null;
+      }
+      scheduleTimerTaskForUnregistration();
     }
-    scheduleTimerTaskForUnregistration();
   }
 
   public void recordMemoryUsage(int memoryMBs) {

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/java/org/apache/hadoop/yarn/server/nodemanager/containermanager/monitor/ContainersMonitorImpl.java

@@ -615,15 +615,16 @@ private void updateContainerMetrics(ContainersMonitorEvent monitoringEvent) {
     }
 
     ContainerId containerId = monitoringEvent.getContainerId();
-    ContainerMetrics usageMetrics = ContainerMetrics
-        .forContainer(containerId, containerMetricsPeriodMs,
-        containerMetricsUnregisterDelayMs);
+    ContainerMetrics usageMetrics;
 
     int vmemLimitMBs;
     int pmemLimitMBs;
     int cpuVcores;
     switch (monitoringEvent.getType()) {
     case START_MONITORING_CONTAINER:
+      usageMetrics = ContainerMetrics
+          .forContainer(containerId, containerMetricsPeriodMs,
+          containerMetricsUnregisterDelayMs);
       ContainerStartMonitoringEvent startEvent =
           (ContainerStartMonitoringEvent) monitoringEvent;
       usageMetrics.recordStateChangeDurations(
@@ -636,9 +637,16 @@ private void updateContainerMetrics(ContainersMonitorEvent monitoringEvent) {
           vmemLimitMBs, pmemLimitMBs, cpuVcores);
       break;
     case STOP_MONITORING_CONTAINER:
-      usageMetrics.finished();
+      usageMetrics = ContainerMetrics.getContainerMetrics(
+          containerId);
+      if (usageMetrics != null) {
+        usageMetrics.finished();
+      }
       break;
     case CHANGE_MONITORING_CONTAINER_RESOURCE:
+      usageMetrics = ContainerMetrics
+          .forContainer(containerId, containerMetricsPeriodMs,
+          containerMetricsUnregisterDelayMs);
       ChangeMonitoringContainerResourceEvent changeEvent =
           (ChangeMonitoringContainerResourceEvent) monitoringEvent;
       Resource resource = changeEvent.getResource();

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/test/java/org/apache/hadoop/yarn/server/nodemanager/containermanager/monitor/TestContainerMetrics.java

@@ -146,7 +146,6 @@ public void testContainerMetricsFinished() throws InterruptedException {
     system.sampleMetrics();
     system.sampleMetrics();
     Thread.sleep(100);
-    system.stop();
     // verify metrics1 is unregistered
     assertTrue(metrics1 != ContainerMetrics.forContainer(
         system, containerId1, 1, 0));
@@ -156,6 +155,9 @@ public void testContainerMetricsFinished() throws InterruptedException {
     // verify metrics3 is still registered
     assertTrue(metrics3 == ContainerMetrics.forContainer(
         system, containerId3, 1, 0));
+    // YARN-5190: move stop() to the end to verify registering containerId1 and
+    // containerId2 won't get MetricsException thrown.
+    system.stop();
     system.shutdown();
   }
 }