Procházet zdrojové kódy

YARN-10393. MR job live lock caused by completed state container leak in heartbeat between node manager and RM. Contributed by zhenzhao wang and Jim Brennan

(cherry picked from commit a1f7e760dffeeffb9cc739f734c0a91b81a0c9d0)
Adam Antal před 4 roky
rodič
revize
3101299f02

+ 18 - 2
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/java/org/apache/hadoop/yarn/server/nodemanager/NodeStatusUpdaterImpl.java

@@ -645,7 +645,7 @@ public class NodeStatusUpdaterImpl extends AbstractService implements
   @VisibleForTesting
   @Private
   public void removeOrTrackCompletedContainersFromContext(
-      List<ContainerId> containerIds) throws IOException {
+      List<ContainerId> containerIds) {
     Set<ContainerId> removedContainers = new HashSet<ContainerId>();
 
     pendingContainersToRemove.addAll(containerIds);
@@ -662,13 +662,13 @@ public class NodeStatusUpdaterImpl extends AbstractService implements
         removedContainers.add(containerId);
         iter.remove();
       }
+      pendingCompletedContainers.remove(containerId);
     }
 
     if (!removedContainers.isEmpty()) {
       LOG.info("Removed completed containers from NM context: "
           + removedContainers);
     }
-    pendingCompletedContainers.clear();
   }
 
   private void trackAppsForKeepAlive(List<ApplicationId> appIds) {
@@ -1037,6 +1037,7 @@ public class NodeStatusUpdaterImpl extends AbstractService implements
     @SuppressWarnings("unchecked")
     public void run() {
       int lastHeartbeatID = 0;
+      boolean missedHearbeat = false;
       while (!isStopped) {
         // Send heartbeat
         try {
@@ -1083,6 +1084,20 @@ public class NodeStatusUpdaterImpl extends AbstractService implements
             removeOrTrackCompletedContainersFromContext(response
                 .getContainersToBeRemovedFromNM());
 
+            // If the last heartbeat was missed, it is possible that the
+            // RM saw this one as a duplicate and did not process it.
+            // If so, we can fail to notify the RM of these completed containers
+            // on the next heartbeat if we clear pendingCompletedContainers.
+            // If it wasn't a duplicate, the only impact is we might notify
+            // the RM twice, which it can handle.
+            if (!missedHearbeat) {
+              pendingCompletedContainers.clear();
+            } else {
+              LOG.info("skipped clearing pending completed containers due to " +
+                  "missed heartbeat");
+              missedHearbeat = false;
+            }
+
             logAggregationReportForAppsTempList.clear();
             lastHeartbeatID = response.getResponseId();
             List<ContainerId> containersToCleanup = response
@@ -1158,6 +1173,7 @@ public class NodeStatusUpdaterImpl extends AbstractService implements
           // TODO Better error handling. Thread can die with the rest of the
           // NM still running.
           LOG.error("Caught exception in status-updater", e);
+          missedHearbeat = true;
         } finally {
           synchronized (heartbeatMonitor) {
             nextHeartBeatInterval = nextHeartBeatInterval <= 0 ?

+ 7 - 17
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/test/java/org/apache/hadoop/yarn/server/nodemanager/TestNodeStatusUpdater.java

@@ -758,15 +758,11 @@ public class TestNodeStatusUpdater {
         } else if (heartBeatID == 2 || heartBeatID == 3) {
           List<ContainerStatus> statuses =
               request.getNodeStatus().getContainersStatuses();
-          if (heartBeatID == 2) {
-            // NM should send completed containers again, since the last
-            // heartbeat is lost.
-            Assert.assertEquals(4, statuses.size());
-          } else {
-            // NM should not send completed containers again, since the last
-            // heartbeat is successful.
-            Assert.assertEquals(2, statuses.size());
-          }
+          // NM should send completed containers on heartbeat 2,
+          // since heartbeat 1 was lost.  It will send them again on
+          // heartbeat 3, because it does not clear them if the previous
+          // heartbeat was lost in case the RM treated it as a duplicate.
+          Assert.assertEquals(4, statuses.size());
           Assert.assertEquals(4, context.getContainers().size());
 
           boolean container2Exist = false, container3Exist = false,
@@ -797,14 +793,8 @@ public class TestNodeStatusUpdater {
               container5Exist = true;
             }
           }
-          if (heartBeatID == 2) {
-            Assert.assertTrue(container2Exist && container3Exist
-                && container4Exist && container5Exist);
-          } else {
-            // NM do not send completed containers again
-            Assert.assertTrue(container2Exist && !container3Exist
-                && container4Exist && !container5Exist);
-          }
+          Assert.assertTrue(container2Exist && container3Exist
+              && container4Exist && container5Exist);
 
           if (heartBeatID == 3) {
             finishedContainersPulledByAM.add(containerStatus3.getContainerId());