YARN-9640. Slow event processing could cause too many attempt unregister events. Contributed by Bibin A Chundatt.

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/ApplicationMasterService.java

@@ -94,6 +94,8 @@ public class ApplicationMasterService extends AbstractService implements
       RecordFactoryProvider.getRecordFactory(null);
   private final ConcurrentMap<ApplicationAttemptId, AllocateResponseLock> responseMap =
       new ConcurrentHashMap<ApplicationAttemptId, AllocateResponseLock>();
+  private final ConcurrentHashMap<ApplicationAttemptId, Boolean>
+      finishedAttemptCache = new ConcurrentHashMap<>();
   protected final RMContext rmContext;
   private final AMSProcessingChain amsProcessingChain;
   private boolean timelineServiceV2Enabled;
@@ -339,11 +341,14 @@ public class ApplicationMasterService extends AbstractService implements
         throw new ApplicationMasterNotRegisteredException(message);
       }
 
-      this.amLivelinessMonitor.receivedPing(applicationAttemptId);
       FinishApplicationMasterResponse response =
           FinishApplicationMasterResponse.newInstance(false);
-      this.amsProcessingChain.finishApplicationMaster(
+      if (finishedAttemptCache.putIfAbsent(applicationAttemptId, true)
-          applicationAttemptId, request, response);
+          == null) {
+        this.amsProcessingChain
+            .finishApplicationMaster(applicationAttemptId, request, response);
+      }
+      this.amLivelinessMonitor.receivedPing(applicationAttemptId);
       return response;
     }
   }
@@ -492,6 +497,7 @@ public class ApplicationMasterService extends AbstractService implements
   public void unregisterAttempt(ApplicationAttemptId attemptId) {
     LOG.info("Unregistering app attempt : " + attemptId);
     responseMap.remove(attemptId);
+    finishedAttemptCache.remove(attemptId);
     rmContext.getNMTokenSecretManager().unregisterApplicationAttempt(attemptId);
   }
 
@@ -506,6 +512,8 @@ public class ApplicationMasterService extends AbstractService implements
     if (this.server != null) {
       this.server.stop();
     }
+    responseMap.clear();
+    finishedAttemptCache.clear();
     super.serviceStop();
   }
   

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/ApplicationMasterServiceTestBase.java

@@ -17,6 +17,10 @@
 package org.apache.hadoop.yarn.server.resourcemanager;
 
 import com.google.common.collect.ImmutableMap;
+import org.apache.hadoop.yarn.event.Dispatcher;
+import org.apache.hadoop.yarn.event.DrainDispatcher;
+import org.apache.hadoop.yarn.event.Event;
+import org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttemptEventType;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 import org.apache.hadoop.yarn.api.protocolrecords.AllocateResponse;
@@ -354,6 +358,55 @@ public abstract class ApplicationMasterServiceTestBase {
     }
   }
 
+  @Test(timeout = 1200000)
+  public void testRepeatedFinishApplicationMaster() throws Exception {
+
+    CountingDispatcher dispatcher = new CountingDispatcher();
+    MockRM rm = new MockRM(conf) {
+      @Override
+      protected Dispatcher createDispatcher() {
+        return dispatcher;
+      }
+    };
+
+    try {
+      rm.start();
+      // Register node1
+      MockNM nm1 = rm.registerNode(DEFAULT_HOST + ":" + DEFAULT_PORT, 6 * GB);
+      // Submit an application
+      RMApp app1 = rm.submitApp(2048);
+      MockAM am1 = MockRM.launchAM(app1, rm, nm1);
+      am1.registerAppAttempt();
+      FinishApplicationMasterRequest req = FinishApplicationMasterRequest
+          .newInstance(FinalApplicationStatus.FAILED, "", "");
+      for (int i = 0; i < 10; i++) {
+        am1.unregisterAppAttempt(req, false);
+      }
+      Assert.assertEquals("Expecting only one event", 1,
+          dispatcher.getEventCount());
+    } finally {
+      rm.stop();
+    }
+  }
+
+  static class CountingDispatcher extends DrainDispatcher {
+    private int eventreceived = 0;
+
+    @SuppressWarnings("rawtypes")
+    @Override
+    protected void dispatch(Event event) {
+      if (event.getType() == RMAppAttemptEventType.UNREGISTERED) {
+        eventreceived++;
+      } else {
+        super.dispatch(event);
+      }
+    }
+
+    public int getEventCount() {
+      return eventreceived;
+    }
+  }
+
   @Test(timeout = 3000000)
   public void testResourceTypes() throws Exception {
     HashMap<YarnConfiguration,