HADOOP-12186. ActiveStandbyElector shouldn't call monitorLockNodeAsync multiple times (Contributed by zhihai xu)

hadoop-common-project/hadoop-common/CHANGES.txt

@@ -924,6 +924,9 @@ Release 2.8.0 - UNRELEASED
     HADOOP-12164. Fix TestMove and TestFsShellReturnCode failed to get command
     name using reflection. (Lei (Eddy) Xu)
 
+    HADOOP-12186. ActiveStandbyElector shouldn't call monitorLockNodeAsync
+    multiple times (zhihai xu via vinayakumarb)
+
 Release 2.7.2 - UNRELEASED
 
   INCOMPATIBLE CHANGES

hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/ha/ActiveStandbyElector.java

@@ -173,6 +173,8 @@ public class ActiveStandbyElector implements StatCallback, StringCallback {
 
   private Lock sessionReestablishLockForTests = new ReentrantLock();
   private boolean wantToBeInElection;
+  private boolean monitorLockNodePending = false;
+  private ZooKeeper monitorLockNodeClient;
 
   /**
    * Create a new ActiveStandbyElector object <br/>
@@ -468,6 +470,7 @@ public class ActiveStandbyElector implements StatCallback, StringCallback {
   public synchronized void processResult(int rc, String path, Object ctx,
       Stat stat) {
     if (isStaleClient(ctx)) return;
+    monitorLockNodePending = false;
 
     assert wantToBeInElection :
         "Got a StatNode result after quitting election";
@@ -744,6 +747,11 @@ public class ActiveStandbyElector implements StatCallback, StringCallback {
     return state;
   }
 
+  @VisibleForTesting
+  synchronized boolean isMonitorLockNodePending() {
+    return monitorLockNodePending;
+  }
+
   private boolean reEstablishSession() {
     int connectionRetryCount = 0;
     boolean success = false;
@@ -949,6 +957,12 @@ public class ActiveStandbyElector implements StatCallback, StringCallback {
   }
 
   private void monitorLockNodeAsync() {
+    if (monitorLockNodePending && monitorLockNodeClient == zkClient) {
+      LOG.info("Ignore duplicate monitor lock-node request.");
+      return;
+    }
+    monitorLockNodePending = true;
+    monitorLockNodeClient = zkClient;
     zkClient.exists(zkLockFilePath,
         watcher, this,
         zkClient);

hadoop-common-project/hadoop-common/src/test/java/org/apache/hadoop/ha/TestActiveStandbyElector.java

@@ -452,6 +452,10 @@ public class TestActiveStandbyElector {
         Event.KeeperState.SyncConnected);
     elector.processWatchEvent(mockZK, mockEvent);
     verifyExistCall(1);
+    Assert.assertTrue(elector.isMonitorLockNodePending());
+    elector.processResult(Code.SESSIONEXPIRED.intValue(), ZK_LOCK_NAME,
+        mockZK, new Stat());
+    Assert.assertFalse(elector.isMonitorLockNodePending());
 
     // session expired should enter safe mode and initiate re-election
     // re-election checked via checking re-creation of new zookeeper and
@@ -495,6 +499,13 @@ public class TestActiveStandbyElector {
         ZK_LOCK_NAME);
     Mockito.verify(mockApp, Mockito.times(1)).becomeStandby();
     verifyExistCall(1);
+    Assert.assertTrue(elector.isMonitorLockNodePending());
+
+    Stat stat = new Stat();
+    stat.setEphemeralOwner(0L);
+    Mockito.when(mockZK.getSessionId()).thenReturn(1L);
+    elector.processResult(Code.OK.intValue(), ZK_LOCK_NAME, mockZK, stat);
+    Assert.assertFalse(elector.isMonitorLockNodePending());
 
     WatchedEvent mockEvent = Mockito.mock(WatchedEvent.class);
     Mockito.when(mockEvent.getPath()).thenReturn(ZK_LOCK_NAME);
@@ -504,12 +515,18 @@ public class TestActiveStandbyElector {
         Event.EventType.NodeDataChanged);
     elector.processWatchEvent(mockZK, mockEvent);
     verifyExistCall(2);
+    Assert.assertTrue(elector.isMonitorLockNodePending());
+    elector.processResult(Code.OK.intValue(), ZK_LOCK_NAME, mockZK, stat);
+    Assert.assertFalse(elector.isMonitorLockNodePending());
 
     // monitoring should be setup again after event is received
     Mockito.when(mockEvent.getType()).thenReturn(
         Event.EventType.NodeChildrenChanged);
     elector.processWatchEvent(mockZK, mockEvent);
     verifyExistCall(3);
+    Assert.assertTrue(elector.isMonitorLockNodePending());
+    elector.processResult(Code.OK.intValue(), ZK_LOCK_NAME, mockZK, stat);
+    Assert.assertFalse(elector.isMonitorLockNodePending());
 
     // lock node deletion when in standby mode should create znode again
     // successful znode creation enters active state and sets monitor
@@ -524,6 +541,10 @@ public class TestActiveStandbyElector {
         ZK_LOCK_NAME);
     Mockito.verify(mockApp, Mockito.times(1)).becomeActive();
     verifyExistCall(4);
+    Assert.assertTrue(elector.isMonitorLockNodePending());
+    stat.setEphemeralOwner(1L);
+    elector.processResult(Code.OK.intValue(), ZK_LOCK_NAME, mockZK, stat);
+    Assert.assertFalse(elector.isMonitorLockNodePending());
 
     // lock node deletion in active mode should enter neutral mode and create
     // znode again successful znode creation enters active state and sets
@@ -538,6 +559,9 @@ public class TestActiveStandbyElector {
         ZK_LOCK_NAME);
     Mockito.verify(mockApp, Mockito.times(2)).becomeActive();
     verifyExistCall(5);
+    Assert.assertTrue(elector.isMonitorLockNodePending());
+    elector.processResult(Code.OK.intValue(), ZK_LOCK_NAME, mockZK, stat);
+    Assert.assertFalse(elector.isMonitorLockNodePending());
 
     // bad path name results in fatal error
     Mockito.when(mockEvent.getPath()).thenReturn(null);
@@ -570,6 +594,13 @@ public class TestActiveStandbyElector {
         ZK_LOCK_NAME);
     Mockito.verify(mockApp, Mockito.times(1)).becomeStandby();
     verifyExistCall(1);
+    Assert.assertTrue(elector.isMonitorLockNodePending());
+
+    Stat stat = new Stat();
+    stat.setEphemeralOwner(0L);
+    Mockito.when(mockZK.getSessionId()).thenReturn(1L);
+    elector.processResult(Code.OK.intValue(), ZK_LOCK_NAME, mockZK, stat);
+    Assert.assertFalse(elector.isMonitorLockNodePending());
 
     WatchedEvent mockEvent = Mockito.mock(WatchedEvent.class);
     Mockito.when(mockEvent.getPath()).thenReturn(ZK_LOCK_NAME);