HADOOP-11149. TestZKFailoverController times out. Contributed by Steve Loughran. closes apache/hadoop#51

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

@@ -1355,6 +1355,9 @@ Release 2.8.0 - UNRELEASED
 
     HADOOP-8419. Fixed GzipCode NPE reset for IBM JDK. (Yu Li via eyang)
 
+    HADOOP-11149. TestZKFailoverController times out. (Steve Loughran
+    via ozawa)
+
   OPTIMIZATIONS
 
     HADOOP-12051. ProtobufRpcEngine.invoke() should use Exception.toString()

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

@@ -133,11 +133,13 @@ public void start(int count) throws Exception {
    * @throws Exception if either of the services had encountered a fatal error
    */
   public void stop() throws Exception {
+    if (thrs != null) {
       for (DummyZKFCThread thr : thrs) {
         if (thr != null) {
           thr.interrupt();
         }
       }
+    }
     if (ctx != null) {
       ctx.stop();
     }

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

@@ -34,14 +34,23 @@
 import org.apache.zookeeper.ZooKeeper;
 import org.apache.zookeeper.data.Stat;
 import org.apache.zookeeper.server.auth.DigestAuthenticationProvider;
+import org.junit.After;
 import org.junit.Before;
+import org.junit.Rule;
 import org.junit.Test;
+import org.junit.rules.Timeout;
 import org.mockito.Mockito;
 
 public class TestZKFailoverController extends ClientBaseWithFixes {
   private Configuration conf;
   private MiniZKFCCluster cluster;
 
+  /**
+   * Set the timeout for every test
+   */
+  @Rule
+  public Timeout testTimeout = new Timeout(3 * 60 * 1000);
+
   // Set up ZK digest-based credentials for the purposes of the tests,
   // to make sure all of our functionality works with auth and ACLs
   // present.
@@ -74,11 +83,21 @@ public void setupConfAndServices() {
     this.cluster = new MiniZKFCCluster(conf, getServer(serverFactory));
   }
 
+  @After
+  public void teardown() {
+    if (cluster != null) {
+      try {
+        cluster.stop();
+      } catch (Exception e) {
+        LOG.warn("When stopping the cluster", e);
+      }
+    }
+  }
   /**
    * Test that the various command lines for formatting the ZK directory
    * function correctly.
    */
-  @Test(timeout=15000)
+  @Test
   public void testFormatZK() throws Exception {
     DummyHAService svc = cluster.getService(1);
     // Run without formatting the base dir,
@@ -101,7 +120,7 @@ public void testFormatZK() throws Exception {
    * Test that if ZooKeeper is not running, the correct error
    * code is returned.
    */
-  @Test(timeout=15000)
+  @Test
   public void testNoZK() throws Exception {
     stopServer();
     DummyHAService svc = cluster.getService(1);
@@ -146,7 +165,7 @@ protected String getScopeInsideParentNode() {
    * Test that automatic failover won't run against a target that hasn't
    * explicitly enabled the feature.
    */
-  @Test(timeout=10000)
+  @Test
   public void testWontRunWhenAutoFailoverDisabled() throws Exception {
     DummyHAService svc = cluster.getService(1);
     svc = Mockito.spy(svc);
@@ -162,7 +181,7 @@ public void testWontRunWhenAutoFailoverDisabled() throws Exception {
    * Test that, if ACLs are specified in the configuration, that
    * it sets the ACLs when formatting the parent node.
    */
-  @Test(timeout=15000)
+  @Test
   public void testFormatSetsAcls() throws Exception {
     // Format the base dir, should succeed
     DummyHAService svc = cluster.getService(1);
@@ -184,7 +203,7 @@ public void testFormatSetsAcls() throws Exception {
    * Test that the ZKFC won't run if fencing is not configured for the
    * local service.
    */
-  @Test(timeout=15000)
+  @Test
   public void testFencingMustBeConfigured() throws Exception {
     DummyHAService svc = Mockito.spy(cluster.getService(0));
     Mockito.doThrow(new BadFencingConfigurationException("no fencing"))
@@ -202,9 +221,8 @@ public void testFencingMustBeConfigured() throws Exception {
    * transition is used when possible, falling back to fencing when
    * the graceful approach fails.
    */
-  @Test(timeout=15000)
+  @Test
   public void testAutoFailoverOnBadHealth() throws Exception {
-    try {
     cluster.start();
     DummyHAService svc1 = cluster.getService(1);
 
@@ -224,9 +242,6 @@ public void testAutoFailoverOnBadHealth() throws Exception {
     cluster.waitForHAState(0, HAServiceState.ACTIVE);
     // and fence svc1
     Mockito.verify(svc1.fencer).fence(Mockito.same(svc1));
-    } finally {
-      cluster.stop();
-    }
   }
 
   /**
@@ -235,9 +250,8 @@ public void testAutoFailoverOnBadHealth() throws Exception {
    * transition is used when possible, falling back to fencing when
    * the graceful approach fails.
    */
-  @Test(timeout=15000)
+  @Test
   public void testAutoFailoverOnBadState() throws Exception {
-    try {
     cluster.start();
     DummyHAService svc0 = cluster.getService(0);
     LOG.info("Faking svc0 to change the state, should failover to svc1");
@@ -245,14 +259,10 @@ public void testAutoFailoverOnBadState() throws Exception {
 
     // Should fail back to svc0 at this point
     cluster.waitForHAState(1, HAServiceState.ACTIVE);
-    } finally {
-      cluster.stop();
-    }
   }
 
-  @Test(timeout=15000)
+  @Test
   public void testAutoFailoverOnLostZKSession() throws Exception {
-    try {
     cluster.start();
 
     // Expire svc0, it should fail over to svc1
@@ -268,18 +278,14 @@ public void testAutoFailoverOnLostZKSession() throws Exception {
 
     // Expire svc1, it should fail back to svc0
     cluster.expireAndVerifyFailover(1, 0);
-    } finally {
-      cluster.stop();
-    }
   }
 
   /**
    * Test that, if the standby node is unhealthy, it doesn't try to become
    * active
    */
-  @Test(timeout=15000)
+  @Test
   public void testDontFailoverToUnhealthyNode() throws Exception {
-    try {
     cluster.start();
 
     // Make svc1 unhealthy, and wait for its FC to notice the bad health.
@@ -304,18 +310,14 @@ public void testDontFailoverToUnhealthyNode() throws Exception {
     }
     // svc0 should get the lock again
     cluster.waitForActiveLockHolder(0);
-    } finally {
-      cluster.stop();
-    }
   }
 
   /**
    * Test that the ZKFC successfully quits the election when it fails to
    * become active. This allows the old node to successfully fail back.
    */
-  @Test(timeout=15000)
+  @Test
   public void testBecomingActiveFails() throws Exception {
-    try {
     cluster.start();
     DummyHAService svc1 = cluster.getService(1);
 
@@ -346,9 +348,6 @@ public void testBecomingActiveFails() throws Exception {
     LOG.info("Allowing svc1 to become active, expiring svc0");
     svc1.failToBecomeActive = false;
     cluster.expireAndVerifyFailover(0, 1);
-    } finally {
-      cluster.stop();
-    }
   }
   
   /**
@@ -356,9 +355,8 @@ public void testBecomingActiveFails() throws Exception {
    * current state, without triggering any failovers, and without
    * causing the active node to enter standby state.
    */
-  @Test(timeout=15000)
+  @Test
   public void testZooKeeperFailure() throws Exception {
-    try {
     cluster.start();
 
     // Record initial ZK sessions
@@ -396,17 +394,13 @@ public void testZooKeeperFailure() throws Exception {
         cluster.getElector(0).getZKSessionIdForTests());
     assertEquals(session1,
         cluster.getElector(1).getZKSessionIdForTests());
-    } finally {
-      cluster.stop();
-    }
   }
   
   /**
    * Test that the ZKFC can gracefully cede its active status.
    */
-  @Test(timeout=15000)
+  @Test
   public void testCedeActive() throws Exception {
-    try {
     cluster.start();
     DummyZKFC zkfc = cluster.getZkfc(0);
     // It should be in active to start.
@@ -434,14 +428,10 @@ public void testCedeActive() throws Exception {
     assertTrue("Should take ~3 seconds to rejoin. Only took " + (et2 - et) +
         "ms before rejoining.",
         et2 - et > 2800);
-    } finally {
-      cluster.stop();
-    }
   }
 
-  @Test(timeout=25000)
+  @Test
   public void testGracefulFailover() throws Exception {
-    try {
     cluster.start();
 
     cluster.waitForActiveLockHolder(0);
@@ -457,14 +447,10 @@ public void testGracefulFailover() throws Exception {
     assertEquals(0, cluster.getService(1).fenceCount);
     assertEquals(2, cluster.getService(0).activeTransitionCount);
     assertEquals(1, cluster.getService(1).activeTransitionCount);
-    } finally {
-      cluster.stop();
-    }
   }
 
-  @Test(timeout=15000)
+  @Test
   public void testGracefulFailoverToUnhealthy() throws Exception {
-    try {
     cluster.start();
 
     cluster.waitForActiveLockHolder(0);
@@ -482,14 +468,10 @@ public void testGracefulFailoverToUnhealthy() throws Exception {
           cluster.getService(1).toString() +
           " is not currently healthy.", sfe);
     }
-    } finally {
-      cluster.stop();
-    }
   }
 
-  @Test(timeout=15000)
+  @Test
   public void testGracefulFailoverFailBecomingActive() throws Exception {
-    try {
     cluster.start();
 
     cluster.waitForActiveLockHolder(0);
@@ -513,14 +495,10 @@ public void testGracefulFailoverFailBecomingActive() throws Exception {
 
     // Service 0 should go back to being active after the failed failover
     cluster.waitForActiveLockHolder(0);
-    } finally {
-      cluster.stop();
-    }
   }
 
-  @Test(timeout=15000)
+  @Test
   public void testGracefulFailoverFailBecomingStandby() throws Exception {
-    try {
     cluster.start();
 
     cluster.waitForActiveLockHolder(0);
@@ -533,15 +511,11 @@ public void testGracefulFailoverFailBecomingStandby() throws Exception {
 
     // Check that the old node was fenced
     assertEquals(1, cluster.getService(0).fenceCount);
-    } finally {
-      cluster.stop();
-    }
   }
 
-  @Test(timeout=15000)
+  @Test
   public void testGracefulFailoverFailBecomingStandbyAndFailFence()
     throws Exception {
-    try {
     cluster.start();
 
     cluster.waitForActiveLockHolder(0);
@@ -559,18 +533,14 @@ public void testGracefulFailoverFailBecomingStandbyAndFailFence()
       GenericTestUtils.assertExceptionContains(
           "Unable to fence " + cluster.getService(0), sfe);
     }
-    } finally {
-      cluster.stop();
-    }
   }
 
   /**
    * Test which exercises all of the inputs into ZKFC. This is particularly
    * useful for running under jcarder to check for lock order violations.
    */
-  @Test(timeout=30000)
+  @Test
   public void testOneOfEverything() throws Exception {
-    try {
     cluster.start();
 
     // Failover by session expiration
@@ -600,14 +570,10 @@ public void testOneOfEverything() throws Exception {
     // Graceful failovers
     cluster.getZkfc(1).gracefulFailoverToYou();
     cluster.getZkfc(0).gracefulFailoverToYou();
-    } finally {
-      cluster.stop();
-    }
   }
 
-  @Test(timeout = 25000)
+  @Test
   public void testGracefulFailoverMultipleZKfcs() throws Exception {
-    try {
     cluster.start(3);
 
     cluster.waitForActiveLockHolder(0);
@@ -632,9 +598,6 @@ public void testGracefulFailoverMultipleZKfcs() throws Exception {
     assertEquals(2, cluster.getService(0).activeTransitionCount);
     assertEquals(1, cluster.getService(1).activeTransitionCount);
     assertEquals(1, cluster.getService(2).activeTransitionCount);
-    } finally {
-      cluster.stop();
-    }
   }
 
   private int runFC(DummyHAService target, String ... args) throws Exception {