Examples with ClusterControllerManager org.apache.helix.integration.manager.ClusterControllerManager used on opensource projects

Example 31 with ClusterControllerManager

use of org.apache.helix.integration.manager.ClusterControllerManager in project helix by apache.

the class TestZkCallbackHandlerLeak method testRemoveUserCbHandlerOnPathRemoval.

@Test
public void testRemoveUserCbHandlerOnPathRemoval() throws Exception {
    String className = TestHelper.getTestClassName();
    String methodName = TestHelper.getTestMethodName();
    String clusterName = className + "_" + methodName;
    final int n = 3;
    final String zkAddr = ZK_ADDR;
    System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
    // resource
    TestHelper.setupCluster(// resource
    clusterName, // resource
    zkAddr, // resource
    12918, // resource
    "localhost", // resource
    "TestDB", // resource
    1, // partitions
    32, // nodes
    n, // replicas
    2, "MasterSlave", true);
    final ClusterControllerManager controller = new ClusterControllerManager(zkAddr, clusterName, "controller_0");
    controller.syncStart();
    MockParticipantManager[] participants = new MockParticipantManager[n];
    for (int i = 0; i < n; i++) {
        String instanceName = "localhost_" + (12918 + i);
        participants[i] = new MockParticipantManager(zkAddr, clusterName, instanceName);
        participants[i].syncStart();
        // register a controller listener on participant_0
        if (i == 0) {
            // ZkHelixTestManager manager = participants[0].getManager();
            MockParticipantManager manager = participants[0];
            manager.addCurrentStateChangeListener(new CurrentStateChangeListener() {

                @Override
                public void onStateChange(String instanceName, List<CurrentState> statesInfo, NotificationContext changeContext) {
                // To change body of implemented methods use File | Settings | File Templates.
                // System.out.println(instanceName + " on current-state change, type: " +
                // changeContext.getType());
                }
            }, manager.getInstanceName(), manager.getSessionId());
        }
    }
    Boolean result = ClusterStateVerifier.verifyByZkCallback(new ClusterStateVerifier.BestPossAndExtViewZkVerifier(zkAddr, clusterName));
    Assert.assertTrue(result);
    MockParticipantManager participantToExpire = participants[0];
    String oldSessionId = participantToExpire.getSessionId();
    PropertyKey.Builder keyBuilder = new PropertyKey.Builder(clusterName);
    // check manager#hanlders
    Assert.assertEquals(participantToExpire.getHandlers().size(), 2, "Should have 2 handlers: CURRENTSTATE/{sessionId}, and MESSAGES");
    // check zkclient#listeners
    Map<String, Set<IZkDataListener>> dataListeners = ZkTestHelper.getZkDataListener(participantToExpire.getZkClient());
    Map<String, Set<IZkChildListener>> childListeners = ZkTestHelper.getZkChildListener(participantToExpire.getZkClient());
    // printZkListeners(participantToExpire.getZkClient());
    Assert.assertEquals(dataListeners.size(), 1, "Should have 1 path (CURRENTSTATE/{sessionId}/TestDB0) which has 1 data-listeners");
    String path = keyBuilder.currentState(participantToExpire.getInstanceName(), oldSessionId, "TestDB0").getPath();
    Assert.assertEquals(dataListeners.get(path).size(), 1, "Should have 1 data-listeners on path: " + path);
    Assert.assertEquals(childListeners.size(), 2, "Should have 2 paths (CURRENTSTATE/{sessionId}, and MESSAGES) each of which has 1 child-listener");
    path = keyBuilder.currentStates(participantToExpire.getInstanceName(), oldSessionId).getPath();
    Assert.assertEquals(childListeners.get(path).size(), 1, "Should have 1 child-listener on path: " + path);
    path = keyBuilder.messages(participantToExpire.getInstanceName()).getPath();
    Assert.assertEquals(childListeners.get(path).size(), 1, "Should have 1 child-listener on path: " + path);
    path = keyBuilder.controller().getPath();
    Assert.assertNull(childListeners.get(path), "Should have no child-listener on path: " + path);
    // check zookeeper#watches on client side
    Map<String, List<String>> watchPaths = ZkTestHelper.getZkWatch(participantToExpire.getZkClient());
    // System.out.println("localhost_12918 zk-client side watchPaths: " + watchPaths + "\n");
    Assert.assertEquals(watchPaths.get("dataWatches").size(), 3, "Should have 3 data-watches: CURRENTSTATE/{sessionId}, CURRENTSTATE/{sessionId}/TestDB, MESSAGES");
    Assert.assertEquals(watchPaths.get("childWatches").size(), 2, "Should have 2 child-watches: MESSAGES, and CURRENTSTATE/{sessionId}");
    // expire localhost_12918
    System.out.println("Expire participant: " + participantToExpire.getInstanceName() + ", session: " + participantToExpire.getSessionId());
    ZkTestHelper.expireSession(participantToExpire.getZkClient());
    String newSessionId = participantToExpire.getSessionId();
    System.out.println(participantToExpire.getInstanceName() + " oldSessionId: " + oldSessionId + ", newSessionId: " + newSessionId);
    HelixClusterVerifier verifier = new BestPossibleExternalViewVerifier.Builder(clusterName).setZkAddr(ZK_ADDR).build();
    Assert.assertTrue(verifier.verify());
    // check manager#hanlders
    Assert.assertEquals(participantToExpire.getHandlers().size(), 1, "Should have 1 handlers: MESSAGES. CURRENTSTATE/{sessionId} handler should be removed by CallbackHandler#handleChildChange()");
    // check zkclient#listeners
    dataListeners = ZkTestHelper.getZkDataListener(participantToExpire.getZkClient());
    childListeners = ZkTestHelper.getZkChildListener(participantToExpire.getZkClient());
    // printZkListeners(participantToExpire.getZkClient());
    Assert.assertTrue(dataListeners.isEmpty(), "Should have no data-listeners");
    Assert.assertEquals(childListeners.size(), 2, "Should have 2 paths (CURRENTSTATE/{oldSessionId}, and MESSAGES). " + "CONTROLLER and MESSAGE has 1 child-listener each. CURRENTSTATE/{oldSessionId} doesn't have listener (ZkClient doesn't remove empty childListener set. probably a ZkClient bug. see ZkClient#unsubscribeChildChange())");
    path = keyBuilder.currentStates(participantToExpire.getInstanceName(), oldSessionId).getPath();
    Assert.assertEquals(childListeners.get(path).size(), 0, "Should have no child-listener on path: " + path);
    path = keyBuilder.messages(participantToExpire.getInstanceName()).getPath();
    Assert.assertEquals(childListeners.get(path).size(), 1, "Should have 1 child-listener on path: " + path);
    path = keyBuilder.controller().getPath();
    Assert.assertNull(childListeners.get(path), "Should have no child-listener on path: " + path);
    // check zookeeper#watches on client side
    watchPaths = ZkTestHelper.getZkWatch(participantToExpire.getZkClient());
    // System.out.println("localhost_12918 zk-client side watchPaths: " + watchPaths + "\n");
    Assert.assertEquals(watchPaths.get("dataWatches").size(), 1, "Should have 1 data-watches: MESSAGES");
    Assert.assertEquals(watchPaths.get("childWatches").size(), 1, "Should have 1 child-watches: MESSAGES");
    Assert.assertEquals(watchPaths.get("existWatches").size(), 2, "Should have 2 exist-watches: CURRENTSTATE/{oldSessionId} and CURRENTSTATE/{oldSessionId}/TestDB0");
    // another session expiry on localhost_12918 should clear the two exist-watches on
    // CURRENTSTATE/{oldSessionId}
    System.out.println("Expire participant: " + participantToExpire.getInstanceName() + ", session: " + participantToExpire.getSessionId());
    ZkTestHelper.expireSession(participantToExpire.getZkClient());
    Assert.assertTrue(verifier.verify());
    // check zookeeper#watches on client side
    watchPaths = ZkTestHelper.getZkWatch(participantToExpire.getZkClient());
    // System.out.println("localhost_12918 zk-client side watchPaths: " + watchPaths + "\n");
    Assert.assertEquals(watchPaths.get("dataWatches").size(), 1, "Should have 1 data-watches: MESSAGES");
    Assert.assertEquals(watchPaths.get("childWatches").size(), 1, "Should have 1 child-watches: MESSAGES");
    Assert.assertEquals(watchPaths.get("existWatches").size(), 0, "Should have no exist-watches. exist-watches on CURRENTSTATE/{oldSessionId} and CURRENTSTATE/{oldSessionId}/TestDB0 should be cleared during handleNewSession");
    // Thread.sleep(1000);
    // clean up
    controller.syncStop();
    for (int i = 0; i < n; i++) {
        participants[i].syncStop();
    }
    System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}

Example 32 with ClusterControllerManager

use of org.apache.helix.integration.manager.ClusterControllerManager in project helix by apache.

the class TestAddStateModelFactoryAfterConnect method testBasic.

@Test
public void testBasic() throws Exception {
    // Logger.getRootLogger().setLevel(Level.INFO);
    String className = TestHelper.getTestClassName();
    String methodName = TestHelper.getTestMethodName();
    String clusterName = className + "_" + methodName;
    final int n = 5;
    System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
    MockParticipantManager[] participants = new MockParticipantManager[n];
    // participant port
    TestHelper.setupCluster(// participant port
    clusterName, // participant port
    ZK_ADDR, // participant port
    12918, // participant name prefix
    "localhost", // resource name prefix
    "TestDB", // resources
    1, // partitions per resource
    10, // number of nodes
    n, // replicas
    3, "MasterSlave", // do rebalance
    true);
    ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
    controller.syncStart();
    // start participants
    for (int i = 0; i < n; i++) {
        String instanceName = "localhost_" + (12918 + i);
        participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
        participants[i].syncStart();
    }
    boolean result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
    Assert.assertTrue(result);
    // add a new idealState without registering message handling factory
    ClusterSetup setupTool = new ClusterSetup(ZK_ADDR);
    setupTool.addResourceToCluster(clusterName, "TestDB1", 16, "MasterSlave");
    ZkBaseDataAccessor<ZNRecord> baseAccessor = new ZkBaseDataAccessor<ZNRecord>(_gZkClient);
    ZKHelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, baseAccessor);
    Builder keyBuilder = accessor.keyBuilder();
    IdealState idealState = accessor.getProperty(keyBuilder.idealStates("TestDB1"));
    idealState.setStateModelFactoryName("TestDB1_Factory");
    accessor.setProperty(keyBuilder.idealStates("TestDB1"), idealState);
    setupTool.rebalanceStorageCluster(clusterName, "TestDB1", 3);
    // assert that we have received OFFLINE->SLAVE messages for all partitions
    int totalMsgs = 0;
    for (int retry = 0; retry < 5; retry++) {
        Thread.sleep(100);
        totalMsgs = 0;
        for (int i = 0; i < n; i++) {
            List<Message> msgs = accessor.getChildValues(keyBuilder.messages(participants[i].getInstanceName()));
            totalMsgs += msgs.size();
        }
        if (// partition# x replicas
        totalMsgs == 48)
            break;
    }
    Assert.assertEquals(totalMsgs, 48, "Should accumulated 48 unprocessed messages (1 O->S per partition per replica) because TestDB1 is added without state-model-factory but was " + totalMsgs);
    // Logger.getRootLogger().setLevel(Level.INFO);
    for (int i = 0; i < n; i++) {
        participants[i].getStateMachineEngine().registerStateModelFactory("MasterSlave", new MockMSModelFactory(), "TestDB1_Factory");
    }
    result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
    Assert.assertTrue(result);
    // clean up
    // wait for all zk callbacks done
    controller.syncStop();
    for (int i = 0; i < 5; i++) {
        participants[i].syncStop();
    }
    System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}

Example 33 with ClusterControllerManager

use of org.apache.helix.integration.manager.ClusterControllerManager in project helix by apache.

the class TestAlertingRebalancerFailure method beforeClass.

@Override
@BeforeClass
public void beforeClass() throws Exception {
    System.out.println("START " + CLASS_NAME + " at " + new Date(System.currentTimeMillis()));
    String namespace = "/" + CLUSTER_NAME;
    if (_gZkClient.exists(namespace)) {
        _gZkClient.deleteRecursively(namespace);
    }
    _setupTool = new ClusterSetup(ZK_ADDR);
    // setup storage cluster
    _setupTool.addCluster(CLUSTER_NAME, true);
    for (int i = 0; i < NODE_NR; i++) {
        String storageNodeName = PARTICIPANT_PREFIX + "_" + (START_PORT + i);
        _setupTool.addInstanceToCluster(CLUSTER_NAME, storageNodeName);
    }
    // start controller
    String controllerName = CONTROLLER_PREFIX + "_0";
    _controller = new ClusterControllerManager(ZK_ADDR, CLUSTER_NAME, controllerName);
    _controller.syncStart();
    // start dummy participants
    for (int i = 0; i < NODE_NR; i++) {
        String instanceName = PARTICIPANT_PREFIX + "_" + (START_PORT + i);
        _instanceNames.add(instanceName);
        _participants[i] = new MockParticipantManager(ZK_ADDR, CLUSTER_NAME, instanceName);
        _participants[i].syncStart();
    }
    enablePersistBestPossibleAssignment(_gZkClient, CLUSTER_NAME, true);
    accessor = new ZKHelixDataAccessor(CLUSTER_NAME, _baseAccessor);
    errorNodeKey = accessor.keyBuilder().controllerTaskError(RebalanceResourceFailure.name());
}

Example 34 with ClusterControllerManager

use of org.apache.helix.integration.manager.ClusterControllerManager in project helix by apache.

the class TestCarryOverBadCurState method testCarryOverBadCurState.

@Test
public void testCarryOverBadCurState() throws Exception {
    System.out.println("START testCarryOverBadCurState at " + new Date(System.currentTimeMillis()));
    String clusterName = getShortClassName();
    MockParticipantManager[] participants = new MockParticipantManager[5];
    // participant port
    TestHelper.setupCluster(// participant port
    clusterName, // participant port
    ZK_ADDR, // participant port
    12918, // participant name prefix
    "localhost", // resource name prefix
    "TestDB", // resources
    1, // partitions per resource
    10, // number of nodes
    5, // replicas
    3, "MasterSlave", // do rebalance
    true);
    // add a bad current state
    ZNRecord badCurState = new ZNRecord("TestDB0");
    String path = PropertyPathBuilder.instanceCurrentState(clusterName, "localhost_12918", "session_0", "TestDB0");
    _gZkClient.createPersistent(path, true);
    _gZkClient.writeData(path, badCurState);
    ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
    controller.syncStart();
    // start participants
    for (int i = 0; i < 5; i++) {
        String instanceName = "localhost_" + (12918 + i);
        participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
        participants[i].syncStart();
    }
    boolean result = ClusterStateVerifier.verifyByZkCallback(new MasterNbInExtViewVerifier(ZK_ADDR, clusterName));
    Assert.assertTrue(result);
    result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
    Assert.assertTrue(result);
    // clean up
    controller.syncStop();
    for (int i = 0; i < 5; i++) {
        participants[i].syncStop();
    }
    System.out.println("END testCarryOverBadCurState at " + new Date(System.currentTimeMillis()));
}

Example 35 with ClusterControllerManager

use of org.apache.helix.integration.manager.ClusterControllerManager in project helix by apache.

the class TestDrop method testDropErrorPartitionFailedAutoIS.

@Test
public void testDropErrorPartitionFailedAutoIS() throws Exception {
    // Logger.getRootLogger().setLevel(Level.INFO);
    String className = TestHelper.getTestClassName();
    String methodName = TestHelper.getTestMethodName();
    String clusterName = className + "_" + methodName;
    final int n = 5;
    System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
    MockParticipantManager[] participants = new MockParticipantManager[n];
    // participant port
    TestHelper.setupCluster(// participant port
    clusterName, // participant port
    ZK_ADDR, // participant port
    12918, // participant name prefix
    "localhost", // resource name prefix
    "TestDB", // resources
    1, // partitions per resource
    8, // number of nodes
    n, // replicas
    3, "MasterSlave", // do rebalance
    true);
    // start controller
    ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
    controller.syncStart();
    // start participants
    Map<String, Set<String>> errTransitions = new HashMap<String, Set<String>>();
    errTransitions.put("SLAVE-MASTER", TestHelper.setOf("TestDB0_4"));
    errTransitions.put("ERROR-DROPPED", TestHelper.setOf("TestDB0_4"));
    for (int i = 0; i < n; i++) {
        String instanceName = "localhost_" + (12918 + i);
        if (i == 0) {
            participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
            participants[i].setTransition(new ErrTransition(errTransitions));
        } else {
            participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
        }
        participants[i].syncStart();
    }
    Map<String, Map<String, String>> errStateMap = new HashMap<>();
    errStateMap.put("TestDB0", new HashMap<String, String>());
    errStateMap.get("TestDB0").put("TestDB0_4", "localhost_12918");
    HelixClusterVerifier verifier = new BestPossibleExternalViewVerifier.Builder(clusterName).setZkAddr(ZK_ADDR).setErrStates(errStateMap).build();
    Assert.assertTrue(verifier.verify());
    // drop resource containing error partitions should invoke error->dropped transition
    // if error happens during error->dropped transition, partition should be disabled
    ClusterSetup.processCommandLineArgs(new String[] { "--zkSvr", ZK_ADDR, "--dropResource", clusterName, "TestDB0" });
    Thread.sleep(100);
    // make sure TestDB0_4 stay in ERROR state and is disabled
    Assert.assertTrue(verifier.verify());
    ZKHelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, new ZkBaseDataAccessor<ZNRecord>(_gZkClient));
    PropertyKey.Builder keyBuilder = accessor.keyBuilder();
    InstanceConfig config = accessor.getProperty(keyBuilder.instanceConfig("localhost_12918"));
    List<String> disabledPartitions = config.getDisabledPartitions();
    // System.out.println("disabledPartitions: " + disabledPartitions);
    Assert.assertEquals(disabledPartitions.size(), 1, "TestDB0_4 should be disabled");
    Assert.assertEquals(disabledPartitions.get(0), "TestDB0_4");
    // ExteranlView should have TestDB0_4->localhost_12918_>ERROR
    Thread.sleep(2000);
    ExternalView ev = accessor.getProperty(keyBuilder.externalView("TestDB0"));
    Set<String> partitions = ev.getPartitionSet();
    Assert.assertEquals(partitions.size(), 1, "Should have TestDB0_4->localhost_12918->ERROR");
    String errPartition = partitions.iterator().next();
    Assert.assertEquals(errPartition, "TestDB0_4");
    Map<String, String> stateMap = ev.getStateMap(errPartition);
    Assert.assertEquals(stateMap.size(), 1);
    Assert.assertEquals(stateMap.keySet().iterator().next(), "localhost_12918");
    Assert.assertEquals(stateMap.get("localhost_12918"), HelixDefinedState.ERROR.name());
    // localhost_12918 should have TestDB0_4 in ERROR state
    CurrentState cs = accessor.getProperty(keyBuilder.currentState(participants[0].getInstanceName(), participants[0].getSessionId(), "TestDB0"));
    Map<String, String> partitionStateMap = cs.getPartitionStateMap();
    Assert.assertEquals(partitionStateMap.size(), 1);
    Assert.assertEquals(partitionStateMap.keySet().iterator().next(), "TestDB0_4");
    Assert.assertEquals(partitionStateMap.get("TestDB0_4"), HelixDefinedState.ERROR.name());
    // all other participants should have cleaned up empty current state
    for (int i = 1; i < n; i++) {
        String instanceName = participants[i].getInstanceName();
        String sessionId = participants[i].getSessionId();
        Assert.assertNull(accessor.getProperty(keyBuilder.currentState(instanceName, sessionId, "TestDB0")));
    }
    // clean up
    controller.syncStop();
    for (int i = 0; i < n; i++) {
        participants[i].syncStop();
    }
    System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}