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

Example 21 with ClusterControllerManager

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

the class TestInvalidResourceRebalance method testResourceRebalanceSkipped.

/**
 * Ensure that the Helix controller doesn't attempt to rebalance resources with invalid ideal
 * states
 */
@Test
public void testResourceRebalanceSkipped() throws Exception {
    final int NUM_PARTICIPANTS = 2;
    final int NUM_PARTITIONS = 4;
    final int NUM_REPLICAS = 2;
    final String RESOURCE_NAME = "TestDB0";
    String className = TestHelper.getTestClassName();
    String methodName = TestHelper.getTestMethodName();
    String clusterName = className + "_" + methodName;
    System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
    // Set up cluster
    // 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
    NUM_PARTITIONS, // number of nodes
    NUM_PARTICIPANTS, // replicas
    NUM_REPLICAS, // use SEMI_AUTO mode
    "MasterSlave", // use SEMI_AUTO mode
    RebalanceMode.SEMI_AUTO, // do rebalance
    true);
    // start controller
    ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller");
    controller.syncStart();
    // add the ideal state spec (prevents non-CUSTOMIZED MasterSlave ideal states)
    HelixAdmin helixAdmin = controller.getClusterManagmentTool();
    Map<String, String> properties = Maps.newHashMap();
    properties.put("IdealStateRule!sampleRuleName", "IDEAL_STATE_MODE=CUSTOMIZED,STATE_MODEL_DEF_REF=MasterSlave");
    helixAdmin.setConfig(new HelixConfigScopeBuilder(ConfigScopeProperty.CLUSTER).forCluster(clusterName).build(), properties);
    // start participants
    MockParticipantManager[] participants = new MockParticipantManager[NUM_PARTICIPANTS];
    for (int i = 0; i < NUM_PARTICIPANTS; i++) {
        final String instanceName = "localhost_" + (12918 + i);
        participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
        participants[i].syncStart();
    }
    Thread.sleep(1000);
    boolean result = ClusterStateVerifier.verifyByZkCallback(new EmptyZkVerifier(clusterName, RESOURCE_NAME));
    Assert.assertTrue(result, "External view and current state must be empty");
    // cleanup
    for (int i = 0; i < NUM_PARTICIPANTS; i++) {
        participants[i].syncStop();
    }
    controller.syncStop();
}

Example 22 with ClusterControllerManager

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

the class TestNullReplica method testNullReplica.

@Test
public void testNullReplica() throws Exception {
    // Logger.getRootLogger().setLevel(Level.INFO);
    String className = TestHelper.getTestClassName();
    String methodName = TestHelper.getTestMethodName();
    String clusterName = className + "_" + methodName;
    System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
    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);
    // set replica in ideal state to null
    String idealStatePath = PropertyPathBuilder.idealState(clusterName, "TestDB0");
    ZNRecord idealState = _gZkClient.readData(idealStatePath);
    idealState.getSimpleFields().remove(IdealState.IdealStateProperty.REPLICAS.toString());
    _gZkClient.writeData(idealStatePath, idealState);
    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 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 " + clusterName + " at " + new Date(System.currentTimeMillis()));
}

Example 23 with ClusterControllerManager

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

the class TestPartitionMovementThrottle 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(_gZkClient);
    // 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);
    }
    // add dummy participants
    for (int i = 0; i < NODE_NR; i++) {
        String instanceName = PARTICIPANT_PREFIX + "_" + (START_PORT + i);
        MockParticipantManager participant = new MockParticipantManager(ZK_ADDR, CLUSTER_NAME, instanceName);
        participant.setTransition(new DelayedTransition());
        _participants[i] = participant;
    }
    _configAccessor = new ConfigAccessor(_gZkClient);
    // start controller
    String controllerName = CONTROLLER_PREFIX + "_0";
    _controller = new ClusterControllerManager(ZK_ADDR, CLUSTER_NAME, controllerName);
    _controller.syncStart();
    setupThrottleConfig();
}

Example 24 with ClusterControllerManager

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

the class TestPauseSignal method testPauseSignal.

@Test()
public void testPauseSignal() throws Exception {
    // Logger.getRootLogger().setLevel(Level.INFO);
    String className = TestHelper.getTestClassName();
    String methodName = TestHelper.getTestMethodName();
    final String clusterName = className + "_" + methodName;
    System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
    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);
    // start controller
    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 BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
    Assert.assertTrue(result);
    // pause the cluster and make sure pause is persistent
    ZkClient zkClient = new ZkClient(ZK_ADDR);
    zkClient.setZkSerializer(new ZNRecordSerializer());
    final HelixDataAccessor tmpAccessor = new ZKHelixDataAccessor(clusterName, new ZkBaseDataAccessor<ZNRecord>(zkClient));
    String cmd = "-zkSvr " + ZK_ADDR + " -enableCluster " + clusterName + " false";
    ClusterSetup.processCommandLineArgs(cmd.split(" "));
    tmpAccessor.setProperty(tmpAccessor.keyBuilder().pause(), new PauseSignal("pause"));
    zkClient.close();
    // wait for controller to be signaled by pause
    Thread.sleep(1000);
    // add a new resource group
    ClusterSetup setupTool = new ClusterSetup(ZK_ADDR);
    setupTool.addResourceToCluster(clusterName, "TestDB1", 10, "MasterSlave");
    setupTool.rebalanceStorageCluster(clusterName, "TestDB1", 3);
    // make sure TestDB1 external view is empty
    TestHelper.verifyWithTimeout("verifyEmptyCurStateAndExtView", 1000, clusterName, "TestDB1", TestHelper.<String>setOf("localhost_12918", "localhost_12919", "localhost_12920", "localhost_12921", "localhost_12922"), ZK_ADDR);
    // resume controller
    final HelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, new ZkBaseDataAccessor<ZNRecord>(_gZkClient));
    cmd = "-zkSvr " + ZK_ADDR + " -enableCluster " + clusterName + " true";
    ClusterSetup.processCommandLineArgs(cmd.split(" "));
    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 " + clusterName + " at " + new Date(System.currentTimeMillis()));
}

Example 25 with ClusterControllerManager

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

the class TestRebalancePipeline method testMsgTriggeredRebalance.

@Test
public void testMsgTriggeredRebalance() throws Exception {
    String clusterName = "CLUSTER_" + _className + "_msgTrigger";
    System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
    HelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, new ZkBaseDataAccessor<ZNRecord>(_gZkClient));
    HelixManager manager = new DummyClusterManager(clusterName, accessor);
    ClusterEvent event = new ClusterEvent(ClusterEventType.Unknown);
    refreshClusterConfig(clusterName, accessor);
    final String resourceName = "testResource_dup";
    String[] resourceGroups = new String[] { resourceName };
    TestHelper.setupEmptyCluster(_gZkClient, clusterName);
    // ideal state: node0 is MASTER, node1 is SLAVE
    // replica=2 means 1 master and 1 slave
    setupIdealState(clusterName, new int[] { 0, 1 }, resourceGroups, 1, 2);
    setupStateModel(clusterName);
    setupInstances(clusterName, new int[] { 0, 1 });
    setupLiveInstances(clusterName, new int[] { 0, 1 });
    ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
    controller.syncStart();
    // round1: controller sends O->S to both node0 and node1
    Thread.sleep(1000);
    Builder keyBuilder = accessor.keyBuilder();
    List<String> messages = accessor.getChildNames(keyBuilder.messages("localhost_0"));
    Assert.assertEquals(messages.size(), 1);
    messages = accessor.getChildNames(keyBuilder.messages("localhost_1"));
    Assert.assertEquals(messages.size(), 1);
    // round2: node0 and node1 update current states but not removing messages
    // Since controller's rebalancer pipeline will GC pending messages after timeout, and both hosts
    // update current states to SLAVE, controller will send out rebalance message to
    // have one host to become master
    setCurrentState(clusterName, "localhost_0", resourceName, resourceName + "_0", "session_0", "SLAVE", true);
    setCurrentState(clusterName, "localhost_1", resourceName, resourceName + "_0", "session_1", "SLAVE", true);
    // Controller has 3s timeout, so after 1s, controller should not have GCed message
    Thread.sleep(1000);
    Assert.assertEquals(accessor.getChildValues(keyBuilder.messages("localhost_0")).size(), 1);
    Assert.assertEquals(accessor.getChildValues(keyBuilder.messages("localhost_1")).size(), 1);
    // After another 2 second, controller should cleanup messages and continue to rebalance
    Thread.sleep(3000);
    // Manually trigger another rebalance by touching current state
    setCurrentState(clusterName, "localhost_0", resourceName, resourceName + "_0", "session_0", "SLAVE");
    Thread.sleep(1000);
    List<Message> host0Msg = accessor.getChildValues(keyBuilder.messages("localhost_0"));
    List<Message> host1Msg = accessor.getChildValues(keyBuilder.messages("localhost_1"));
    List<Message> allMsgs = new ArrayList<>(host0Msg);
    allMsgs.addAll(host1Msg);
    Assert.assertEquals(allMsgs.size(), 1);
    Assert.assertEquals(allMsgs.get(0).getToState(), "MASTER");
    Assert.assertEquals(allMsgs.get(0).getFromState(), "SLAVE");
    // round3: node0 changes state to master, but failed to delete message,
    // controller will clean it up
    setCurrentState(clusterName, "localhost_0", resourceName, resourceName + "_0", "session_0", "MASTER", true);
    Thread.sleep(3500);
    // touch current state to trigger rebalance
    setCurrentState(clusterName, "localhost_0", resourceName, resourceName + "_0", "session_0", "MASTER", false);
    Thread.sleep(1000);
    Assert.assertTrue(accessor.getChildNames(keyBuilder.messages("localhost_0")).isEmpty());
    // round4: node0 has duplicated but valid message, i.e. there is a P2P message sent to it
    // due to error in the triggered pipeline, controller should remove duplicated message
    // immediately as the partition has became master 3 sec ago (there is already a timeout)
    Message sourceMsg = allMsgs.get(0);
    Message dupMsg = new Message(sourceMsg.getMsgType(), UUID.randomUUID().toString());
    dupMsg.getRecord().setSimpleFields(sourceMsg.getRecord().getSimpleFields());
    dupMsg.getRecord().setListFields(sourceMsg.getRecord().getListFields());
    dupMsg.getRecord().setMapFields(sourceMsg.getRecord().getMapFields());
    accessor.setProperty(dupMsg.getKey(accessor.keyBuilder(), dupMsg.getTgtName()), dupMsg);
    Thread.sleep(1000);
    messages = accessor.getChildNames(keyBuilder.messages("localhost_0"));
    Assert.assertTrue(messages.isEmpty());
    // round5: node0 has completely invalid message, controller should immediately delete it
    dupMsg.setFromState("SLAVE");
    dupMsg.setToState("OFFLINE");
    accessor.setProperty(dupMsg.getKey(accessor.keyBuilder(), dupMsg.getTgtName()), dupMsg);
    Thread.sleep(1000);
    messages = accessor.getChildNames(keyBuilder.messages("localhost_0"));
    Assert.assertTrue(messages.isEmpty());
    System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}