Examples with HelixDataAccessor org.apache.helix.HelixDataAccessor used on opensource projects

Example 66 with HelixDataAccessor

use of org.apache.helix.HelixDataAccessor in project helix by apache.

the class TestReelectedPipelineCorrectness method testReelection.

@Test
public void testReelection() throws Exception {
    final int NUM_CONTROLLERS = 2;
    final int NUM_PARTICIPANTS = 4;
    final int NUM_PARTITIONS = 8;
    final int NUM_REPLICAS = 2;
    String className = TestHelper.getTestClassName();
    String methodName = TestHelper.getTestMethodName();
    String clusterName = className + "_" + methodName;
    System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
    ClusterSetup setupTool = new ClusterSetup(ZK_ADDR);
    // 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, "MasterSlave", RebalanceMode.FULL_AUTO, // do rebalance
    true);
    // configure distributed controllers
    String controllerCluster = clusterName + "_controllers";
    setupTool.addCluster(controllerCluster, true);
    for (int i = 0; i < NUM_CONTROLLERS; i++) {
        setupTool.addInstanceToCluster(controllerCluster, "controller_" + i);
    }
    setupTool.activateCluster(clusterName, controllerCluster, true);
    // 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();
    }
    // start controllers
    ClusterDistributedController[] controllers = new ClusterDistributedController[NUM_CONTROLLERS];
    for (int i = 0; i < NUM_CONTROLLERS; i++) {
        controllers[i] = new ClusterDistributedController(ZK_ADDR, controllerCluster, "controller_" + i);
        controllers[i].syncStart();
    }
    Thread.sleep(1000);
    // Ensure a balanced cluster
    boolean result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
    Assert.assertTrue(result);
    // Disable the leader, resulting in a leader election
    HelixDataAccessor accessor = participants[0].getHelixDataAccessor();
    LiveInstance leader = accessor.getProperty(accessor.keyBuilder().controllerLeader());
    int totalWait = 0;
    while (leader == null && totalWait < CHECK_TIMEOUT) {
        Thread.sleep(CHECK_INTERVAL);
        totalWait += CHECK_INTERVAL;
        leader = accessor.getProperty(accessor.keyBuilder().controllerLeader());
    }
    if (totalWait >= CHECK_TIMEOUT) {
        Assert.fail("No leader was ever elected!");
    }
    String leaderId = leader.getId();
    String standbyId = (leaderId.equals("controller_0")) ? "controller_1" : "controller_0";
    HelixAdmin admin = setupTool.getClusterManagementTool();
    admin.enableInstance(controllerCluster, leaderId, false);
    // Stop a participant to make sure that the leader election worked
    Thread.sleep(500);
    participants[0].syncStop();
    Thread.sleep(500);
    result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
    Assert.assertTrue(result);
    // Disable the original standby (leaving 0 active controllers) and kill another participant
    admin.enableInstance(controllerCluster, standbyId, false);
    Thread.sleep(500);
    participants[1].syncStop();
    // Also change the ideal state
    IdealState idealState = admin.getResourceIdealState(clusterName, "TestDB0");
    idealState.setMaxPartitionsPerInstance(1);
    admin.setResourceIdealState(clusterName, "TestDB0", idealState);
    Thread.sleep(500);
    // Also disable an instance in the main cluster
    admin.enableInstance(clusterName, "localhost_12920", false);
    // Re-enable the original leader
    admin.enableInstance(controllerCluster, leaderId, true);
    // Now check that both the ideal state and the live instances are adhered to by the rebalance
    Thread.sleep(500);
    result = ClusterStateVerifier.verifyByZkCallback(new BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
    Assert.assertTrue(result);
    // cleanup
    for (int i = 0; i < NUM_CONTROLLERS; i++) {
        controllers[i].syncStop();
    }
    for (int i = 2; i < NUM_PARTICIPANTS; i++) {
        participants[i].syncStop();
    }
    System.out.println("STOP " + clusterName + " at " + new Date(System.currentTimeMillis()));
}

Example 67 with HelixDataAccessor

use of org.apache.helix.HelixDataAccessor in project helix by apache.

the class TestHelixInstanceTag method testInstanceTag.

@Test
public void testInstanceTag() throws Exception {
    HelixManager manager = _controller;
    HelixDataAccessor accessor = manager.getHelixDataAccessor();
    String DB2 = "TestDB2";
    int partitions = 100;
    String DB2tag = "TestDB2_tag";
    int replica = 2;
    for (int i = 0; i < 2; i++) {
        String instanceName = "localhost_" + (12918 + i);
        _setupTool.getClusterManagementTool().addInstanceTag(CLUSTER_NAME, instanceName, DB2tag);
    }
    _setupTool.addResourceToCluster(CLUSTER_NAME, DB2, partitions, STATE_MODEL);
    _setupTool.rebalanceStorageCluster(CLUSTER_NAME, DB2, DB2tag, replica);
    boolean result = ClusterStateVerifier.verifyByZkCallback((new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, CLUSTER_NAME)));
    Assert.assertTrue(result, "Cluster verification fails");
    ExternalView ev = accessor.getProperty(accessor.keyBuilder().externalView(DB2));
    Set<String> hosts = new HashSet<String>();
    for (String p : ev.getPartitionSet()) {
        for (String hostName : ev.getStateMap(p).keySet()) {
            InstanceConfig config = accessor.getProperty(accessor.keyBuilder().instanceConfig(hostName));
            Assert.assertTrue(config.containsTag(DB2tag));
            hosts.add(hostName);
        }
    }
    Assert.assertEquals(hosts.size(), 2);
    String DB3 = "TestDB3";
    String DB3Tag = "TestDB3_tag";
    partitions = 10;
    replica = 3;
    for (int i = 1; i < 5; i++) {
        String instanceName = "localhost_" + (12918 + i);
        _setupTool.getClusterManagementTool().addInstanceTag(CLUSTER_NAME, instanceName, DB3Tag);
    }
    _setupTool.addResourceToCluster(CLUSTER_NAME, DB3, partitions, STATE_MODEL);
    _setupTool.rebalanceStorageCluster(CLUSTER_NAME, DB3, DB3Tag, replica);
    result = ClusterStateVerifier.verifyByZkCallback((new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, CLUSTER_NAME)));
    Assert.assertTrue(result, "Cluster verification fails");
    ev = accessor.getProperty(accessor.keyBuilder().externalView(DB3));
    hosts = new HashSet<String>();
    for (String p : ev.getPartitionSet()) {
        for (String hostName : ev.getStateMap(p).keySet()) {
            InstanceConfig config = accessor.getProperty(accessor.keyBuilder().instanceConfig(hostName));
            Assert.assertTrue(config.containsTag(DB3Tag));
            hosts.add(hostName);
        }
    }
    Assert.assertEquals(hosts.size(), 4);
}

Example 68 with HelixDataAccessor

use of org.apache.helix.HelixDataAccessor in project helix by apache.

the class TestMessagePartitionStateMismatch method testStateMismatch.

@Test
public void testStateMismatch() throws InterruptedException {
    // String controllerName = CONTROLLER_PREFIX + "_0";
    // _startCMResultMap.get(controllerName)._manager;
    HelixManager manager = _controller;
    HelixDataAccessor accessor = manager.getHelixDataAccessor();
    Builder kb = accessor.keyBuilder();
    ExternalView ev = accessor.getProperty(kb.externalView(TEST_DB));
    Map<String, LiveInstance> liveinstanceMap = accessor.getChildValuesMap(accessor.keyBuilder().liveInstances());
    for (String instanceName : liveinstanceMap.keySet()) {
        String sessionid = liveinstanceMap.get(instanceName).getSessionId();
        for (String partition : ev.getPartitionSet()) {
            if (ev.getStateMap(partition).containsKey(instanceName)) {
                String uuid = UUID.randomUUID().toString();
                Message message = new Message(MessageType.STATE_TRANSITION, uuid);
                boolean rand = new Random().nextInt(10) > 5;
                if (ev.getStateMap(partition).get(instanceName).equals("MASTER")) {
                    message.setSrcName(manager.getInstanceName());
                    message.setTgtName(instanceName);
                    message.setMsgState(MessageState.NEW);
                    message.setPartitionName(partition);
                    message.setResourceName(TEST_DB);
                    message.setFromState(rand ? "SLAVE" : "OFFLINE");
                    message.setToState(rand ? "MASTER" : "SLAVE");
                    message.setTgtSessionId(sessionid);
                    message.setSrcSessionId(manager.getSessionId());
                    message.setStateModelDef("MasterSlave");
                    message.setStateModelFactoryName("DEFAULT");
                } else if (ev.getStateMap(partition).get(instanceName).equals("SLAVE")) {
                    message.setSrcName(manager.getInstanceName());
                    message.setTgtName(instanceName);
                    message.setMsgState(MessageState.NEW);
                    message.setPartitionName(partition);
                    message.setResourceName(TEST_DB);
                    message.setFromState(rand ? "MASTER" : "OFFLINE");
                    message.setToState(rand ? "SLAVE" : "SLAVE");
                    message.setTgtSessionId(sessionid);
                    message.setSrcSessionId(manager.getSessionId());
                    message.setStateModelDef("MasterSlave");
                    message.setStateModelFactoryName("DEFAULT");
                }
                accessor.setProperty(accessor.keyBuilder().message(instanceName, message.getMsgId()), message);
            }
        }
    }
    Thread.sleep(3000);
    ExternalView ev2 = accessor.getProperty(kb.externalView(TEST_DB));
    Assert.assertTrue(ev.equals(ev2));
}

Example 69 with HelixDataAccessor

use of org.apache.helix.HelixDataAccessor 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 70 with HelixDataAccessor

use of org.apache.helix.HelixDataAccessor 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()));
}