Examples with ErrTransition org.apache.helix.mock.participant.ErrTransition used on opensource projects

Example 1 with ErrTransition

use of org.apache.helix.mock.participant.ErrTransition in project helix by apache.

the class TestResetInstance method testResetInstance.

@Test
public void testResetInstance() throws Exception {
    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()));
    // 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);
    // start controller
    ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
    controller.syncStart();
    Map<String, Set<String>> errPartitions = new HashMap<String, Set<String>>() {

        {
            put("SLAVE-MASTER", TestHelper.setOf("TestDB0_4"));
            put("OFFLINE-SLAVE", TestHelper.setOf("TestDB0_8"));
        }
    };
    // start mock participants
    MockParticipantManager[] participants = new MockParticipantManager[n];
    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(errPartitions));
        } else {
            participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
        }
        participants[i].syncStart();
    }
    // verify cluster
    Map<String, Map<String, String>> errStateMap = new HashMap<String, Map<String, String>>();
    errStateMap.put("TestDB0", new HashMap<String, String>());
    errStateMap.get("TestDB0").put("TestDB0_4", "localhost_12918");
    errStateMap.get("TestDB0").put("TestDB0_8", "localhost_12918");
    boolean result = ClusterStateVerifier.verifyByZkCallback((new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName, errStateMap)));
    Assert.assertTrue(result, "Cluster verification fails");
    // reset node "localhost_12918"
    participants[0].setTransition(null);
    String hostName = "localhost_12918";
    String instanceUrl = "http://localhost:" + ADMIN_PORT + "/clusters/" + clusterName + "/instances/" + hostName;
    Map<String, String> paramMap = new HashMap<String, String>();
    paramMap.put(JsonParameters.MANAGEMENT_COMMAND, ClusterSetup.resetInstance);
    TestHelixAdminScenariosRest.assertSuccessPostOperation(instanceUrl, paramMap, false);
    result = ClusterStateVerifier.verifyByZkCallback((new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName)));
    Assert.assertTrue(result, "Cluster verification fails");
    // 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 2 with ErrTransition

use of org.apache.helix.mock.participant.ErrTransition in project helix by apache.

the class TestResetPartitionState method testResetPartitionState.

@Test()
public void testResetPartitionState() throws Exception {
    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()));
    // 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);
    // start controller
    ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
    controller.syncStart();
    Map<String, Set<String>> errPartitions = new HashMap<String, Set<String>>();
    errPartitions.put("SLAVE-MASTER", TestHelper.setOf("TestDB0_4"));
    errPartitions.put("OFFLINE-SLAVE", TestHelper.setOf("TestDB0_8"));
    // start mock participants
    MockParticipantManager[] participants = new MockParticipantManager[n];
    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(errPartitions));
        } else {
            participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
        }
        participants[i].syncStart();
    }
    // verify cluster
    Map<String, Map<String, String>> errStateMap = new HashMap<String, Map<String, String>>();
    errStateMap.put("TestDB0", new HashMap<String, String>());
    errStateMap.get("TestDB0").put("TestDB0_4", "localhost_12918");
    errStateMap.get("TestDB0").put("TestDB0_8", "localhost_12918");
    boolean result = ClusterStateVerifier.verifyByZkCallback((new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName, errStateMap)));
    Assert.assertTrue(result, "Cluster verification fails");
    // reset a non-exist partition, should throw exception
    String hostName = "localhost_12918";
    String instanceUrl = getInstanceUrl(clusterName, hostName);
    Map<String, String> paramMap = new HashMap<String, String>();
    paramMap.put(JsonParameters.MANAGEMENT_COMMAND, ClusterSetup.resetPartition);
    paramMap.put(JsonParameters.PARTITION, "TestDB0_nonExist");
    paramMap.put(JsonParameters.RESOURCE, "TestDB0");
    LOG.info("IGNORABLE exception: test reset non-exist partition");
    TestHelixAdminScenariosRest.assertSuccessPostOperation(instanceUrl, paramMap, true);
    // reset 2 error partitions
    errPartitions.clear();
    participants[0].setTransition(new ErrTransitionWithResetCnt(errPartitions));
    clearStatusUpdate(clusterName, "localhost_12918", "TestDB0", "TestDB0_4");
    _errToOfflineInvoked.set(0);
    paramMap.put(JsonParameters.PARTITION, "TestDB0_4 TestDB0_8");
    TestHelixAdminScenariosRest.assertSuccessPostOperation(instanceUrl, paramMap, false);
    for (int i = 0; i < 10; i++) {
        // wait reset to be done
        Thread.sleep(400);
        LOG.info("IGNORABLE exception: test reset non-error partition");
        TestHelixAdminScenariosRest.assertSuccessPostOperation(instanceUrl, paramMap, true);
        result = ClusterStateVerifier.verifyByZkCallback(new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
        if (result == true) {
            break;
        }
    }
    Assert.assertTrue(result);
    Assert.assertEquals(_errToOfflineInvoked.get(), 2, "reset() should be invoked 2 times");
    // 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 3 with ErrTransition

use of org.apache.helix.mock.participant.ErrTransition in project helix by apache.

the class TestResetPartitionState method testResetPartitionState.

@Test()
public void testResetPartitionState() throws Exception {
    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()));
    // 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);
    // start controller
    ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
    controller.syncStart();
    Map<String, Set<String>> errPartitions = new HashMap<String, Set<String>>() {

        {
            put("SLAVE-MASTER", TestHelper.setOf("TestDB0_4"));
            put("OFFLINE-SLAVE", TestHelper.setOf("TestDB0_8"));
        }
    };
    // start mock participants
    MockParticipantManager[] participants = new MockParticipantManager[n];
    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(errPartitions));
        } else {
            participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
        }
        participants[i].syncStart();
    }
    // verify cluster
    Map<String, Map<String, String>> errStateMap = new HashMap<String, Map<String, String>>();
    errStateMap.put("TestDB0", new HashMap<String, String>());
    errStateMap.get("TestDB0").put("TestDB0_4", "localhost_12918");
    errStateMap.get("TestDB0").put("TestDB0_8", "localhost_12918");
    boolean result = ClusterStateVerifier.verifyByZkCallback((new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName, errStateMap)));
    Assert.assertTrue(result, "Cluster verification fails");
    // reset a non-exist partition, should throw exception
    try {
        String command = "--zkSvr " + ZK_ADDR + " --resetPartition " + clusterName + " localhost_12918 TestDB0 TestDB0_nonExist";
        ClusterSetup.processCommandLineArgs(command.split("\\s+"));
        Assert.fail("Should throw exception on reset a non-exist partition");
    } catch (Exception e) {
    // OK
    }
    // reset one error partition
    errPartitions.remove("SLAVE-MASTER");
    participants[0].setTransition(new ErrTransitionWithResetCnt(errPartitions));
    clearStatusUpdate(clusterName, "localhost_12918", "TestDB0", "TestDB0_4");
    _errToOfflineInvoked = 0;
    String command = "--zkSvr " + ZK_ADDR + " --resetPartition " + clusterName + " localhost_12918 TestDB0 TestDB0_4";
    ClusterSetup.processCommandLineArgs(command.split("\\s+"));
    // wait reset to be done
    Thread.sleep(200);
    try {
        ClusterSetup.processCommandLineArgs(command.split("\\s+"));
        Assert.fail("Should throw exception on reset a partition not in ERROR state");
    } catch (Exception e) {
    // OK
    }
    errStateMap.get("TestDB0").remove("TestDB0_4");
    result = ClusterStateVerifier.verifyByZkCallback((new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName, errStateMap)));
    Assert.assertTrue(result, "Cluster verification fails");
    Assert.assertEquals(_errToOfflineInvoked, 1);
    // reset the other error partition
    participants[0].setTransition(new ErrTransitionWithResetCnt(null));
    clearStatusUpdate(clusterName, "localhost_12918", "TestDB0", "TestDB0_8");
    command = "--zkSvr " + ZK_ADDR + " --resetPartition " + clusterName + " localhost_12918 TestDB0 TestDB0_8";
    ClusterSetup.processCommandLineArgs(command.split("\\s+"));
    result = ClusterStateVerifier.verifyByPolling(new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR, clusterName));
    Assert.assertTrue(result, "Cluster verification fails");
    Assert.assertEquals(_errToOfflineInvoked, 2, "Should reset 2 partitions");
    // 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 4 with ErrTransition

use of org.apache.helix.mock.participant.ErrTransition in project helix by apache.

the class TestStateTransitionThrottle method testTransitionThrottleOnErrorPartition.

@Test
public void testTransitionThrottleOnErrorPartition() throws Exception {
    String clusterName = getShortClassName() + "testMaxErrorPartition";
    MockParticipantManager[] participants = new MockParticipantManager[participantCount];
    System.out.println("START " + clusterName + " at " + new Date(System.currentTimeMillis()));
    final ZKHelixDataAccessor accessor = new ZKHelixDataAccessor(clusterName, new ZkBaseDataAccessor<ZNRecord>(_gZkClient));
    setupCluster(clusterName, accessor);
    // Set throttle config to enable throttling
    PropertyKey.Builder keyBuilder = accessor.keyBuilder();
    ClusterConfig clusterConfig = accessor.getProperty(accessor.keyBuilder().clusterConfig());
    clusterConfig.setResourcePriorityField("Name");
    List<StateTransitionThrottleConfig> throttleConfigs = new ArrayList<>();
    throttleConfigs.add(new StateTransitionThrottleConfig(StateTransitionThrottleConfig.RebalanceType.LOAD_BALANCE, StateTransitionThrottleConfig.ThrottleScope.CLUSTER, 100));
    throttleConfigs.add(new StateTransitionThrottleConfig(StateTransitionThrottleConfig.RebalanceType.RECOVERY_BALANCE, StateTransitionThrottleConfig.ThrottleScope.CLUSTER, 100));
    clusterConfig.setStateTransitionThrottleConfigs(throttleConfigs);
    accessor.setProperty(keyBuilder.clusterConfig(), clusterConfig);
    // set one partition to be always Error, so load balance won't be triggered
    Map<String, Set<String>> errPartitions = new HashMap<>();
    errPartitions.put("OFFLINE-SLAVE", TestHelper.setOf(resourceName + "_0"));
    // start part of participants
    for (int i = 0; i < participantCount - 1; i++) {
        participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, "localhost_" + (12918 + i));
        if (i == 0) {
            participants[i].setTransition(new ErrTransition(errPartitions));
        }
        participants[i].syncStart();
    }
    ClusterControllerManager controller = new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
    controller.syncStart();
    BestPossibleExternalViewVerifier verifier = new BestPossibleExternalViewVerifier.Builder(clusterName).setZkClient(_gZkClient).build();
    Assert.assertTrue(verifier.verify(3000));
    // Adding one more participant.
    participants[participantCount - 1] = new MockParticipantManager(ZK_ADDR, clusterName, "localhost_" + (12918 + participantCount - 1));
    participants[participantCount - 1].syncStart();
    // Since error partition exists, no load balance transition will be done
    Assert.assertFalse(pollForPartitionAssignment(accessor, participants[participantCount - 1], resourceName, 5000));
    // Update cluster config to tolerate error partition, so load balance transition will be done
    clusterConfig = accessor.getProperty(accessor.keyBuilder().clusterConfig());
    clusterConfig.setErrorPartitionThresholdForLoadBalance(1);
    accessor.setProperty(keyBuilder.clusterConfig(), clusterConfig);
    _gSetupTool.rebalanceResource(clusterName, resourceName, 3);
    Assert.assertTrue(pollForPartitionAssignment(accessor, participants[participantCount - 1], resourceName, 3000));
    // clean up
    controller.syncStop();
    for (int i = 0; i < participantCount; i++) {
        participants[i].syncStop();
    }
    System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));
}

Example 5 with ErrTransition

use of org.apache.helix.mock.participant.ErrTransition 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()));
}