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

Example 56 with HelixManager

use of org.apache.helix.HelixManager in project ambry by linkedin.

the class BlobStoreTest method storeErrorTriggerDisableReplicaTest.

/**
 * Test that replica is correctly disabled when store is shut down due to disk I/O error.
 * @throws Exception
 */
@Test
public void storeErrorTriggerDisableReplicaTest() throws Exception {
    final String RESOURCE_NAME = "0";
    final String CLUSTER_NAME = "BlobStoreTest";
    // setup testing environment
    store.shutdown();
    List<TestUtils.ZkInfo> zkInfoList = new ArrayList<>();
    zkInfoList.add(new TestUtils.ZkInfo(null, "DC1", (byte) 0, 2199, false));
    JSONObject zkJson = constructZkLayoutJSON(zkInfoList);
    properties.setProperty("clustermap.cluster.name", CLUSTER_NAME);
    properties.setProperty("clustermap.datacenter.name", "DC1");
    properties.setProperty("clustermap.host.name", "localhost");
    properties.setProperty("clustermap.dcs.zk.connect.strings", zkJson.toString(2));
    properties.setProperty("store.io.error.count.to.trigger.shutdown", "1");
    properties.setProperty("store.replica.status.delegate.enable", "true");
    properties.setProperty("store.set.local.partition.state.enabled", "true");
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(new VerifiableProperties(properties));
    AtomicReference<InstanceConfig> instanceConfig = new AtomicReference<>(new InstanceConfig("localhost"));
    instanceConfig.get().setPort("2222");
    Map<String, List<String>> listMap = new HashMap<>();
    listMap.put(storeId, null);
    ZNRecord znRecord = new ZNRecord("localhost");
    znRecord.setListFields(listMap);
    IdealState idealState = new IdealState(znRecord);
    idealState.setRebalanceMode(IdealState.RebalanceMode.SEMI_AUTO);
    // mock helix related components
    HelixAdmin mockHelixAdmin = mock(HelixAdmin.class);
    when(mockHelixAdmin.getInstanceConfig(eq(CLUSTER_NAME), anyString())).then(invocation -> instanceConfig.get());
    when(mockHelixAdmin.getResourcesInCluster(eq(CLUSTER_NAME))).thenReturn(Collections.singletonList(RESOURCE_NAME));
    when(mockHelixAdmin.getResourceIdealState(eq(CLUSTER_NAME), eq(RESOURCE_NAME))).thenReturn(idealState);
    when(mockHelixAdmin.setInstanceConfig(any(), any(), any())).then(invocation -> {
        instanceConfig.set(invocation.getArgument(2));
        return true;
    });
    HelixManager mockHelixManager = mock(HelixManager.class);
    when(mockHelixManager.getClusterManagmentTool()).thenReturn(mockHelixAdmin);
    HelixFactory mockHelixFactory = new HelixFactory() {

        @Override
        public HelixManager getZKHelixManager(String clusterName, String instanceName, InstanceType instanceType, String zkAddr) {
            return mockHelixManager;
        }
    };
    MockHelixParticipant.metricRegistry = new MetricRegistry();
    MockHelixParticipant mockParticipant = new MockHelixParticipant(clusterMapConfig, mockHelixFactory);
    mockParticipant.overrideDisableReplicaMethod = false;
    ReplicaStatusDelegate replicaStatusDelegate = new ReplicaStatusDelegate(mockParticipant);
    BlobStore testStore = createBlobStore(getMockAmbryReplica(clusterMapConfig, tempDirStr), new StoreConfig(new VerifiableProperties(properties)), Collections.singletonList(replicaStatusDelegate));
    testStore.start();
    assertTrue("Store should start successfully", testStore.isStarted());
    // create corrupted write set
    MessageInfo corruptedInfo = new MessageInfo(getUniqueId(), PUT_RECORD_SIZE, Utils.getRandomShort(TestUtils.RANDOM), Utils.getRandomShort(TestUtils.RANDOM), Utils.Infinite_Time);
    MessageWriteSet corruptedWriteSet = new MockMessageWriteSet(Collections.singletonList(corruptedInfo), Collections.singletonList(ByteBuffer.allocate(PUT_RECORD_SIZE)), new StoreException(StoreException.IO_ERROR_STR, StoreErrorCodes.IOError));
    // 1. mock failure case
    when(mockHelixAdmin.getInstanceConfig(eq(CLUSTER_NAME), anyString())).thenReturn(null);
    // trigger store exception when calling store.put()
    try {
        testStore.put(corruptedWriteSet);
        fail("should throw exception");
    } catch (StoreException e) {
        assertEquals("Mismatch in error code", StoreErrorCodes.IOError, e.getErrorCode());
    }
    assertNull("Disabled partition list should be null as disabling replica didn't succeed", instanceConfig.get().getDisabledPartitions(RESOURCE_NAME));
    // 2. mock success case
    when(mockHelixAdmin.getInstanceConfig(eq(CLUSTER_NAME), anyString())).then(invocation -> instanceConfig.get());
    testStore.start();
    assertTrue("Store should start successfully", testStore.isStarted());
    try {
        testStore.put(corruptedWriteSet);
        fail("should throw exception");
    } catch (StoreException e) {
        assertEquals("Mismatch in error code", StoreErrorCodes.IOError, e.getErrorCode());
    }
    assertEquals("Disabled partition name is not expected", storeId, instanceConfig.get().getDisabledPartitions(RESOURCE_NAME).get(0));
    // verify "DISABLED" list in InstanceConfig has correct partition id.
    assertEquals("Disabled replica list is not expected", Collections.singletonList(storeId), getDisabledReplicas(instanceConfig.get()));
    // 3. mock disk is replaced case, restart should succeed
    testStore.start();
    assertNull("Disabled partition list should be null as restart will enable same replica", instanceConfig.get().getDisabledPartitions(RESOURCE_NAME));
    assertTrue("Disabled replica list should be empty", getDisabledReplicas(instanceConfig.get()).isEmpty());
    testStore.shutdown();
    reloadStore();
}

Example 57 with HelixManager

use of org.apache.helix.HelixManager in project ambry by linkedin.

the class HelixClusterManagerTest method listenCrossColoTest.

/**
 * Ensure that effects of the listenCrossColo config is as expected. When it is set to false, the Helix cluster manager
 * initializes fine, but listens to subsequent InstanceConfig changes in the local colo only.
 */
@Test
public void listenCrossColoTest() {
    assumeTrue(!useComposite);
    HelixClusterManager helixClusterManager = (HelixClusterManager) clusterManager;
    Counter instanceTriggerCounter = helixClusterManager.helixClusterManagerMetrics.instanceConfigChangeTriggerCount;
    Map<String, DcInfo> dcInfosMap = helixClusterManager.getDcInfosMap();
    Map<String, HelixManager> helixManagerMap = dcInfosMap.entrySet().stream().filter(e -> e.getValue() instanceof HelixDcInfo).collect(Collectors.toMap(Map.Entry::getKey, e -> ((HelixDcInfo) e.getValue()).helixManager));
    for (Map.Entry<String, HelixManager> entry : helixManagerMap.entrySet()) {
        if (entry.getKey().equals(localDc)) {
            assertTrue("Helix cluster manager should always be connected to the local Helix manager", entry.getValue().isConnected());
        } else {
            assertEquals("Helix cluster manager should be connected to the remote Helix managers if and only if listenCrossColo is" + "set to true", listenCrossColo, entry.getValue().isConnected());
        }
    }
    long instanceConfigChangeTriggerCount = instanceTriggerCounter.getCount();
    helixCluster.triggerInstanceConfigChangeNotification();
    assertEquals("Number of trigger count should be in accordance to listenCrossColo value", instanceConfigChangeTriggerCount + (listenCrossColo ? helixDcs.length : 1), instanceTriggerCounter.getCount());
    InstanceConfig remoteInstanceConfig = helixCluster.getInstanceConfigsFromDcs(helixDcs).stream().filter(e -> ClusterMapUtils.getDcName(e).equals(remoteDc)).findAny().get();
    DataNodeId remote = helixClusterManager.getDataNodeId(remoteInstanceConfig.getHostName(), Integer.parseInt(remoteInstanceConfig.getPort()));
    Set<PartitionId> writablePartitions = new HashSet<>(helixClusterManager.getWritablePartitionIds(null));
    PartitionId partitionIdToSealInRemote = helixClusterManager.getReplicaIds(remote).stream().filter(e -> writablePartitions.contains(e.getPartitionId())).findAny().get().getPartitionId();
    remoteInstanceConfig.getRecord().setListField(SEALED_STR, Collections.singletonList(partitionIdToSealInRemote.toPathString()));
    helixCluster.triggerInstanceConfigChangeNotification();
    assertEquals("If replica in remote is sealed, partition should be sealed if and only if listenCrossColo is true " + "and override is disabled", !listenCrossColo || overrideEnabled, helixClusterManager.getWritablePartitionIds(null).contains(partitionIdToSealInRemote));
}

Example 58 with HelixManager

use of org.apache.helix.HelixManager in project ambry by linkedin.

the class AmbryStateModelFactoryTest method testAmbryPartitionStateModel.

/**
 * Test that {@link HelixParticipantMetrics} keeps track of partition during state transition
 */
@Test
public void testAmbryPartitionStateModel() {
    assumeTrue(stateModelDef.equals(ClusterMapConfig.AMBRY_STATE_MODEL_DEF));
    MetricRegistry metricRegistry = new MetricRegistry();
    MockHelixParticipant.metricRegistry = metricRegistry;
    DataNodeConfig mockDataNodeConfig = Mockito.mock(DataNodeConfig.class);
    Set<String> disabledPartitionSet = new HashSet<>();
    Set<String> enabledPartitionSet = new HashSet<>();
    when(mockDataNodeConfig.getDisabledReplicas()).thenReturn(disabledPartitionSet);
    DataNodeConfigSource mockConfigSource = Mockito.mock(DataNodeConfigSource.class);
    when(mockConfigSource.get(anyString())).thenReturn(mockDataNodeConfig);
    HelixAdmin mockHelixAdmin = Mockito.mock(HelixAdmin.class);
    InstanceConfig mockInstanceConfig = Mockito.mock(InstanceConfig.class);
    doAnswer(invocation -> {
        String partitionName = invocation.getArgument(1);
        boolean enable = invocation.getArgument(2);
        if (enable) {
            enabledPartitionSet.add(partitionName);
        }
        return null;
    }).when(mockInstanceConfig).setInstanceEnabledForPartition(any(), any(), anyBoolean());
    when(mockHelixAdmin.getInstanceConfig(anyString(), anyString())).thenReturn(mockInstanceConfig);
    when(mockHelixAdmin.setInstanceConfig(anyString(), anyString(), any())).thenReturn(true);
    HelixManager mockHelixManager = Mockito.mock(HelixManager.class);
    when(mockHelixManager.getClusterManagmentTool()).thenReturn(mockHelixAdmin);
    MockHelixManagerFactory.overrideGetHelixManager = true;
    MockHelixParticipant.mockHelixFactory = new MockHelixManagerFactory(mockConfigSource, mockHelixManager);
    MockHelixParticipant mockHelixParticipant = new MockHelixParticipant(config);
    HelixParticipantMetrics participantMetrics = mockHelixParticipant.getHelixParticipantMetrics();
    String resourceName = "0";
    String partitionName = "1";
    Message mockMessage = Mockito.mock(Message.class);
    when(mockMessage.getPartitionName()).thenReturn(partitionName);
    when(mockMessage.getResourceName()).thenReturn(resourceName);
    AmbryPartitionStateModel stateModel = new AmbryPartitionStateModel(resourceName, partitionName, mockHelixParticipant, config);
    mockHelixParticipant.setInitialLocalPartitions(new HashSet<>(Collections.singletonList(partitionName)));
    assertEquals("Offline count is not expected", 1, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    // OFFLINE -> BOOTSTRAP
    stateModel.onBecomeBootstrapFromOffline(mockMessage, null);
    assertEquals("Offline count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    assertEquals("Bootstrap count should be 1", 1, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "bootstrapPartitionCount"));
    // BOOTSTRAP -> STANDBY
    stateModel.onBecomeStandbyFromBootstrap(mockMessage, null);
    assertEquals("Offline count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    assertEquals("Standby count should be 1", 1, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "standbyPartitionCount"));
    assertEquals("Bootstrap count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "bootstrapPartitionCount"));
    // STANDBY -> LEADER
    stateModel.onBecomeLeaderFromStandby(mockMessage, null);
    assertEquals("Offline count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    assertEquals("Leader count should be 1", 1, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "leaderPartitionCount"));
    assertEquals("Standby count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "standbyPartitionCount"));
    // LEADER -> STANDBY
    stateModel.onBecomeStandbyFromLeader(mockMessage, null);
    assertEquals("Offline count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    assertEquals("Standby count should be 1", 1, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "standbyPartitionCount"));
    assertEquals("Leader count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "leaderPartitionCount"));
    // STANDBY -> INACTIVE
    stateModel.onBecomeInactiveFromStandby(mockMessage, null);
    assertEquals("Offline count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    assertEquals("Inactive count should be 1", 1, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "inactivePartitionCount"));
    assertEquals("Standby count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "standbyPartitionCount"));
    // INACTIVE -> OFFLINE
    stateModel.onBecomeOfflineFromInactive(mockMessage, null);
    assertEquals("Offline count should be 1", 1, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    assertEquals("Inactive count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "inactivePartitionCount"));
    // OFFLINE -> DROPPED
    disabledPartitionSet.add(partitionName);
    stateModel.onBecomeDroppedFromOffline(mockMessage, null);
    assertEquals("Offline count should be 0", 0, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    assertEquals("Dropped count should be updated", 1, participantMetrics.partitionDroppedCount.getCount());
    assertTrue("Partition should be removed from disabled partition set", disabledPartitionSet.isEmpty());
    assertEquals("Mismatch in enabled partition", partitionName, enabledPartitionSet.iterator().next());
    // ERROR -> DROPPED
    stateModel.onBecomeDroppedFromError(mockMessage, null);
    assertEquals("Dropped count should be updated", 2, participantMetrics.partitionDroppedCount.getCount());
    // ERROR -> OFFLINE (this occurs when we use Helix API to reset certain partition in ERROR state)
    stateModel.onBecomeOfflineFromError(mockMessage, null);
    assertEquals("Offline count should be 1", 1, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    // reset method
    stateModel.reset();
    assertEquals("Offline count should be 1 after reset", 1, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    // call reset method again to mock the case where same partition is reset multiple times during zk disconnection or shutdown
    stateModel.reset();
    assertEquals("Offline count should still be 1 after reset twice", 1, getHelixParticipantMetricValue(metricRegistry, HelixParticipant.class.getName(), "offlinePartitionCount"));
    MockHelixManagerFactory.overrideGetHelixManager = false;
}

Example 59 with HelixManager

use of org.apache.helix.HelixManager in project pinot by linkedin.

the class ControllerRequestBuilderUtil method addFakeBrokerInstancesToAutoJoinHelixCluster.

public static void addFakeBrokerInstancesToAutoJoinHelixCluster(String helixClusterName, String zkServer, int numInstances, boolean isSingleTenant) throws Exception {
    for (int i = 0; i < numInstances; ++i) {
        final String brokerId = "Broker_localhost_" + i;
        final HelixManager helixZkManager = HelixManagerFactory.getZKHelixManager(helixClusterName, brokerId, InstanceType.PARTICIPANT, zkServer);
        final StateMachineEngine stateMachineEngine = helixZkManager.getStateMachineEngine();
        final StateModelFactory<?> stateModelFactory = new EmptyBrokerOnlineOfflineStateModelFactory();
        stateMachineEngine.registerStateModelFactory(EmptyBrokerOnlineOfflineStateModelFactory.getStateModelDef(), stateModelFactory);
        helixZkManager.connect();
        if (isSingleTenant) {
            helixZkManager.getClusterManagmentTool().addInstanceTag(helixClusterName, brokerId, ControllerTenantNameBuilder.getBrokerTenantNameForTenant(ControllerTenantNameBuilder.DEFAULT_TENANT_NAME));
        } else {
            helixZkManager.getClusterManagmentTool().addInstanceTag(helixClusterName, brokerId, UNTAGGED_BROKER_INSTANCE);
        }
        Thread.sleep(1000);
    }
}

Example 60 with HelixManager

use of org.apache.helix.HelixManager in project pinot by linkedin.

the class ControllerRequestBuilderUtil method addFakeDataInstancesToAutoJoinHelixCluster.

public static void addFakeDataInstancesToAutoJoinHelixCluster(String helixClusterName, String zkServer, int numInstances, boolean isSingleTenant, int adminPort) throws Exception {
    for (int i = 0; i < numInstances; ++i) {
        final String instanceId = "Server_localhost_" + i;
        final HelixManager helixZkManager = HelixManagerFactory.getZKHelixManager(helixClusterName, instanceId, InstanceType.PARTICIPANT, zkServer);
        final StateMachineEngine stateMachineEngine = helixZkManager.getStateMachineEngine();
        final StateModelFactory<?> stateModelFactory = new EmptySegmentOnlineOfflineStateModelFactory();
        stateMachineEngine.registerStateModelFactory(EmptySegmentOnlineOfflineStateModelFactory.getStateModelDef(), stateModelFactory);
        helixZkManager.connect();
        if (isSingleTenant) {
            helixZkManager.getClusterManagmentTool().addInstanceTag(helixClusterName, instanceId, TableNameBuilder.OFFLINE_TABLE_NAME_BUILDER.forTable(ControllerTenantNameBuilder.DEFAULT_TENANT_NAME));
            helixZkManager.getClusterManagmentTool().addInstanceTag(helixClusterName, instanceId, TableNameBuilder.REALTIME_TABLE_NAME_BUILDER.forTable(ControllerTenantNameBuilder.DEFAULT_TENANT_NAME));
        } else {
            helixZkManager.getClusterManagmentTool().addInstanceTag(helixClusterName, instanceId, UNTAGGED_SERVER_INSTANCE);
        }
        HelixConfigScope scope = new HelixConfigScopeBuilder(HelixConfigScope.ConfigScopeProperty.PARTICIPANT, helixClusterName).forParticipant(instanceId).build();
        Map<String, String> props = new HashMap<>();
        props.put(CommonConstants.Helix.Instance.ADMIN_PORT_KEY, String.valueOf(adminPort + i));
        helixZkManager.getClusterManagmentTool().setConfig(scope, props);
    }
}