Examples with InstanceConfig org.apache.helix.model.InstanceConfig used on opensource projects

Example 71 with InstanceConfig

use of org.apache.helix.model.InstanceConfig in project ambry by linkedin.

the class HelixClusterManagerTest method verifyInitialClusterChanges.

// Helpers
/**
 * Verify that {@link HelixClusterManager} receives and handles initial cluster changes (i.e InstanceConfig, IdealState
 * change), and populates the in-mem clustermap correctly.
 * @param clusterManager the {@link HelixClusterManager} to use for verification.
 * @param helixCluster the {@link MockHelixCluster} to provide cluster infos
 */
private void verifyInitialClusterChanges(HelixClusterManager clusterManager, MockHelixCluster helixCluster, String[] dcs) {
    // get in-mem data structures populated based on initial notification
    Map<String, Map<String, String>> partitionToResouceByDc = clusterManager.getPartitionToResourceMap();
    Map<String, Set<AmbryDataNode>> dataNodesByDc = clusterManager.getDcToDataNodesMap();
    for (String dc : dcs) {
        // 1. verify all instanceConfigs from Helix are present in cluster manager
        List<InstanceConfig> instanceConfigsInCluster = helixCluster.getInstanceConfigsFromDcs(new String[] { dc });
        assertEquals("Mismatch in number of instances", instanceConfigsInCluster.size(), dataNodesByDc.get(dc).size());
        Set<String> hostsFromClusterManager = dataNodesByDc.get(dc).stream().map(AmbryDataNode::getHostname).collect(Collectors.toSet());
        Set<String> hostsFromHelix = instanceConfigsInCluster.stream().map(InstanceConfig::getHostName).collect(Collectors.toSet());
        assertEquals("Mismatch in hosts set", hostsFromHelix, hostsFromClusterManager);
        // 2. verify all resources and partitions from Helix are present in cluster manager
        Map<String, String> partitionToResourceMap = partitionToResouceByDc.get(dc);
        MockHelixAdmin helixAdmin = helixCluster.getHelixAdminFromDc(dc);
        List<IdealState> idealStates = helixAdmin.getIdealStates();
        for (IdealState idealState : idealStates) {
            String resourceName = idealState.getResourceName();
            Set<String> partitionSet = idealState.getPartitionSet();
            for (String partitionStr : partitionSet) {
                assertEquals("Mismatch in resource name", resourceName, partitionToResourceMap.get(partitionStr));
            }
        }
    }
    if (cloudDc != null) {
        // If one cloud DC is present, there should be exactly one virtual replica for every partition.
        for (PartitionId partitionId : clusterManager.getAllPartitionIds(null)) {
            List<? extends ReplicaId> replicaIds = partitionId.getReplicaIds();
            int count = 0;
            for (ReplicaId replicaId : replicaIds) {
                if (replicaId instanceof CloudServiceReplica) {
                    count++;
                }
            }
            assertEquals("Unexpected number of CloudServiceReplicas in partition: " + replicaIds, 1, count);
        }
    }
}

Example 72 with InstanceConfig

use of org.apache.helix.model.InstanceConfig in project ambry by linkedin.

the class HelixClusterManagerTest method routingTableProviderChangeTest.

/**
 * Test that routing table change reflects correct state of each replica and {@link HelixClusterManager} is able to get
 * replica in required state.
 */
@Test
public void routingTableProviderChangeTest() throws Exception {
    assumeTrue(!useComposite && !overrideEnabled && !listenCrossColo);
    // Change zk connect strings to ensure HelixClusterManager sees local DC only
    JSONObject zkJson = constructZkLayoutJSON(Collections.singletonList(dcsToZkInfo.get(localDc)));
    Properties props = new Properties();
    props.setProperty("clustermap.host.name", hostname);
    props.setProperty("clustermap.cluster.name", clusterNamePrefixInHelix + clusterNameStatic);
    props.setProperty("clustermap.datacenter.name", localDc);
    props.setProperty("clustermap.port", Integer.toString(portNum));
    props.setProperty("clustermap.dcs.zk.connect.strings", zkJson.toString(2));
    props.setProperty("clustermap.current.xid", Long.toString(CURRENT_XID));
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(new VerifiableProperties(props));
    // Mock metricRegistry here to introduce a latch based counter for testing purpose
    metricRegistry = new MetricRegistry();
    HelixClusterManager helixClusterManager = new HelixClusterManager(clusterMapConfig, selfInstanceName, new MockHelixManagerFactory(helixCluster, null, null), metricRegistry);
    Map<String, RoutingTableSnapshot> snapshotsByDc = helixClusterManager.getRoutingTableSnapshots();
    RoutingTableSnapshot localDcSnapshot = snapshotsByDc.get(localDc);
    Set<InstanceConfig> instanceConfigsInSnapshot = new HashSet<>(localDcSnapshot.getInstanceConfigs());
    Set<InstanceConfig> instanceConfigsInCluster = new HashSet<>(helixCluster.getInstanceConfigsFromDcs(new String[] { localDc }));
    assertEquals("Mismatch in instance configs", instanceConfigsInCluster, instanceConfigsInSnapshot);
    // verify leader replica of each partition is correct
    verifyLeaderReplicasInDc(helixClusterManager, localDc);
    // test live instance triggered routing table change
    // we purposely bring down one instance and wait for expected number of live instance unless times out.
    int initialLiveCnt = localDcSnapshot.getLiveInstances().size();
    MockHelixAdmin mockHelixAdmin = helixCluster.getHelixAdminFromDc(localDc);
    String instance = instanceConfigsInCluster.stream().filter(insConfig -> !insConfig.getInstanceName().equals(selfInstanceName)).findFirst().get().getInstanceName();
    mockHelixAdmin.bringInstanceDown(instance);
    mockHelixAdmin.triggerRoutingTableNotification();
    int sleepCnt = 0;
    while (helixClusterManager.getRoutingTableSnapshots().get(localDc).getLiveInstances().size() != initialLiveCnt - 1) {
        assertTrue("Routing table change (triggered by bringing down node) didn't come within 1 sec", sleepCnt < 5);
        Thread.sleep(200);
        sleepCnt++;
    }
    // then bring up the same instance, the number of live instances should equal to initial count
    mockHelixAdmin.bringInstanceUp(instance);
    mockHelixAdmin.triggerRoutingTableNotification();
    sleepCnt = 0;
    while (helixClusterManager.getRoutingTableSnapshots().get(localDc).getLiveInstances().size() != initialLiveCnt) {
        assertTrue("Routing table change (triggered by bringing up node) didn't come within 1 sec", sleepCnt < 5);
        Thread.sleep(200);
        sleepCnt++;
    }
    // randomly choose a partition and change the leader replica of it in cluster
    List<? extends PartitionId> defaultPartitionIds = helixClusterManager.getAllPartitionIds(DEFAULT_PARTITION_CLASS);
    PartitionId partitionToChange = defaultPartitionIds.get((new Random()).nextInt(defaultPartitionIds.size()));
    String currentLeaderInstance = mockHelixAdmin.getPartitionToLeaderReplica().get(partitionToChange.toPathString());
    int currentLeaderPort = Integer.parseInt(currentLeaderInstance.split("_")[1]);
    String newLeaderInstance = mockHelixAdmin.getInstancesForPartition(partitionToChange.toPathString()).stream().filter(k -> !k.equals(currentLeaderInstance)).findFirst().get();
    mockHelixAdmin.changeLeaderReplicaForPartition(partitionToChange.toPathString(), newLeaderInstance);
    mockHelixAdmin.triggerRoutingTableNotification();
    sleepCnt = 0;
    while (partitionToChange.getReplicaIdsByState(ReplicaState.LEADER, localDc).get(0).getDataNodeId().getPort() == currentLeaderPort) {
        assertTrue("Routing table change (triggered by leadership change) didn't come within 1 sec", sleepCnt < 5);
        Thread.sleep(200);
        sleepCnt++;
    }
    verifyLeaderReplicasInDc(helixClusterManager, localDc);
    helixClusterManager.close();
}

Example 73 with InstanceConfig

use of org.apache.helix.model.InstanceConfig 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 74 with InstanceConfig

use of org.apache.helix.model.InstanceConfig in project pinot by linkedin.

the class Instance method toInstanceConfig.

public InstanceConfig toInstanceConfig() {
    final InstanceConfig iConfig = new InstanceConfig(toInstanceId());
    iConfig.setHostName(_host);
    iConfig.setPort(_port);
    iConfig.setInstanceEnabled(true);
    iConfig.addTag(getTagOrDefaultTag());
    return iConfig;
}

Example 75 with InstanceConfig

use of org.apache.helix.model.InstanceConfig in project pinot by linkedin.

the class PinotInstanceRestletResource method getInstanceInformation.

/**
   * Gets the information for an instance.
   *
   * @param instanceName The instance name
   */
@HttpVerb("get")
@Summary("Gets information for an instance")
@Tags({ "instance" })
@Paths({ "/instances/{instanceName}", "/instances/{instanceName}/" })
@Responses({ @Response(statusCode = "200", description = "Information about the specified instance"), @Response(statusCode = "404", description = "The specified instance does not exist"), @Response(statusCode = "500", description = "There was an error while fetching information for the given instance") })
private Representation getInstanceInformation(@Parameter(name = "instanceName", description = "The name of the instance (eg. Server_1.2.3.4_1234 or Broker_someHost.example.com_2345)", in = "path", required = true) String instanceName) {
    try {
        if (!_pinotHelixResourceManager.instanceExists(instanceName)) {
            setStatus(Status.CLIENT_ERROR_NOT_FOUND);
            return new StringRepresentation("Error: Instance " + instanceName + " not found.");
        }
        InstanceConfig instanceConfig = _pinotHelixResourceManager.getHelixInstanceConfig(instanceName);
        JSONObject response = new JSONObject();
        response.put("instanceName", instanceConfig.getInstanceName());
        response.put("hostName", instanceConfig.getHostName());
        response.put("enabled", instanceConfig.getInstanceEnabled());
        response.put("port", instanceConfig.getPort());
        response.put("tags", new JSONArray(instanceConfig.getTags()));
        return new StringRepresentation(response.toString());
    } catch (Exception e) {
        LOGGER.warn("Caught exception while fetching information for instance {}", instanceName, e);
        setStatus(Status.SERVER_ERROR_INTERNAL);
        return new StringRepresentation("{}");
    }
}