Examples with ClusterMapConfig com.github.ambry.config.ClusterMapConfig used on opensource projects

Example 26 with ClusterMapConfig

use of com.github.ambry.config.ClusterMapConfig in project ambry by linkedin.

the class ReplicationTest method addAndRemoveReplicaTest.

/**
 * Test dynamically add/remove replica in {@link ReplicationManager}
 * @throws Exception
 */
@Test
public void addAndRemoveReplicaTest() throws Exception {
    MockClusterMap clusterMap = new MockClusterMap();
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(verifiableProperties);
    StoreConfig storeConfig = new StoreConfig(verifiableProperties);
    DataNodeId dataNodeId = clusterMap.getDataNodeIds().get(0);
    MockStoreKeyConverterFactory storeKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
    storeKeyConverterFactory.setConversionMap(new HashMap<>());
    StorageManager storageManager = new StorageManager(storeConfig, new DiskManagerConfig(verifiableProperties), Utils.newScheduler(1, true), new MetricRegistry(), null, clusterMap, dataNodeId, null, null, new MockTime(), null, new InMemAccountService(false, false));
    storageManager.start();
    MockReplicationManager replicationManager = new MockReplicationManager(replicationConfig, clusterMapConfig, storeConfig, storageManager, clusterMap, dataNodeId, storeKeyConverterFactory, null);
    ReplicaId replicaToTest = clusterMap.getReplicaIds(dataNodeId).get(0);
    // Attempting to add replica that already exists should fail
    assertFalse("Adding an existing replica should fail", replicationManager.addReplica(replicaToTest));
    // Create a brand new replica that sits on one of the disk of datanode, add it into replication manager
    PartitionId newPartition = clusterMap.createNewPartition(clusterMap.getDataNodes());
    for (ReplicaId replicaId : newPartition.getReplicaIds()) {
        if (replicaId.getDataNodeId() == dataNodeId) {
            replicaToTest = replicaId;
            break;
        }
    }
    // Before adding replica, partitionToPartitionInfo and mountPathToPartitionInfos should not contain new partition
    assertFalse("partitionToPartitionInfo should not contain new partition", replicationManager.getPartitionToPartitionInfoMap().containsKey(newPartition));
    for (PartitionInfo partitionInfo : replicationManager.getMountPathToPartitionInfosMap().get(replicaToTest.getMountPath())) {
        assertNotSame("mountPathToPartitionInfos should not contain new partition", partitionInfo.getPartitionId(), newPartition);
    }
    // Add new replica to replication manager
    assertTrue("Adding new replica to replication manager should succeed", replicationManager.addReplica(replicaToTest));
    // After adding replica, partitionToPartitionInfo and mountPathToPartitionInfos should contain new partition
    assertTrue("partitionToPartitionInfo should contain new partition", replicationManager.getPartitionToPartitionInfoMap().containsKey(newPartition));
    Optional<PartitionInfo> newPartitionInfo = replicationManager.getMountPathToPartitionInfosMap().get(replicaToTest.getMountPath()).stream().filter(partitionInfo -> partitionInfo.getPartitionId() == newPartition).findAny();
    assertTrue("mountPathToPartitionInfos should contain new partition info", newPartitionInfo.isPresent());
    // Verify that all remoteReplicaInfos of new added replica have assigned thread
    for (RemoteReplicaInfo remoteReplicaInfo : newPartitionInfo.get().getRemoteReplicaInfos()) {
        assertNotNull("The remote replica should be assigned to one replica thread", remoteReplicaInfo.getReplicaThread());
    }
    // Remove replica
    assertTrue("Remove replica from replication manager should succeed", replicationManager.removeReplica(replicaToTest));
    // Verify replica is removed, so partitionToPartitionInfo and mountPathToPartitionInfos should not contain new partition
    assertFalse("partitionToPartitionInfo should not contain new partition", replicationManager.getPartitionToPartitionInfoMap().containsKey(newPartition));
    for (PartitionInfo partitionInfo : replicationManager.getMountPathToPartitionInfosMap().get(replicaToTest.getMountPath())) {
        assertNotSame("mountPathToPartitionInfos should not contain new partition", partitionInfo.getPartitionId(), newPartition);
    }
    // Verify that none of remoteReplicaInfo should have assigned thread
    for (RemoteReplicaInfo remoteReplicaInfo : newPartitionInfo.get().getRemoteReplicaInfos()) {
        assertNull("The remote replica should be assigned to one replica thread", remoteReplicaInfo.getReplicaThread());
    }
    // Remove the same replica that doesn't exist should be no-op
    ReplicationManager mockManager = Mockito.spy(replicationManager);
    assertFalse("Remove non-existent replica should return false", replicationManager.removeReplica(replicaToTest));
    verify(mockManager, never()).removeRemoteReplicaInfoFromReplicaThread(anyList());
    storageManager.shutdown();
}

Example 27 with ClusterMapConfig

use of com.github.ambry.config.ClusterMapConfig in project ambry by linkedin.

the class ReplicationTest method replicaResumeDecommissionTest.

/**
 * Test that resuming decommission on certain replica behaves correctly.
 * @throws Exception
 */
@Test
public void replicaResumeDecommissionTest() throws Exception {
    MockClusterMap clusterMap = new MockClusterMap();
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(verifiableProperties);
    MockHelixParticipant.metricRegistry = new MetricRegistry();
    MockHelixParticipant mockHelixParticipant = Mockito.spy(new MockHelixParticipant(clusterMapConfig));
    doNothing().when(mockHelixParticipant).setPartitionDisabledState(anyString(), anyBoolean());
    // choose a replica on local node and put decommission file into its dir
    ReplicaId localReplica = clusterMap.getReplicaIds(clusterMap.getDataNodeIds().get(0)).get(0);
    String partitionName = localReplica.getPartitionId().toPathString();
    File decommissionFile = new File(localReplica.getReplicaPath(), "decommission_in_progress");
    assertTrue("Can't create decommission file", decommissionFile.createNewFile());
    Pair<StorageManager, ReplicationManager> managers = createStorageManagerAndReplicationManager(clusterMap, clusterMapConfig, mockHelixParticipant);
    StorageManager storageManager = managers.getFirst();
    // failure case 1: store is not started when resuming decommission
    storageManager.shutdownBlobStore(localReplica.getPartitionId());
    try {
        mockHelixParticipant.onPartitionBecomeDroppedFromOffline(partitionName);
        fail("should fail");
    } catch (StateTransitionException e) {
        assertEquals("Mismatch in error code", ReplicaOperationFailure, e.getErrorCode());
    }
    storageManager.startBlobStore(localReplica.getPartitionId());
    // failure case 2: fail to remove replica from InstanceConfig in Helix
    AmbryReplicaSyncUpManager replicaSyncUpManager = (AmbryReplicaSyncUpManager) mockHelixParticipant.getReplicaSyncUpManager();
    mockHelixParticipant.updateNodeInfoReturnVal = false;
    CountDownLatch executionLatch = new CountDownLatch(1);
    AtomicBoolean exceptionOccurred = new AtomicBoolean(false);
    Utils.newThread(() -> {
        try {
            mockHelixParticipant.onPartitionBecomeDroppedFromOffline(partitionName);
            fail("should fail because updating node info returns false");
        } catch (StateTransitionException e) {
            exceptionOccurred.getAndSet(true);
            assertEquals("Mismatch in error code", ReplicaOperationFailure, e.getErrorCode());
        } finally {
            executionLatch.countDown();
        }
    }, false).start();
    while (!replicaSyncUpManager.getPartitionToDeactivationLatch().containsKey(partitionName)) {
        Thread.sleep(100);
    }
    replicaSyncUpManager.onDeactivationComplete(localReplica);
    while (!replicaSyncUpManager.getPartitionToDisconnectionLatch().containsKey(partitionName)) {
        Thread.sleep(100);
    }
    replicaSyncUpManager.onDisconnectionComplete(localReplica);
    assertTrue("Offline-To-Dropped transition didn't complete within 1 sec", executionLatch.await(1, TimeUnit.SECONDS));
    assertTrue("State transition exception should be thrown", exceptionOccurred.get());
    mockHelixParticipant.updateNodeInfoReturnVal = null;
    storageManager.startBlobStore(localReplica.getPartitionId());
    // success case
    mockHelixParticipant.mockStatsManagerListener = Mockito.mock(PartitionStateChangeListener.class);
    doNothing().when(mockHelixParticipant.mockStatsManagerListener).onPartitionBecomeDroppedFromOffline(anyString());
    mockHelixParticipant.registerPartitionStateChangeListener(StateModelListenerType.StatsManagerListener, mockHelixParticipant.mockStatsManagerListener);
    CountDownLatch participantLatch = new CountDownLatch(1);
    Utils.newThread(() -> {
        mockHelixParticipant.onPartitionBecomeDroppedFromOffline(partitionName);
        participantLatch.countDown();
    }, false).start();
    while (!replicaSyncUpManager.getPartitionToDeactivationLatch().containsKey(partitionName)) {
        Thread.sleep(100);
    }
    replicaSyncUpManager.onDeactivationComplete(localReplica);
    while (!replicaSyncUpManager.getPartitionToDisconnectionLatch().containsKey(partitionName)) {
        Thread.sleep(100);
    }
    replicaSyncUpManager.onDisconnectionComplete(localReplica);
    assertTrue("Offline-To-Dropped transition didn't complete within 1 sec", participantLatch.await(1, TimeUnit.SECONDS));
    // verify stats manager listener is called
    verify(mockHelixParticipant.mockStatsManagerListener).onPartitionBecomeDroppedFromOffline(anyString());
    // verify setPartitionDisabledState method is called
    verify(mockHelixParticipant).setPartitionDisabledState(partitionName, false);
    File storeDir = new File(localReplica.getReplicaPath());
    assertFalse("Store dir should not exist", storeDir.exists());
    storageManager.shutdown();
}

Example 28 with ClusterMapConfig

use of com.github.ambry.config.ClusterMapConfig in project ambry by linkedin.

the class ReplicationTest method replicaFromInactiveToOfflineTest.

/**
 * Test INACTIVE -> OFFLINE transition on existing replica (both success and failure cases)
 */
@Test
public void replicaFromInactiveToOfflineTest() throws Exception {
    MockClusterMap clusterMap = new MockClusterMap();
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(verifiableProperties);
    MockHelixParticipant.metricRegistry = new MetricRegistry();
    MockHelixParticipant mockHelixParticipant = new MockHelixParticipant(clusterMapConfig);
    Pair<StorageManager, ReplicationManager> managers = createStorageManagerAndReplicationManager(clusterMap, clusterMapConfig, mockHelixParticipant);
    StorageManager storageManager = managers.getFirst();
    MockReplicationManager replicationManager = (MockReplicationManager) managers.getSecond();
    // 1. test replica not found case
    try {
        mockHelixParticipant.onPartitionBecomeOfflineFromInactive("-1");
        fail("should fail because of invalid partition");
    } catch (StateTransitionException e) {
        assertEquals("Error code doesn't match", ReplicaNotFound, e.getErrorCode());
    }
    // 2. test store not started case
    PartitionId existingPartition = replicationManager.partitionToPartitionInfo.keySet().iterator().next();
    storageManager.shutdownBlobStore(existingPartition);
    try {
        mockHelixParticipant.onPartitionBecomeOfflineFromInactive(existingPartition.toPathString());
        fail("should fail because store is not started");
    } catch (StateTransitionException e) {
        assertEquals("Error code doesn't match", StoreNotStarted, e.getErrorCode());
    }
    storageManager.startBlobStore(existingPartition);
    // before testing success case, let's write a blob (size = 100) into local store and add a delete record for new blob
    Store localStore = storageManager.getStore(existingPartition);
    MockId id = new MockId(TestUtils.getRandomString(10), Utils.getRandomShort(TestUtils.RANDOM), Utils.getRandomShort(TestUtils.RANDOM));
    long crc = (new Random()).nextLong();
    long blobSize = 100;
    MessageInfo info = new MessageInfo(id, blobSize, false, false, Utils.Infinite_Time, crc, id.getAccountId(), id.getContainerId(), Utils.Infinite_Time);
    List<MessageInfo> infos = new ArrayList<>();
    List<ByteBuffer> buffers = new ArrayList<>();
    ByteBuffer buffer = ByteBuffer.wrap(TestUtils.getRandomBytes((int) blobSize));
    infos.add(info);
    buffers.add(buffer);
    localStore.put(new MockMessageWriteSet(infos, buffers));
    // delete the blob
    int deleteRecordSize = (int) (new DeleteMessageFormatInputStream(id, (short) 0, (short) 0, 0).getSize());
    MessageInfo deleteInfo = new MessageInfo(id, deleteRecordSize, id.getAccountId(), id.getContainerId(), time.milliseconds());
    localStore.delete(Collections.singletonList(deleteInfo));
    int sizeOfPutAndHeader = 100 + 18;
    int sizeOfWhole = sizeOfPutAndHeader + deleteRecordSize;
    // note that end offset of last PUT = 100 + 18 = 118, end offset of the store is sizeOfWhole
    // 3. test success case (create a new thread and trigger INACTIVE -> OFFLINE transition)
    ReplicaId localReplica = storageManager.getReplica(existingPartition.toPathString());
    // put a decommission-in-progress file into local store dir
    File decommissionFile = new File(localReplica.getReplicaPath(), "decommission_in_progress");
    assertTrue("Couldn't create decommission file in local store", decommissionFile.createNewFile());
    decommissionFile.deleteOnExit();
    assertNotSame("Before disconnection, the local store state shouldn't be OFFLINE", ReplicaState.OFFLINE, localStore.getCurrentState());
    mockHelixParticipant.registerPartitionStateChangeListener(StateModelListenerType.ReplicationManagerListener, replicationManager.replicationListener);
    CountDownLatch participantLatch = new CountDownLatch(1);
    replicationManager.listenerExecutionLatch = new CountDownLatch(1);
    Utils.newThread(() -> {
        mockHelixParticipant.onPartitionBecomeOfflineFromInactive(existingPartition.toPathString());
        participantLatch.countDown();
    }, false).start();
    assertTrue("Partition state change listener in ReplicationManager didn't get called within 1 sec", replicationManager.listenerExecutionLatch.await(1, TimeUnit.SECONDS));
    // the state of local store should be updated to OFFLINE
    assertEquals("Local store state is not expected", ReplicaState.OFFLINE, localStore.getCurrentState());
    // update replication lag between local and peer replicas
    List<RemoteReplicaInfo> remoteReplicaInfos = replicationManager.partitionToPartitionInfo.get(existingPartition).getRemoteReplicaInfos();
    ReplicaId peerReplica1 = remoteReplicaInfos.get(0).getReplicaId();
    ReplicaId peerReplica2 = remoteReplicaInfos.get(1).getReplicaId();
    // peer1 catches up with last PUT, peer2 catches up with end offset of local store. In this case, SyncUp is not complete
    replicationManager.updateTotalBytesReadByRemoteReplica(existingPartition, peerReplica1.getDataNodeId().getHostname(), peerReplica1.getReplicaPath(), sizeOfPutAndHeader);
    replicationManager.updateTotalBytesReadByRemoteReplica(existingPartition, peerReplica2.getDataNodeId().getHostname(), peerReplica2.getReplicaPath(), sizeOfWhole);
    assertFalse("Only one peer replica has fully caught up with end offset so sync-up should not complete", mockHelixParticipant.getReplicaSyncUpManager().isSyncUpComplete(localReplica));
    // make peer1 catch up with end offset
    replicationManager.updateTotalBytesReadByRemoteReplica(existingPartition, peerReplica1.getDataNodeId().getHostname(), peerReplica1.getReplicaPath(), sizeOfWhole);
    // Now, sync-up should complete and transition should be able to proceed.
    assertTrue("Inactive-To-Offline transition didn't complete within 1 sec", participantLatch.await(1, TimeUnit.SECONDS));
    assertFalse("Local store should be stopped after transition", localStore.isStarted());
    storageManager.shutdown();
}

Example 29 with ClusterMapConfig

use of com.github.ambry.config.ClusterMapConfig in project ambry by linkedin.

the class ReplicationTest method replicaFromStandbyToLeaderTest.

/**
 * Test state transition in replication manager from STANDBY to LEADER
 * Test setup: When creating partitions, make sure that there is exactly one replica in LEADER STATE on each data center
 * Test condition: When a partition on current node moves from standby to leader, verify that in-memory map storing
 * partition to peer leader replicas is updated correctly
 * @throws Exception
 */
@Test
public void replicaFromStandbyToLeaderTest() throws Exception {
    MockClusterMap clusterMap = new MockClusterMap();
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(verifiableProperties);
    MockHelixParticipant.metricRegistry = new MetricRegistry();
    MockHelixParticipant mockHelixParticipant = new MockHelixParticipant(clusterMapConfig);
    ReplicationConfig initialReplicationConfig = replicationConfig;
    properties.setProperty("replication.model.across.datacenters", "LEADER_BASED");
    replicationConfig = new ReplicationConfig(new VerifiableProperties(properties));
    Pair<StorageManager, ReplicationManager> managers = createStorageManagerAndReplicationManager(clusterMap, clusterMapConfig, mockHelixParticipant);
    StorageManager storageManager = managers.getFirst();
    MockReplicationManager replicationManager = (MockReplicationManager) managers.getSecond();
    List<ReplicaId> replicaIds = clusterMap.getReplicaIds(replicationManager.dataNodeId);
    for (ReplicaId replicaId : replicaIds) {
        MockReplicaId mockReplicaId = (MockReplicaId) replicaId;
        if (mockReplicaId.getReplicaState() == ReplicaState.LEADER) {
            PartitionId existingPartition = mockReplicaId.getPartitionId();
            mockHelixParticipant.onPartitionBecomeLeaderFromStandby(existingPartition.toPathString());
            Set<ReplicaId> peerLeaderReplicasInReplicationManager = replicationManager.leaderBasedReplicationAdmin.getLeaderPartitionToPeerLeaderReplicas().get(existingPartition.toPathString());
            Set<ReplicaId> peerLeaderReplicasInClusterMap = new HashSet<>(existingPartition.getReplicaIdsByState(ReplicaState.LEADER, null));
            peerLeaderReplicasInClusterMap.remove(mockReplicaId);
            assertThat("Mismatch in list of leader peer replicas stored by partition in replication manager and cluster map", peerLeaderReplicasInReplicationManager, is(peerLeaderReplicasInClusterMap));
        }
    }
    storageManager.shutdown();
    replicationConfig = initialReplicationConfig;
}

Example 30 with ClusterMapConfig

use of com.github.ambry.config.ClusterMapConfig in project ambry by linkedin.

the class RestServerMain method main.

public static void main(String[] args) {
    final RestServer restServer;
    int exitCode = 0;
    ClusterMap clusterMap = null;
    try {
        InvocationOptions options = new InvocationOptions(args);
        Properties properties = Utils.loadProps(options.serverPropsFilePath);
        VerifiableProperties verifiableProperties = new VerifiableProperties(properties);
        ClusterMapConfig clusterMapConfig = new ClusterMapConfig(verifiableProperties);
        ClusterAgentsFactory clusterAgentsFactory = Utils.getObj(clusterMapConfig.clusterMapClusterAgentsFactory, clusterMapConfig, options.hardwareLayoutFilePath, options.partitionLayoutFilePath);
        clusterMap = clusterAgentsFactory.getClusterMap();
        SSLFactory sslFactory = getSSLFactoryIfRequired(verifiableProperties);
        logger.info("Bootstrapping RestServer");
        restServer = new RestServer(verifiableProperties, clusterMap, new LoggingNotificationSystem(), sslFactory);
        // attach shutdown handler to catch control-c
        Runtime.getRuntime().addShutdownHook(new Thread(() -> {
            logger.info("Received shutdown signal. Shutting down RestServer");
            restServer.shutdown();
        }));
        restServer.start();
        restServer.awaitShutdown();
    } catch (Exception e) {
        logger.error("Exception during bootstrap of RestServer", e);
        exitCode = 1;
    } finally {
        if (clusterMap != null) {
            clusterMap.close();
        }
    }
    logger.info("Exiting RestServerMain");
    System.exit(exitCode);
}