Examples with MockClusterMap com.github.ambry.clustermap.MockClusterMap used on opensource projects

Example 26 with MockClusterMap

use of com.github.ambry.clustermap.MockClusterMap in project ambry by linkedin.

the class ReplicationTest method limitMaxPartitionCountPerRequestTest.

/**
 * Test that max partition count per request is honored in {@link ReplicaThread} if there are too many partitions to
 * replicate from the remote node.
 * @throws Exception
 */
@Test
public void limitMaxPartitionCountPerRequestTest() throws Exception {
    MockClusterMap clusterMap = new MockClusterMap();
    Pair<MockHost, MockHost> localAndRemoteHosts = getLocalAndRemoteHosts(clusterMap);
    MockHost localHost = localAndRemoteHosts.getFirst();
    MockHost remoteHost = localAndRemoteHosts.getSecond();
    List<PartitionId> partitionIds = clusterMap.getAllPartitionIds(null);
    for (PartitionId partitionId : partitionIds) {
        // add 5 messages into each partition and place it on remote host only
        addPutMessagesToReplicasOfPartition(partitionId, Collections.singletonList(remoteHost), 5);
    }
    StoreKeyFactory storeKeyFactory = Utils.getObj("com.github.ambry.commons.BlobIdFactory", clusterMap);
    MockStoreKeyConverterFactory mockStoreKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
    mockStoreKeyConverterFactory.setReturnInputIfAbsent(true);
    mockStoreKeyConverterFactory.setConversionMap(new HashMap<>());
    // we set batchSize to 10 in order to get all messages from one partition within single replication cycle
    int batchSize = 10;
    StoreKeyConverter storeKeyConverter = mockStoreKeyConverterFactory.getStoreKeyConverter();
    Transformer transformer = new ValidatingTransformer(storeKeyFactory, storeKeyConverter);
    // we set max partition count per request to 5, which forces thread to replicate replicas in two cycles. (Note that
    // number of partition to replicate is 10, they will be replicated in two batches)
    ReplicationConfig initialReplicationConfig = replicationConfig;
    properties.setProperty("replication.max.partition.count.per.request", String.valueOf(5));
    replicationConfig = new ReplicationConfig(new VerifiableProperties(properties));
    CountDownLatch replicationCompleted = new CountDownLatch(partitionIds.size());
    AtomicReference<Exception> exception = new AtomicReference<>();
    Pair<Map<DataNodeId, List<RemoteReplicaInfo>>, ReplicaThread> replicasAndThread = getRemoteReplicasAndReplicaThread(batchSize, clusterMap, localHost, remoteHost, storeKeyConverter, transformer, (store, messageInfos) -> {
        try {
            replicationCompleted.countDown();
            // for each partition, replication should complete within single cycle (fetch once should suffice), so
            // we shut down local store once blobs are written. This can avoid unnecessary metadata requests sent to
            // remote host.
            store.shutdown();
        } catch (Exception e) {
            exception.set(e);
        }
    }, null);
    ReplicaThread replicaThread = replicasAndThread.getSecond();
    Thread thread = Utils.newThread(replicaThread, false);
    thread.start();
    assertTrue("Replication didn't complete within 10 secs", replicationCompleted.await(10, TimeUnit.SECONDS));
    // verify the # of replicas per metadata request is limited to 5 (note that there are 10 replicas to replicate, they
    // are split into to 2 small batches and get replicated in separate requests)
    assertEquals("There should be 2 metadata requests and each has 5 replicas to replicate", Arrays.asList(5, 5), remoteHost.replicaCountPerRequestTracker);
    // shutdown
    replicaThread.shutdown();
    if (exception.get() != null) {
        throw exception.get();
    }
    replicationConfig = initialReplicationConfig;
}

Example 27 with MockClusterMap

use of com.github.ambry.clustermap.MockClusterMap in project ambry by linkedin.

the class ReplicationTest method replicaThreadSleepTest.

@Test
public void replicaThreadSleepTest() throws Exception {
    MockClusterMap clusterMap = new MockClusterMap();
    Pair<MockHost, MockHost> localAndRemoteHosts = getLocalAndRemoteHosts(clusterMap);
    MockHost localHost = localAndRemoteHosts.getFirst();
    MockHost remoteHost = localAndRemoteHosts.getSecond();
    long expectedThrottleDurationMs = localHost.dataNodeId.getDatacenterName().equals(remoteHost.dataNodeId.getDatacenterName()) ? replicationConfig.replicationIntraReplicaThreadThrottleSleepDurationMs : replicationConfig.replicationInterReplicaThreadThrottleSleepDurationMs;
    MockStoreKeyConverterFactory storeKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
    storeKeyConverterFactory.setConversionMap(new HashMap<>());
    storeKeyConverterFactory.setReturnInputIfAbsent(true);
    MockStoreKeyConverterFactory.MockStoreKeyConverter storeKeyConverter = storeKeyConverterFactory.getStoreKeyConverter();
    StoreKeyFactory storeKeyFactory = new BlobIdFactory(clusterMap);
    Transformer transformer = new BlobIdTransformer(storeKeyFactory, storeKeyConverter);
    int batchSize = 4;
    Pair<Map<DataNodeId, List<RemoteReplicaInfo>>, ReplicaThread> replicasAndThread = getRemoteReplicasAndReplicaThread(batchSize, clusterMap, localHost, remoteHost, storeKeyConverter, transformer, null, null);
    Map<DataNodeId, List<RemoteReplicaInfo>> replicasToReplicate = replicasAndThread.getFirst();
    ReplicaThread replicaThread = replicasAndThread.getSecond();
    // populate data, add 1 messages to both hosts.
    for (PartitionId partitionId : clusterMap.getAllPartitionIds(null)) {
        addPutMessagesToReplicasOfPartition(partitionId, Arrays.asList(localHost, remoteHost), 1);
    }
    // tests to verify replica thread throttling and idling functions in the following steps:
    // 1. all replicas are in sync, thread level sleep and replica quarantine are both enabled.
    // 2. add put messages to some replica and verify that replication for replicas remain disabled.
    // 3. forward the time so replication for replicas are re-enabled and check replication resumes.
    // 4. add more put messages to ensure replication happens continuously when needed and is throttled appropriately.
    // 1. verify that the replica thread sleeps and replicas are temporarily disable when all replicas are synced.
    List<List<RemoteReplicaInfo>> replicasToReplicateList = new ArrayList<>(replicasToReplicate.values());
    // replicate is called and time is moved forward to prepare the replicas for testing.
    replicaThread.replicate();
    time.sleep(replicationConfig.replicationSyncedReplicaBackoffDurationMs + 1);
    long currentTimeMs = time.milliseconds();
    replicaThread.replicate();
    for (List<RemoteReplicaInfo> replicaInfos : replicasToReplicateList) {
        for (RemoteReplicaInfo replicaInfo : replicaInfos) {
            assertEquals("Unexpected re-enable replication time", currentTimeMs + replicationConfig.replicationSyncedReplicaBackoffDurationMs, replicaInfo.getReEnableReplicationTime());
        }
    }
    currentTimeMs = time.milliseconds();
    replicaThread.replicate();
    assertEquals("Replicas are in sync, replica thread should sleep by replication.thread.idle.sleep.duration.ms", currentTimeMs + replicationConfig.replicationReplicaThreadIdleSleepDurationMs, time.milliseconds());
    // 2. add 3 messages to a partition in the remote host only and verify replication for all replicas should be disabled.
    PartitionId partitionId = clusterMap.getWritablePartitionIds(null).get(0);
    addPutMessagesToReplicasOfPartition(partitionId, Collections.singletonList(remoteHost), 3);
    int[] missingKeys = new int[replicasToReplicate.get(remoteHost.dataNodeId).size()];
    for (int i = 0; i < missingKeys.length; i++) {
        missingKeys[i] = replicasToReplicate.get(remoteHost.dataNodeId).get(i).getReplicaId().getPartitionId().isEqual(partitionId.toPathString()) ? 3 : 0;
    }
    currentTimeMs = time.milliseconds();
    replicaThread.replicate();
    assertEquals("Replication for all replicas should be disabled and the thread should sleep", currentTimeMs + replicationConfig.replicationReplicaThreadIdleSleepDurationMs, time.milliseconds());
    assertMissingKeys(missingKeys, batchSize, replicaThread, remoteHost, replicasToReplicate);
    // 3. forward the time and run replicate and verify the replication.
    time.sleep(replicationConfig.replicationSyncedReplicaBackoffDurationMs);
    replicaThread.replicate();
    missingKeys = new int[replicasToReplicate.get(remoteHost.dataNodeId).size()];
    assertMissingKeys(missingKeys, batchSize, replicaThread, remoteHost, replicasToReplicate);
    // Since, now we moved setting of remoteReplicaInfo::setReEnableReplicationTime inside replicaThread::exchangeMetaData and
    // above assertMissingKeys() does exchangeMetadata() for replicas up to date, each replica will have
    // ReEnableReplicationTime set by replicationSyncedReplicaBackoffDurationMs. Forward the time here.
    time.sleep(replicationConfig.replicationSyncedReplicaBackoffDurationMs);
    // 4. add more put messages and verify that replication continues and is throttled appropriately.
    addPutMessagesToReplicasOfPartition(partitionId, Arrays.asList(localHost, remoteHost), 3);
    currentTimeMs = time.milliseconds();
    replicaThread.replicate();
    assertEquals("Replica thread should sleep exactly " + expectedThrottleDurationMs + " since remote has new token", currentTimeMs + expectedThrottleDurationMs, time.milliseconds());
    assertMissingKeys(missingKeys, batchSize, replicaThread, remoteHost, replicasToReplicate);
    // Since, now we moved setting of remoteReplicaInfo::setReEnableReplicationTime inside replicaThread::exchangeMetaData and
    // above assertMissingKeys() does exchangeMetadata() for replicas up to date, each replica will have
    // ReEnableReplicationTime set by replicationSyncedReplicaBackoffDurationMs. Forward the time here.
    time.sleep(replicationConfig.replicationSyncedReplicaBackoffDurationMs);
    // verify that throttling on the replica thread is disabled when relevant configs are 0.
    Properties properties = new Properties();
    properties.setProperty("replication.intra.replica.thread.throttle.sleep.duration.ms", "0");
    properties.setProperty("replication.inter.replica.thread.throttle.sleep.duration.ms", "0");
    replicationConfig = new ReplicationConfig(new VerifiableProperties(properties));
    replicasAndThread = getRemoteReplicasAndReplicaThread(batchSize, clusterMap, localHost, remoteHost, storeKeyConverter, transformer, null, null);
    replicaThread = replicasAndThread.getSecond();
    currentTimeMs = time.milliseconds();
    replicaThread.replicate();
    assertEquals("Replica thread should not sleep when throttling is disabled and replicas are out of sync", currentTimeMs, time.milliseconds());
}

Example 28 with MockClusterMap

use of com.github.ambry.clustermap.MockClusterMap 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 29 with MockClusterMap

use of com.github.ambry.clustermap.MockClusterMap 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 30 with MockClusterMap

use of com.github.ambry.clustermap.MockClusterMap 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;
}