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

Example 86 with ClusterMapConfig

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

the class LeaderBasedReplicationTest method replicaThreadLeaderBasedReplicationStandByCrossColoFetchTest.

/**
 * Test leader based replication to verify cross colo gets for standby replicas after they have have timed out
 * waiting for missing keys.
 * @throws Exception
 */
@Test
public void replicaThreadLeaderBasedReplicationStandByCrossColoFetchTest() throws Exception {
    Map<DataNodeId, MockHost> hosts = new HashMap<>();
    hosts.put(remoteNodeInLocalDC, remoteHostInLocalDC);
    hosts.put(remoteNodeInRemoteDC, remoteHostInRemoteDC);
    int batchSize = 5;
    int numOfMessagesOnRemoteNodeInLocalDC = 3;
    int numOfMessagesOnRemoteNodeInRemoteDC = 10;
    ConnectionPool mockConnectionPool = new MockConnectionPool(hosts, clusterMap, batchSize);
    Pair<StorageManager, ReplicationManager> managers = createStorageManagerAndReplicationManager(clusterMap, clusterMapConfig, mockHelixParticipant, mockConnectionPool);
    StorageManager storageManager = managers.getFirst();
    MockReplicationManager replicationManager = (MockReplicationManager) managers.getSecond();
    // set mock local stores on all remoteReplicaInfos which will used during replication.
    for (PartitionId partitionId : replicationManager.partitionToPartitionInfo.keySet()) {
        localHost.addStore(partitionId, null);
        Store localStore = localHost.getStore(partitionId);
        localStore.start();
        List<RemoteReplicaInfo> remoteReplicaInfos = replicationManager.partitionToPartitionInfo.get(partitionId).getRemoteReplicaInfos();
        remoteReplicaInfos.forEach(remoteReplicaInfo -> remoteReplicaInfo.setLocalStore(localStore));
    }
    // get remote replicas and replica thread for remote host on local datacenter
    ReplicaThread intraColoReplicaThread = replicationManager.dataNodeIdToReplicaThread.get(remoteNodeInLocalDC);
    List<RemoteReplicaInfo> remoteReplicaInfosForLocalDC = intraColoReplicaThread.getRemoteReplicaInfos().get(remoteNodeInLocalDC);
    // get remote replicas and replica thread for remote host on remote datacenter
    ReplicaThread crossColoReplicaThread = replicationManager.dataNodeIdToReplicaThread.get(remoteNodeInRemoteDC);
    List<RemoteReplicaInfo> remoteReplicaInfosForRemoteDC = crossColoReplicaThread.getRemoteReplicaInfos().get(remoteNodeInRemoteDC);
    // mock helix transition state from standby to leader for local leader partitions
    List<? extends ReplicaId> replicaIds = clusterMap.getReplicaIds(replicationManager.dataNodeId);
    for (ReplicaId replicaId : replicaIds) {
        MockReplicaId mockReplicaId = (MockReplicaId) replicaId;
        if (mockReplicaId.getReplicaState() == ReplicaState.LEADER) {
            MockPartitionId mockPartitionId = (MockPartitionId) replicaId.getPartitionId();
            mockHelixParticipant.onPartitionBecomeLeaderFromStandby(mockPartitionId.toPathString());
        }
    }
    // Add put messages to all partitions on remoteHost1 and remoteHost2
    List<PartitionId> partitionIds = clusterMap.getWritablePartitionIds(null);
    for (PartitionId partitionId : partitionIds) {
        // add 3 put messages to the remoteNodeInLocalDC and remoteNodeInRemoteDC from which local host will replicate.
        addPutMessagesToReplicasOfPartition(partitionId, Arrays.asList(remoteHostInLocalDC, remoteHostInRemoteDC), numOfMessagesOnRemoteNodeInLocalDC);
        // add 1 put message to the remoteNodeInRemoteDC only. Since this message is not present in remoteNodeInLocalDC, it
        // doesn't come to local node via intra-dc replication. We should see time out for remote standby replicas waiting for this
        // message and see a cross colo fetch happening.
        addPutMessagesToReplicasOfPartition(partitionId, Collections.singletonList(remoteHostInRemoteDC), numOfMessagesOnRemoteNodeInRemoteDC - numOfMessagesOnRemoteNodeInLocalDC);
    }
    // Choose partitions that are leaders on both local and remote nodes
    Set<ReplicaId> leaderReplicasOnLocalAndRemoteNodes = getRemoteLeaderReplicasWithLeaderPartitionsOnLocalNode(clusterMap, replicationManager.dataNodeId, remoteNodeInRemoteDC);
    // replicate with remote node in remote DC
    crossColoReplicaThread.replicate();
    // missing messages are not fetched yet.
    for (RemoteReplicaInfo remoteReplicaInfo : remoteReplicaInfosForRemoteDC) {
        if (leaderReplicasOnLocalAndRemoteNodes.contains(remoteReplicaInfo.getReplicaId())) {
            assertEquals("remote token mismatch for leader replicas", ((MockFindToken) remoteReplicaInfo.getToken()).getIndex(), batchSize - 1);
        } else {
            assertEquals("remote token should not move forward for standby replicas until missing keys are fetched", ((MockFindToken) remoteReplicaInfo.getToken()).getIndex(), 0);
        }
    }
    // Replicate with remote node in local dc
    intraColoReplicaThread.replicate();
    // verify that remote token will be moved for all replicas as it is intra-dc replication
    for (RemoteReplicaInfo remoteReplicaInfo : remoteReplicaInfosForLocalDC) {
        assertEquals("mismatch in remote token set for intra colo replicas", ((MockFindToken) remoteReplicaInfo.getToken()).getIndex(), numOfMessagesOnRemoteNodeInLocalDC - 1);
    }
    // via intra-dc replication
    for (RemoteReplicaInfo remoteReplicaInfo : remoteReplicaInfosForRemoteDC) {
        crossColoReplicaThread.processMissingKeysFromPreviousMetadataResponse(remoteReplicaInfo);
    }
    // verify that the remote token will remain 0 for standby replicas as one message in its missing set is not fetched yet.
    for (RemoteReplicaInfo remoteReplicaInfo : remoteReplicaInfosForRemoteDC) {
        if (!leaderReplicasOnLocalAndRemoteNodes.contains(remoteReplicaInfo.getReplicaId())) {
            assertTrue("missing store messages should still exist for standby replicas", crossColoReplicaThread.containsMissingKeysFromPreviousMetadataExchange(remoteReplicaInfo));
            assertEquals("remote token should not move forward for standby replicas until missing keys are fetched", ((MockFindToken) remoteReplicaInfo.getToken()).getIndex(), 0);
            assertEquals("incorrect number of missing store messages found for standby replicas", remoteReplicaInfo.getExchangeMetadataResponse().missingStoreMessages.size(), batchSize - numOfMessagesOnRemoteNodeInLocalDC);
        }
    }
    // Attempt replication with remoteNodeInRemoteDC, we should not see any replication attempt for standby replicas
    // and their remote token stays as 0.
    crossColoReplicaThread.replicate();
    for (RemoteReplicaInfo remoteReplicaInfo : remoteReplicaInfosForRemoteDC) {
        if (!leaderReplicasOnLocalAndRemoteNodes.contains(remoteReplicaInfo.getReplicaId())) {
            assertEquals("remote token should not move forward for standby replicas until missing keys are fetched", ((MockFindToken) remoteReplicaInfo.getToken()).getIndex(), 0);
            assertTrue("missing store messages should still exist for standby replicas", crossColoReplicaThread.containsMissingKeysFromPreviousMetadataExchange(remoteReplicaInfo));
        }
    }
    // Move time forward by replicationStandbyWaitTimeoutToTriggerCrossColoFetchSeconds+1 seconds and attempt replication.
    // We should see cross colo fetch for standby replicas now since missing keys haven't arrived for
    // replicationConfig.replicationStandbyWaitTimeoutToTriggerCrossColoFetchSeconds.
    time.sleep((replicationConfig.replicationStandbyWaitTimeoutToTriggerCrossColoFetchSeconds + 1) * 1000);
    // verify that we get the list of standby replicas that timed out on no progress
    Set<RemoteReplicaInfo> allStandbyReplicas = remoteReplicaInfosForRemoteDC.stream().filter(info -> !leaderReplicasOnLocalAndRemoteNodes.contains(info.getReplicaId())).collect(Collectors.toSet());
    assertEquals("mismatch in list of standby replicas timed out on no progress", new HashSet<>(crossColoReplicaThread.getRemoteStandbyReplicasTimedOutOnNoProgress(remoteReplicaInfosForRemoteDC)), allStandbyReplicas);
    crossColoReplicaThread.replicate();
    // token index for all standby replicas will move forward after fetching missing keys themselves
    for (RemoteReplicaInfo remoteReplicaInfo : remoteReplicaInfosForRemoteDC) {
        if (!leaderReplicasOnLocalAndRemoteNodes.contains(remoteReplicaInfo.getReplicaId())) {
            assertEquals("mismatch in remote token set for standby cross colo replicas", ((MockFindToken) remoteReplicaInfo.getToken()).getIndex(), batchSize - 1);
            assertFalse("missing store messages should be empty for standby replicas now", crossColoReplicaThread.containsMissingKeysFromPreviousMetadataExchange(remoteReplicaInfo));
        }
    }
    // verify replication metrics to track number of cross colo get requests for standby replicas. If all replicas are
    // leaders, we should have 0 cross colo get requests.
    String remoteDataCenter = remoteReplicaInfosForRemoteDC.get(0).getReplicaId().getDataNodeId().getDatacenterName();
    assertEquals("mismatch in number of cross colo get requests tracked for standby replicas", crossColoReplicaThread.getReplicationMetrics().interColoReplicationGetRequestCountForStandbyReplicas.get(remoteDataCenter).getCount(), leaderReplicasOnLocalAndRemoteNodes.size() != remoteReplicaInfosForRemoteDC.size() ? 1 : 0);
    storageManager.shutdown();
}

Example 87 with ClusterMapConfig

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

the class ServerTestUtil method endToEndReplicationWithMultiNodeMultiPartitionMultiDCTest.

static void endToEndReplicationWithMultiNodeMultiPartitionMultiDCTest(String sourceDatacenter, String sslEnabledDatacenters, PortType portType, MockCluster cluster, MockNotificationSystem notificationSystem, Properties routerProps) throws Exception {
    Properties props = new Properties();
    props.setProperty("router.hostname", "localhost");
    props.setProperty("router.datacenter.name", sourceDatacenter);
    props.setProperty("router.put.request.parallelism", "1");
    props.setProperty("router.put.success.target", "1");
    props.setProperty("clustermap.cluster.name", "test");
    props.setProperty("clustermap.datacenter.name", sourceDatacenter);
    props.setProperty("clustermap.host.name", "localhost");
    props.setProperty("kms.default.container.key", TestUtils.getRandomKey(32));
    props.putAll(routerProps);
    VerifiableProperties verifiableProperties = new VerifiableProperties(props);
    AccountService accountService = new InMemAccountService(false, true);
    Router router = new NonBlockingRouterFactory(verifiableProperties, cluster.getClusterMap(), notificationSystem, getSSLFactoryIfRequired(verifiableProperties), accountService).getRouter();
    int numberOfRequestsToSend = 15;
    int numberOfVerifierThreads = 3;
    final LinkedBlockingQueue<Payload> payloadQueue = new LinkedBlockingQueue<Payload>();
    final AtomicReference<Exception> exceptionRef = new AtomicReference<>(null);
    final CountDownLatch callbackLatch = new CountDownLatch(numberOfRequestsToSend);
    List<Future<String>> putFutures = new ArrayList<>(numberOfRequestsToSend);
    short accountId = Utils.getRandomShort(TestUtils.RANDOM);
    short containerId = Utils.getRandomShort(TestUtils.RANDOM);
    for (int i = 0; i < numberOfRequestsToSend; i++) {
        int size = new Random().nextInt(5000);
        final BlobProperties properties = new BlobProperties(size, "service1", "owner id check", "image/jpeg", false, TestUtils.TTL_SECS, cluster.time.milliseconds(), accountId, containerId, false, null, null, null);
        final byte[] metadata = new byte[new Random().nextInt(1000)];
        final byte[] blob = new byte[size];
        TestUtils.RANDOM.nextBytes(metadata);
        TestUtils.RANDOM.nextBytes(blob);
        Future<String> future = router.putBlob(properties, metadata, new ByteBufferReadableStreamChannel(ByteBuffer.wrap(blob)), new PutBlobOptionsBuilder().build(), new Callback<String>() {

            @Override
            public void onCompletion(String result, Exception exception) {
                if (exception == null) {
                    payloadQueue.add(new Payload(properties, metadata, blob, result));
                } else {
                    exceptionRef.set(exception);
                }
                callbackLatch.countDown();
            }
        }, QUOTA_CHARGE_EVENT_LISTENER);
        putFutures.add(future);
    }
    for (Future<String> future : putFutures) {
        future.get(20, TimeUnit.SECONDS);
    }
    assertTrue("Did not receive all callbacks in time", callbackLatch.await(1, TimeUnit.SECONDS));
    if (exceptionRef.get() != null) {
        throw exceptionRef.get();
    }
    // put away for future use
    Payload payload1 = payloadQueue.peek();
    MockClusterMap clusterMap = cluster.getClusterMap();
    BlobId blobId1 = new BlobId(payload1.blobId, clusterMap);
    assertEquals("Did not put expected number of blobs", numberOfRequestsToSend, payloadQueue.size());
    Properties sslProps = new Properties();
    sslProps.putAll(routerProps);
    sslProps.setProperty("clustermap.ssl.enabled.datacenters", sslEnabledDatacenters);
    sslProps.setProperty("clustermap.cluster.name", "test");
    sslProps.setProperty("clustermap.datacenter.name", sourceDatacenter);
    sslProps.setProperty("clustermap.host.name", "localhost");
    sslProps.setProperty("connectionpool.read.timeout.ms", "15000");
    VerifiableProperties vProps = new VerifiableProperties(sslProps);
    ConnectionPool connectionPool = new BlockingChannelConnectionPool(new ConnectionPoolConfig(vProps), new SSLConfig(vProps), new ClusterMapConfig(vProps), new MetricRegistry());
    CountDownLatch verifierLatch = new CountDownLatch(numberOfVerifierThreads);
    AtomicInteger totalRequests = new AtomicInteger(numberOfRequestsToSend);
    AtomicInteger verifiedRequests = new AtomicInteger(0);
    AtomicBoolean cancelTest = new AtomicBoolean(false);
    for (int i = 0; i < numberOfVerifierThreads; i++) {
        Thread thread = new Thread(new Verifier(payloadQueue, verifierLatch, totalRequests, verifiedRequests, cluster.getClusterMap(), cancelTest, portType, connectionPool, notificationSystem, cluster.time));
        thread.start();
    }
    assertTrue("Did not verify in 2 minutes", verifierLatch.await(2, TimeUnit.MINUTES));
    assertEquals(totalRequests.get(), verifiedRequests.get());
    BlobIdFactory blobIdFactory = new BlobIdFactory(clusterMap);
    MockDataNodeId dataNodeId = clusterMap.getDataNodes().get(0);
    Port port = new Port(portType == PortType.PLAINTEXT ? dataNodeId.getPort() : dataNodeId.getSSLPort(), portType);
    ConnectedChannel channel = connectionPool.checkOutConnection("localhost", port, 10000);
    PartitionId partitionId = blobId1.getPartition();
    // stop the store via AdminRequest
    System.out.println("Begin to stop a BlobStore");
    AdminRequest adminRequest = new AdminRequest(AdminRequestOrResponseType.BlobStoreControl, partitionId, 1, "clientid2");
    BlobStoreControlAdminRequest controlRequest = new BlobStoreControlAdminRequest((short) 0, BlobStoreControlAction.StopStore, adminRequest);
    DataInputStream stream = channel.sendAndReceive(controlRequest).getInputStream();
    AdminResponse adminResponse = AdminResponse.readFrom(stream);
    releaseNettyBufUnderneathStream(stream);
    assertEquals("Stop store admin request should succeed", ServerErrorCode.No_Error, adminResponse.getError());
    // put a blob on a stopped store, which should fail
    byte[] userMetadata = new byte[1000];
    byte[] data = new byte[3187];
    BlobProperties properties = new BlobProperties(3187, "serviceid1", accountId, containerId, false, cluster.time.milliseconds());
    BlobId blobId2 = new BlobId(CommonTestUtils.getCurrentBlobIdVersion(), BlobId.BlobIdType.NATIVE, clusterMap.getLocalDatacenterId(), accountId, containerId, partitionId, false, BlobId.BlobDataType.DATACHUNK);
    PutRequest putRequest2 = new PutRequest(1, "clientId2", blobId2, properties, ByteBuffer.wrap(userMetadata), Unpooled.wrappedBuffer(data), properties.getBlobSize(), BlobType.DataBlob, null);
    DataInputStream putResponseStream = channel.sendAndReceive(putRequest2).getInputStream();
    PutResponse response2 = PutResponse.readFrom(putResponseStream);
    releaseNettyBufUnderneathStream(putResponseStream);
    assertEquals("Put blob on stopped store should fail", ServerErrorCode.Replica_Unavailable, response2.getError());
    // get a blob properties on a stopped store, which should fail
    ArrayList<BlobId> ids = new ArrayList<>();
    ids.add(blobId1);
    ArrayList<PartitionRequestInfo> partitionRequestInfoList = new ArrayList<>();
    PartitionRequestInfo partitionRequestInfo = new PartitionRequestInfo(partitionId, ids);
    partitionRequestInfoList.add(partitionRequestInfo);
    GetRequest getRequest1 = new GetRequest(1, "clientId1", MessageFormatFlags.BlobProperties, partitionRequestInfoList, GetOption.None);
    stream = channel.sendAndReceive(getRequest1).getInputStream();
    GetResponse resp1 = GetResponse.readFrom(stream, clusterMap);
    assertEquals("Get blob properties on stopped store should fail", ServerErrorCode.Replica_Unavailable, resp1.getPartitionResponseInfoList().get(0).getErrorCode());
    releaseNettyBufUnderneathStream(stream);
    // delete a blob on a stopped store, which should fail
    DeleteRequest deleteRequest = new DeleteRequest(1, "clientId1", blobId1, System.currentTimeMillis());
    stream = channel.sendAndReceive(deleteRequest).getInputStream();
    DeleteResponse deleteResponse = DeleteResponse.readFrom(stream);
    releaseNettyBufUnderneathStream(stream);
    assertEquals("Delete blob on stopped store should fail", ServerErrorCode.Replica_Unavailable, deleteResponse.getError());
    // start the store via AdminRequest
    System.out.println("Begin to restart the BlobStore");
    adminRequest = new AdminRequest(AdminRequestOrResponseType.BlobStoreControl, partitionId, 1, "clientId");
    controlRequest = new BlobStoreControlAdminRequest((short) 0, BlobStoreControlAction.StartStore, adminRequest);
    stream = channel.sendAndReceive(controlRequest).getInputStream();
    adminResponse = AdminResponse.readFrom(stream);
    releaseNettyBufUnderneathStream(stream);
    assertEquals("Start store admin request should succeed", ServerErrorCode.No_Error, adminResponse.getError());
    List<? extends ReplicaId> replicaIds = partitionId.getReplicaIds();
    for (ReplicaId replicaId : replicaIds) {
        // forcibly mark replicas and disks as up.
        MockReplicaId mockReplicaId = (MockReplicaId) replicaId;
        mockReplicaId.markReplicaDownStatus(false);
        ((MockDiskId) mockReplicaId.getDiskId()).setDiskState(HardwareState.AVAILABLE, false);
    }
    // put a blob on a restarted store , which should succeed
    putRequest2 = new PutRequest(1, "clientId2", blobId2, properties, ByteBuffer.wrap(userMetadata), Unpooled.wrappedBuffer(data), properties.getBlobSize(), BlobType.DataBlob, null);
    putResponseStream = channel.sendAndReceive(putRequest2).getInputStream();
    response2 = PutResponse.readFrom(putResponseStream);
    releaseNettyBufUnderneathStream(putResponseStream);
    assertEquals("Put blob on restarted store should succeed", ServerErrorCode.No_Error, response2.getError());
    // verify the put blob has been replicated successfully.
    notificationSystem.awaitBlobCreations(blobId2.getID());
    // get a blob on a restarted store , which should succeed
    ids = new ArrayList<BlobId>();
    ids.add(blobId2);
    partitionRequestInfoList = new ArrayList<PartitionRequestInfo>();
    partitionRequestInfo = new PartitionRequestInfo(partitionId, ids);
    partitionRequestInfoList.add(partitionRequestInfo);
    GetRequest getRequest2 = new GetRequest(1, "clientId2", MessageFormatFlags.All, partitionRequestInfoList, GetOption.None);
    stream = channel.sendAndReceive(getRequest2).getInputStream();
    GetResponse resp2 = GetResponse.readFrom(stream, clusterMap);
    InputStream responseStream = resp2.getInputStream();
    BlobAll blobAll = MessageFormatRecord.deserializeBlobAll(responseStream, blobIdFactory);
    byte[] actualBlobData = getBlobDataAndRelease(blobAll.getBlobData());
    assertArrayEquals("Content mismatch.", data, actualBlobData);
    releaseNettyBufUnderneathStream(stream);
    // delete a blob on a restarted store , which should succeed
    deleteRequest = new DeleteRequest(1, "clientId2", blobId2, System.currentTimeMillis());
    stream = channel.sendAndReceive(deleteRequest).getInputStream();
    deleteResponse = DeleteResponse.readFrom(stream);
    releaseNettyBufUnderneathStream(stream);
    assertEquals("Delete blob on restarted store should succeed", ServerErrorCode.No_Error, deleteResponse.getError());
    router.close();
    connectionPool.shutdown();
}

Example 88 with ClusterMapConfig

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

the class StatsManagerTest method testReplicaFromOfflineToDropped.

/**
 * Test Offline-To-Dropped transition (both failure and success cases)
 * @throws Exception
 */
@Test
public void testReplicaFromOfflineToDropped() throws Exception {
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(verifiableProperties);
    ReplicationConfig replicationConfig = new ReplicationConfig(verifiableProperties);
    StoreConfig storeConfig = new StoreConfig(verifiableProperties);
    MockClusterMap clusterMap = new MockClusterMap();
    DataNodeId currentNode = clusterMap.getDataNodeIds().get(0);
    List<ReplicaId> localReplicas = clusterMap.getReplicaIds(currentNode);
    StorageManager storageManager = new StorageManager(storeConfig, new DiskManagerConfig(verifiableProperties), Utils.newScheduler(1, true), new MetricRegistry(), null, clusterMap, currentNode, null, Collections.singletonList(clusterParticipant), new MockTime(), null, new InMemAccountService(false, false));
    storageManager.start();
    MockStoreKeyConverterFactory storeKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
    storeKeyConverterFactory.setConversionMap(new HashMap<>());
    MockReplicationManager mockReplicationManager = new MockReplicationManager(replicationConfig, clusterMapConfig, storeConfig, storageManager, clusterMap, currentNode, storeKeyConverterFactory, clusterParticipant);
    MockStatsManager mockStatsManager = new MockStatsManager(storageManager, localReplicas, new MetricRegistry(), statsManagerConfig, clusterParticipant);
    // 1. attempt to remove replica while store is still running (remove store failure case)
    ReplicaId replicaToDrop = localReplicas.get(0);
    try {
        clusterParticipant.onPartitionBecomeDroppedFromOffline(replicaToDrop.getPartitionId().toPathString());
        fail("should fail because store is still running");
    } catch (StateTransitionException e) {
        assertEquals("Error code doesn't match", ReplicaOperationFailure, e.getErrorCode());
    }
    // 2. shutdown the store but introduce file deletion failure (put a invalid dir in store dir)
    storageManager.shutdownBlobStore(replicaToDrop.getPartitionId());
    File invalidDir = new File(replicaToDrop.getReplicaPath(), "invalidDir");
    invalidDir.deleteOnExit();
    assertTrue("Couldn't create dir within store dir", invalidDir.mkdir());
    assertTrue("Could not make unreadable", invalidDir.setReadable(false));
    try {
        clusterParticipant.onPartitionBecomeDroppedFromOffline(replicaToDrop.getPartitionId().toPathString());
        fail("should fail because store deletion fails");
    } catch (StateTransitionException e) {
        assertEquals("Error code doesn't match", ReplicaOperationFailure, e.getErrorCode());
    }
    // reset permission to allow deletion to succeed.
    assertTrue("Could not make readable", invalidDir.setReadable(true));
    assertTrue("Could not delete invalid dir", invalidDir.delete());
    // 3. success case (remove another replica because previous replica has been removed from in-mem data structures)
    ReplicaId replica = localReplicas.get(1);
    storageManager.shutdownBlobStore(replica.getPartitionId());
    MockHelixParticipant mockHelixParticipant = Mockito.spy(clusterParticipant);
    doNothing().when(mockHelixParticipant).setPartitionDisabledState(anyString(), anyBoolean());
    mockHelixParticipant.onPartitionBecomeDroppedFromOffline(replica.getPartitionId().toPathString());
    // verify that the replica is no longer present in StorageManager
    assertNull("Store of removed replica should not exist", storageManager.getStore(replica.getPartitionId(), true));
    // purposely remove the same replica in ReplicationManager again to verify it no longer exists
    assertFalse("Should return false because replica no longer exists", mockReplicationManager.removeReplica(replica));
    // purposely remove the same replica in StatsManager again to verify it no longer exists
    assertFalse("Should return false because replica no longer exists", mockStatsManager.removeReplica(replica));
    verify(mockHelixParticipant).setPartitionDisabledState(replica.getPartitionId().toPathString(), false);
    storageManager.shutdown();
    mockStatsManager.shutdown();
}

Example 89 with ClusterMapConfig

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

the class HelixBootstrapUpgradeToolTest method testDisablePartitionAdminOp.

/**
 * Test when AdminOperation is specified to DisablePartition, Helix bootstrap tool is able to disable certain partition
 * only without changing IdealState and InstanceConfig. (In practice, this is first step to decommission a replica)
 * @throws Exception
 */
@Test
public void testDisablePartitionAdminOp() throws Exception {
    String clusterName = CLUSTER_NAME_PREFIX + CLUSTER_NAME_IN_STATIC_CLUSTER_MAP;
    // Test regular bootstrap. This is to ensure DataNodeConfig and IdealState are there before testing disabling
    // certain replica on specific node.
    long expectedResourceCount = (testPartitionLayout.getPartitionLayout().getPartitionCount() - 1) / DEFAULT_MAX_PARTITIONS_PER_RESOURCE + 1;
    writeBootstrapOrUpgrade(expectedResourceCount, false);
    int totalPartitionCount = testPartitionLayout.getPartitionCount();
    // Randomly pick a partition to remove one of its replicas
    Partition testPartition = (Partition) testPartitionLayout.getPartitionLayout().getPartitions(null).get(RANDOM.nextInt(totalPartitionCount));
    ReplicaId removedReplica = testPartition.getReplicaIds().stream().filter(r -> r.getDataNodeId().getDatacenterName().equals("DC1")).findFirst().get();
    testPartition.getReplicas().remove(removedReplica);
    ZkInfo zkInfo = dcsToZkInfo.get(removedReplica.getDataNodeId().getDatacenterName());
    // create a participant that hosts this removed replica
    Properties props = new Properties();
    props.setProperty("clustermap.host.name", "localhost");
    props.setProperty("clustermap.port", String.valueOf(removedReplica.getDataNodeId().getPort()));
    props.setProperty("clustermap.cluster.name", clusterName);
    props.setProperty("clustermap.datacenter.name", "DC1");
    props.setProperty("clustermap.update.datanode.info", Boolean.toString(true));
    props.setProperty("clustermap.dcs.zk.connect.strings", zkJson.toString(2));
    props.setProperty("clustermap.retry.disable.partition.completion.backoff.ms", Integer.toString(100));
    props.setProperty("clustermap.data.node.config.source.type", dataNodeConfigSourceType.name());
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(new VerifiableProperties(props));
    HelixParticipant helixParticipant = new HelixParticipant(clusterMapConfig, new HelixFactory(), new MetricRegistry(), "localhost:" + zkInfo.getPort(), true);
    PropertyStoreToDataNodeConfigAdapter propertyStoreAdapter = dataNodeConfigSourceType == DataNodeConfigSourceType.INSTANCE_CONFIG ? null : new PropertyStoreToDataNodeConfigAdapter("localhost:" + zkInfo.getPort(), clusterMapConfig);
    InstanceConfigToDataNodeConfigAdapter.Converter instanceConfigConverter = new InstanceConfigToDataNodeConfigAdapter.Converter(clusterMapConfig);
    // create HelixAdmin
    ZKHelixAdmin admin = new ZKHelixAdmin("localhost:" + zkInfo.getPort());
    // Write changes to static files
    Utils.writeJsonObjectToFile(zkJson, zkLayoutPath);
    Utils.writeJsonObjectToFile(testHardwareLayout.getHardwareLayout().toJSONObject(), hardwareLayoutPath);
    Utils.writeJsonObjectToFile(testPartitionLayout.getPartitionLayout().toJSONObject(), partitionLayoutPath);
    // make bootstrap tool blocked by count down latch before removing znodes for disabling partitions
    blockRemovingNodeLatch = new CountDownLatch(1);
    disablePartitionLatch = new CountDownLatch(activeDcSet.size());
    CountDownLatch bootstrapCompletionLatch = new CountDownLatch(1);
    Utils.newThread(() -> {
        try {
            // Upgrade Helix by updating IdealState: AdminOperation = DisablePartition
            HelixBootstrapUpgradeUtil.bootstrapOrUpgrade(hardwareLayoutPath, partitionLayoutPath, zkLayoutPath, CLUSTER_NAME_PREFIX, dcStr, DEFAULT_MAX_PARTITIONS_PER_RESOURCE, false, false, new HelixAdminFactory(), false, ClusterMapConfig.DEFAULT_STATE_MODEL_DEF, DisablePartition, dataNodeConfigSourceType, false);
            bootstrapCompletionLatch.countDown();
        } catch (Exception e) {
        // do nothing, if there is any exception subsequent test should fail.
        }
    }, false).start();
    assertTrue("Disable partition latch didn't come down within 5 seconds", disablePartitionLatch.await(5, TimeUnit.SECONDS));
    // Let's attempt to update InstanceConfig/DataNodeConfig via HelixParticipant, which should be blocked
    CountDownLatch updateCompletionLatch = new CountDownLatch(1);
    Utils.newThread(() -> {
        helixParticipant.updateDataNodeInfoInCluster(removedReplica, false);
        updateCompletionLatch.countDown();
    }, false).start();
    // sleep 100 ms to ensure updateDataNodeInfoInCluster is blocked due to disabling partition hasn't completed yet
    Thread.sleep(100);
    // Ensure the DataNodeConfig still has the replica
    String instanceName = getInstanceName(removedReplica.getDataNodeId());
    DataNodeConfig currentDataNodeConfig = dataNodeConfigSourceType == DataNodeConfigSourceType.INSTANCE_CONFIG ? instanceConfigConverter.convert(admin.getInstanceConfig(clusterName, instanceName)) : propertyStoreAdapter.get(instanceName);
    verifyReplicaInfoInDataNodeConfig(currentDataNodeConfig, removedReplica, true);
    // verify the znode is created for the node on which partition has been disabled.
    Properties properties = new Properties();
    properties.setProperty("helix.property.store.root.path", "/" + clusterName + "/" + PROPERTYSTORE_STR);
    HelixPropertyStoreConfig propertyStoreConfig = new HelixPropertyStoreConfig(new VerifiableProperties(properties));
    HelixPropertyStore<ZNRecord> helixPropertyStore = CommonUtils.createHelixPropertyStore("localhost:" + zkInfo.getPort(), propertyStoreConfig, null);
    String path = PARTITION_DISABLED_ZNODE_PATH + getInstanceName(removedReplica.getDataNodeId());
    assertTrue("ZNode is not found for disabled partition node.", helixPropertyStore.exists(path, AccessOption.PERSISTENT));
    helixPropertyStore.stop();
    // unblock HelixBootstrapTool
    blockRemovingNodeLatch.countDown();
    // waiting for bootstrap tool to complete
    assertTrue("Helix tool didn't complete within 5 seconds", bootstrapCompletionLatch.await(5, TimeUnit.SECONDS));
    verifyResourceCount(testHardwareLayout.getHardwareLayout(), expectedResourceCount);
    assertTrue("Helix participant didn't complete update within 5 seconds", updateCompletionLatch.await(5, TimeUnit.SECONDS));
    InstanceConfig currentInstanceConfig = admin.getInstanceConfig(clusterName, getInstanceName(removedReplica.getDataNodeId()));
    // Verify that replica has been disabled
    String resourceName = null;
    for (String rs : admin.getResourcesInCluster(clusterName)) {
        IdealState is = admin.getResourceIdealState(clusterName, rs);
        if (is.getPartitionSet().contains(removedReplica.getPartitionId().toPathString())) {
            resourceName = rs;
            break;
        }
    }
    List<String> disabledPartition = currentInstanceConfig.getDisabledPartitions(resourceName);
    assertEquals("Disabled partition not as expected", Collections.singletonList(removedReplica.getPartitionId().toPathString()), disabledPartition);
    // Verify that IdealState has no change
    verifyIdealStateForPartition(removedReplica, true, 3, expectedResourceCount);
    // Verify the InstanceConfig is changed in MapFields (Disabled partitions are added to this field, also the replica entry has been removed)
    String disabledPartitionStr = currentInstanceConfig.getRecord().getMapFields().keySet().stream().filter(k -> !k.startsWith("/mnt")).findFirst().get();
    // Verify the disabled partition string contains correct partition
    Map<String, String> expectedDisabledPartitionMap = new HashMap<>();
    expectedDisabledPartitionMap.put(resourceName, removedReplica.getPartitionId().toPathString());
    assertEquals("Mismatch in disabled partition string in InstanceConfig", expectedDisabledPartitionMap, currentInstanceConfig.getRecord().getMapField(disabledPartitionStr));
    // verify the removed replica is no longer in InstanceConfig
    currentDataNodeConfig = dataNodeConfigSourceType == DataNodeConfigSourceType.INSTANCE_CONFIG ? instanceConfigConverter.convert(admin.getInstanceConfig(clusterName, instanceName)) : propertyStoreAdapter.get(instanceName);
    verifyReplicaInfoInDataNodeConfig(currentDataNodeConfig, removedReplica, false);
    if (propertyStoreAdapter != null) {
        propertyStoreAdapter.close();
    }
}

Example 90 with ClusterMapConfig

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

the class HelixBootstrapUpgradeToolTest method testCreateInstanceConfig.

/**
 * Tests the method to create instance config and ensures that derived fields are set correctly.
 */
@Test
public void testCreateInstanceConfig() {
    DataNode dataNode = (DataNode) ((Partition) testPartitionLayout.getPartitionLayout().getPartitions(DEFAULT_PARTITION_CLASS).get(0)).getReplicas().get(0).getDataNodeId();
    Map<String, Set<String>> partitionToInstances = new HashMap<>();
    JSONObject jsonObject = dataNode.toJSONObject();
    jsonObject.put("xid", ClusterMapUtils.DEFAULT_XID);
    ClusterMapConfig clusterMapConfig = testHardwareLayout.clusterMapConfig;
    dataNode = new DataNode(dataNode.getDatacenter(), jsonObject, clusterMapConfig);
    InstanceConfig referenceInstanceConfig = HelixBootstrapUpgradeUtil.createInstanceConfigFromStaticInfo(dataNode, partitionToInstances, new ConcurrentHashMap<>(), null);
    // Assert that xid field does not get set in InstanceConfig when it is the default.
    assertNull(referenceInstanceConfig.getRecord().getSimpleField(ClusterMapUtils.XID_STR));
    // Assert that xid field does get set if not the default.
    jsonObject.put("xid", "10");
    dataNode = new DataNode(dataNode.getDatacenter(), jsonObject, clusterMapConfig);
    InstanceConfig instanceConfig = HelixBootstrapUpgradeUtil.createInstanceConfigFromStaticInfo(dataNode, partitionToInstances, new ConcurrentHashMap<>(), null);
    assertEquals("10", instanceConfig.getRecord().getSimpleField(ClusterMapUtils.XID_STR));
    assertThat(referenceInstanceConfig.getRecord(), not(equalTo(instanceConfig.getRecord())));
    referenceInstanceConfig = instanceConfig;
    // Assert that sealed list being different does not affect equality
    List<String> sealedList = Arrays.asList("5", "10");
    referenceInstanceConfig.getRecord().setListField(ClusterMapUtils.SEALED_STR, sealedList);
    // set the field to null. The created InstanceConfig should not have null fields.
    referenceInstanceConfig.getRecord().setListField(ClusterMapUtils.STOPPED_REPLICAS_STR, null);
    instanceConfig = HelixBootstrapUpgradeUtil.createInstanceConfigFromStaticInfo(dataNode, partitionToInstances, new ConcurrentHashMap<>(), referenceInstanceConfig);
    // Stopped replicas should be an empty list and not null, so set that in referenceInstanceConfig for comparison.
    referenceInstanceConfig.getRecord().setListField(ClusterMapUtils.STOPPED_REPLICAS_STR, Collections.emptyList());
    assertEquals(instanceConfig.getRecord(), referenceInstanceConfig.getRecord());
    // Assert that stopped list being different does not affect equality
    List<String> stoppedReplicas = Arrays.asList("11", "15");
    referenceInstanceConfig.getRecord().setListField(ClusterMapUtils.STOPPED_REPLICAS_STR, stoppedReplicas);
    instanceConfig = HelixBootstrapUpgradeUtil.createInstanceConfigFromStaticInfo(dataNode, partitionToInstances, new ConcurrentHashMap<>(), referenceInstanceConfig);
    assertEquals(instanceConfig.getRecord(), referenceInstanceConfig.getRecord());
}