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

Example 1 with CloudDataNode

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

the class CloudToStoreReplicationManager method handleChangeInVcrNodes.

/**
 * When there is a change in vcr nodes state, update the new list of live vcr nodes.
 * Also if there are nodes that are removed as part of change, then replication from
 * those nodes should stop and the partitions should find new nodes to replicate from.
 * Note that this method is not thread safe in the wake of arriving helix notifications.
 * @param newVcrNodes Set of new vcr nodes.
 */
private void handleChangeInVcrNodes(Set<CloudDataNode> newVcrNodes) {
    Set<CloudDataNode> removedNodes = new HashSet<>(vcrNodes.get());
    removedNodes.removeAll(newVcrNodes);
    vcrNodes.set(new ArrayList<>(newVcrNodes));
    logger.info("Handling VCR nodes change. The removed vcr nodes: {}", removedNodes);
    List<PartitionId> partitionsOnRemovedNodes = getPartitionsOnNodes(removedNodes);
    for (PartitionId partitionId : partitionsOnRemovedNodes) {
        try {
            // We first remove replica to stop replication from removed node, and then add replica so that it can pick a
            // new cloud node to start replicating from.
            removeCloudReplica(partitionId.toPathString());
            addCloudReplica(partitionId.toPathString());
        } catch (ReplicationException rex) {
            replicationMetrics.addCloudPartitionErrorCount.inc();
            logger.error("Exception {} during remove/add replica for partitionId {}", rex, partitionId);
        }
    }
}

Example 2 with CloudDataNode

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

the class ReplicaThread method getMessagesForMissingKeys.

/**
 * Gets the messages for the keys that are missing from the local store by issuing a {@link GetRequest} to the remote
 * node, if there are any missing keys. If there are no missing keys to be fetched, then no request is issued and a
 * null response is returned.
 * @param connectedChannel The connection channel to the remote node
 * @param exchangeMetadataResponseList The list of metadata response from the remote node
 * @param replicasToReplicatePerNode The list of remote replicas for the remote node
 * @param remoteNode The remote node from which replication needs to happen
 * @param remoteColoGetRequestForStandby boolean which indicates if we are getting missing keys for standby or
 *                                       non-leader replica pairs during leader-based replication.
 * @return The response that contains the missing messages; or null if no request was issued because there were no
 * keys missing.
 * @throws ReplicationException
 * @throws IOException
 */
private GetResponse getMessagesForMissingKeys(ConnectedChannel connectedChannel, List<ExchangeMetadataResponse> exchangeMetadataResponseList, List<RemoteReplicaInfo> replicasToReplicatePerNode, DataNodeId remoteNode, boolean remoteColoGetRequestForStandby) throws ReplicationException, IOException {
    List<PartitionRequestInfo> partitionRequestInfoList = new ArrayList<PartitionRequestInfo>();
    for (int i = 0; i < exchangeMetadataResponseList.size(); i++) {
        ExchangeMetadataResponse exchangeMetadataResponse = exchangeMetadataResponseList.get(i);
        RemoteReplicaInfo remoteReplicaInfo = replicasToReplicatePerNode.get(i);
        if (exchangeMetadataResponse.serverErrorCode == ServerErrorCode.No_Error) {
            Set<StoreKey> missingStoreKeys = exchangeMetadataResponse.getMissingStoreKeys();
            if (missingStoreKeys.size() > 0) {
                if (remoteNode instanceof CloudDataNode) {
                    logger.trace("Replicating blobs from CloudDataNode: {}", missingStoreKeys);
                }
                ArrayList<BlobId> keysToFetch = new ArrayList<BlobId>();
                for (StoreKey storeKey : missingStoreKeys) {
                    keysToFetch.add((BlobId) storeKey);
                }
                PartitionRequestInfo partitionRequestInfo = new PartitionRequestInfo(remoteReplicaInfo.getReplicaId().getPartitionId(), keysToFetch);
                partitionRequestInfoList.add(partitionRequestInfo);
            }
        }
    }
    GetResponse getResponse = null;
    if (!partitionRequestInfoList.isEmpty()) {
        GetRequest getRequest = new GetRequest(correlationIdGenerator.incrementAndGet(), GetRequest.Replication_Client_Id_Prefix + dataNodeId.getHostname() + "[" + dataNodeId.getDatacenterName() + "]", MessageFormatFlags.All, partitionRequestInfoList, replicationConfig.replicationIncludeAll ? GetOption.Include_All : GetOption.None);
        long startTime = time.milliseconds();
        try {
            ChannelOutput channelOutput = connectedChannel.sendAndReceive(getRequest);
            getResponse = GetResponse.readFrom(channelOutput.getInputStream(), clusterMap);
            long getRequestTime = time.milliseconds() - startTime;
            replicationMetrics.updateGetRequestTime(getRequestTime, replicatingFromRemoteColo, replicatingOverSsl, datacenterName, remoteColoGetRequestForStandby);
            if (getResponse.getError() != ServerErrorCode.No_Error) {
                logger.error("Remote node: {} Thread name: {} Remote replicas: {} GetResponse from replication: {}", remoteNode, threadName, replicasToReplicatePerNode, getResponse.getError());
                throw new ReplicationException(" Get Request returned error when trying to get missing keys " + getResponse.getError());
            }
        } catch (IOException e) {
            responseHandler.onEvent(replicasToReplicatePerNode.get(0).getReplicaId(), e);
            throw e;
        }
    }
    return getResponse;
}

Example 3 with CloudDataNode

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

the class CloudRouterFactory method getRequestHandlerPool.

/**
 * Utility method to build  a {@link RequestHandlerPool}.
 * @param verifiableProperties the properties to use.
 * @param clusterMap the {@link ClusterMap} to use.
 * @return the constructed {@link RequestHandlerPool}.
 * @throws Exception if the construction fails.
 */
public RequestHandlerPool getRequestHandlerPool(VerifiableProperties verifiableProperties, ClusterMap clusterMap, CloudDestination cloudDestination, CloudConfig cloudConfig) throws Exception {
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(verifiableProperties);
    MetricRegistry registry = clusterMap.getMetricRegistry();
    DataNodeId nodeId = new CloudDataNode(cloudConfig, clusterMapConfig);
    VcrMetrics vcrMetrics = new VcrMetrics(registry);
    StoreManager cloudStorageManager = new CloudStorageManager(verifiableProperties, vcrMetrics, cloudDestination, clusterMap);
    LocalRequestResponseChannel channel = new LocalRequestResponseChannel();
    ServerMetrics serverMetrics = new ServerMetrics(registry, AmbryRequests.class);
    StoreKeyFactory storeKeyFactory = new BlobIdFactory(clusterMap);
    StoreKeyConverterFactory storeKeyConverterFactory = Utils.getObj(routerConfig.routerStoreKeyConverterFactory, verifiableProperties, registry);
    // A null notification system is passed into AmbryRequests so that replication events are not emitted from a
    // frontend.
    AmbryRequests requests = new AmbryRequests(cloudStorageManager, channel, clusterMap, nodeId, registry, serverMetrics, null, null, null, storeKeyFactory, storeKeyConverterFactory);
    return new RequestHandlerPool(routerConfig.routerRequestHandlerNumOfThreads, channel, requests);
}

Example 4 with CloudDataNode

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

the class CloudBlobStoreTest method testPutWithTtl.

/**
 * Test PUT(with TTL) and TtlUpdate record replication.
 * Replication may happen after PUT and after TtlUpdate, or after TtlUpdate only.
 * PUT may already expired, expiration time < upload threshold or expiration time >= upload threshold.
 * @throws Exception
 */
@Test
public void testPutWithTtl() throws Exception {
    // Set up remote host
    MockClusterMap clusterMap = new MockClusterMap();
    MockHost remoteHost = getLocalAndRemoteHosts(clusterMap).getSecond();
    List<PartitionId> partitionIds = clusterMap.getWritablePartitionIds(null);
    PartitionId partitionId = partitionIds.get(0);
    StoreKeyFactory storeKeyFactory = new BlobIdFactory(clusterMap);
    MockStoreKeyConverterFactory storeKeyConverterFactory = new MockStoreKeyConverterFactory(null, null);
    storeKeyConverterFactory.setConversionMap(new HashMap<>());
    storeKeyConverterFactory.setReturnInputIfAbsent(true);
    MockStoreKeyConverterFactory.MockStoreKeyConverter storeKeyConverter = storeKeyConverterFactory.getStoreKeyConverter();
    Transformer transformer = new BlobIdTransformer(storeKeyFactory, storeKeyConverter);
    Map<DataNodeId, MockHost> hosts = new HashMap<>();
    hosts.put(remoteHost.dataNodeId, remoteHost);
    MockConnectionPool connectionPool = new MockConnectionPool(hosts, clusterMap, 4);
    // Generate BlobIds for following PUT.
    short blobIdVersion = CommonTestUtils.getCurrentBlobIdVersion();
    short accountId = Utils.getRandomShort(TestUtils.RANDOM);
    short containerId = Utils.getRandomShort(TestUtils.RANDOM);
    boolean toEncrypt = TestUtils.RANDOM.nextBoolean();
    List<BlobId> blobIdList = new ArrayList<>();
    for (int i = 0; i < 6; i++) {
        blobIdList.add(new BlobId(blobIdVersion, BlobId.BlobIdType.NATIVE, ClusterMap.UNKNOWN_DATACENTER_ID, accountId, containerId, partitionId, toEncrypt, BlobId.BlobDataType.DATACHUNK));
    }
    // Set up VCR
    Properties props = new Properties();
    setBasicProperties(props);
    props.setProperty("clustermap.port", "12300");
    props.setProperty("vcr.ssl.port", "12345");
    ReplicationConfig replicationConfig = new ReplicationConfig(new VerifiableProperties(props));
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(new VerifiableProperties(props));
    CloudConfig cloudConfig = new CloudConfig(new VerifiableProperties(props));
    CloudDataNode cloudDataNode = new CloudDataNode(cloudConfig, clusterMapConfig);
    LatchBasedInMemoryCloudDestination latchBasedInMemoryCloudDestination = new LatchBasedInMemoryCloudDestination(blobIdList, clusterMap);
    CloudReplica cloudReplica = new CloudReplica(partitionId, cloudDataNode);
    CloudBlobStore cloudBlobStore = new CloudBlobStore(new VerifiableProperties(props), partitionId, latchBasedInMemoryCloudDestination, clusterMap, new VcrMetrics(new MetricRegistry()));
    cloudBlobStore.start();
    // Create ReplicaThread and add RemoteReplicaInfo to it.
    ReplicationMetrics replicationMetrics = new ReplicationMetrics(new MetricRegistry(), Collections.emptyList());
    ReplicaThread replicaThread = new ReplicaThread("threadtest", new MockFindTokenHelper(storeKeyFactory, replicationConfig), clusterMap, new AtomicInteger(0), cloudDataNode, connectionPool, replicationConfig, replicationMetrics, null, storeKeyConverter, transformer, clusterMap.getMetricRegistry(), false, cloudDataNode.getDatacenterName(), new ResponseHandler(clusterMap), new MockTime(), null, null, null);
    for (ReplicaId replica : partitionId.getReplicaIds()) {
        if (replica.getDataNodeId() == remoteHost.dataNodeId) {
            RemoteReplicaInfo remoteReplicaInfo = new RemoteReplicaInfo(replica, cloudReplica, cloudBlobStore, new MockFindToken(0, 0), Long.MAX_VALUE, SystemTime.getInstance(), new Port(remoteHost.dataNodeId.getPort(), PortType.PLAINTEXT));
            replicaThread.addRemoteReplicaInfo(remoteReplicaInfo);
            break;
        }
    }
    long referenceTime = System.currentTimeMillis();
    // Case 1: Put already expired. Replication happens after Put and after TtlUpdate.
    // Upload to Cloud only after replicating ttlUpdate.
    BlobId id = blobIdList.get(0);
    addPutMessagesToReplicasOfPartition(id, accountId, containerId, partitionId, Collections.singletonList(remoteHost), referenceTime - 2000, referenceTime - 1000);
    replicaThread.replicate();
    assertFalse("Blob should not exist.", latchBasedInMemoryCloudDestination.doesBlobExist(id));
    addTtlUpdateMessagesToReplicasOfPartition(partitionId, id, Collections.singletonList(remoteHost), Utils.Infinite_Time);
    replicaThread.replicate();
    assertTrue("Blob should exist.", latchBasedInMemoryCloudDestination.doesBlobExist(id));
    // Case 2: Put already expired. Replication happens after TtlUpdate.
    // Upload to Cloud only after replicating ttlUpdate.
    id = blobIdList.get(1);
    addPutMessagesToReplicasOfPartition(id, accountId, containerId, partitionId, Collections.singletonList(remoteHost), referenceTime - 2000, referenceTime - 1000);
    addTtlUpdateMessagesToReplicasOfPartition(partitionId, id, Collections.singletonList(remoteHost), Utils.Infinite_Time);
    replicaThread.replicate();
    assertTrue("Blob should exist.", latchBasedInMemoryCloudDestination.doesBlobExist(id));
    // Case 3: Put TTL less than cloudConfig.vcrMinTtlDays. Replication happens after Put and after TtlUpdate.
    // Upload to Cloud only after replicating ttlUpdate.
    id = blobIdList.get(2);
    addPutMessagesToReplicasOfPartition(id, accountId, containerId, partitionId, Collections.singletonList(remoteHost), referenceTime, referenceTime + TimeUnit.DAYS.toMillis(cloudConfig.vcrMinTtlDays) - 1);
    replicaThread.replicate();
    if (isVcr) {
        assertFalse("Blob should not exist (vcr).", latchBasedInMemoryCloudDestination.doesBlobExist(id));
    } else {
        assertTrue("Blob should exist (not vcr).", latchBasedInMemoryCloudDestination.doesBlobExist(id));
    }
    addTtlUpdateMessagesToReplicasOfPartition(partitionId, id, Collections.singletonList(remoteHost), Utils.Infinite_Time);
    replicaThread.replicate();
    assertTrue("Blob should exist.", latchBasedInMemoryCloudDestination.doesBlobExist(id));
    // Case 4: Put TTL less than cloudConfig.vcrMinTtlDays. Replication happens after TtlUpdate.
    // Upload to Cloud only after replicating ttlUpdate.
    id = blobIdList.get(3);
    addPutMessagesToReplicasOfPartition(id, accountId, containerId, partitionId, Collections.singletonList(remoteHost), referenceTime, referenceTime + TimeUnit.DAYS.toMillis(cloudConfig.vcrMinTtlDays) - 1);
    addTtlUpdateMessagesToReplicasOfPartition(partitionId, id, Collections.singletonList(remoteHost), Utils.Infinite_Time);
    replicaThread.replicate();
    assertTrue("Blob should exist.", latchBasedInMemoryCloudDestination.doesBlobExist(id));
    // Case 5: Put TTL greater than or equals to cloudConfig.vcrMinTtlDays. Replication happens after Put and after TtlUpdate.
    // Upload to Cloud after Put and update ttl after TtlUpdate.
    id = blobIdList.get(4);
    addPutMessagesToReplicasOfPartition(id, accountId, containerId, partitionId, Collections.singletonList(remoteHost), referenceTime, referenceTime + TimeUnit.DAYS.toMillis(cloudConfig.vcrMinTtlDays));
    replicaThread.replicate();
    assertTrue(latchBasedInMemoryCloudDestination.doesBlobExist(id));
    addTtlUpdateMessagesToReplicasOfPartition(partitionId, id, Collections.singletonList(remoteHost), Utils.Infinite_Time);
    replicaThread.replicate();
    assertTrue("Blob should exist.", latchBasedInMemoryCloudDestination.doesBlobExist(id));
    // Case 6: Put TTL greater than or equals to cloudConfig.vcrMinTtlDays. Replication happens after TtlUpdate.
    // Upload to Cloud after TtlUpdate.
    id = blobIdList.get(5);
    addPutMessagesToReplicasOfPartition(id, accountId, containerId, partitionId, Collections.singletonList(remoteHost), referenceTime, referenceTime + TimeUnit.DAYS.toMillis(cloudConfig.vcrMinTtlDays));
    addTtlUpdateMessagesToReplicasOfPartition(partitionId, id, Collections.singletonList(remoteHost), Utils.Infinite_Time);
    replicaThread.replicate();
    assertTrue("Blob should exist.", latchBasedInMemoryCloudDestination.doesBlobExist(id));
    // Verify expiration time of all blobs.
    Map<String, CloudBlobMetadata> map = latchBasedInMemoryCloudDestination.getBlobMetadata(blobIdList);
    for (BlobId blobId : blobIdList) {
        assertEquals("Blob ttl should be infinite now.", Utils.Infinite_Time, map.get(blobId.toString()).getExpirationTime());
    }
}

Example 5 with CloudDataNode

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

the class CloudTokenPersistorTest method basicTest.

@Test
public void basicTest() throws Exception {
    Properties props = VcrTestUtil.createVcrProperties("DC1", "vcrClusterName", "zkConnectString", 12310, 12410, 12510, null);
    props.setProperty("replication.cloud.token.factory", replicationCloudTokenFactory);
    CloudConfig cloudConfig = new CloudConfig(new VerifiableProperties(props));
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(new VerifiableProperties(props));
    ClusterMap clusterMap = new MockClusterMap();
    DataNodeId dataNodeId = new CloudDataNode(cloudConfig, clusterMapConfig);
    Map<String, Set<PartitionInfo>> mountPathToPartitionInfoList = new HashMap<>();
    BlobIdFactory blobIdFactory = new BlobIdFactory(clusterMap);
    StoreFindTokenFactory factory = new StoreFindTokenFactory(blobIdFactory);
    PartitionId partitionId = clusterMap.getAllPartitionIds(null).get(0);
    ReplicaId cloudReplicaId = new CloudReplica(partitionId, dataNodeId);
    List<? extends ReplicaId> peerReplicas = cloudReplicaId.getPeerReplicaIds();
    List<RemoteReplicaInfo> remoteReplicas = new ArrayList<RemoteReplicaInfo>();
    List<RemoteReplicaInfo.ReplicaTokenInfo> replicaTokenInfos = new ArrayList<>();
    for (ReplicaId remoteReplica : peerReplicas) {
        RemoteReplicaInfo remoteReplicaInfo = new RemoteReplicaInfo(remoteReplica, cloudReplicaId, null, factory.getNewFindToken(), 10, SystemTime.getInstance(), remoteReplica.getDataNodeId().getPortToConnectTo());
        remoteReplicas.add(remoteReplicaInfo);
        replicaTokenInfos.add(new RemoteReplicaInfo.ReplicaTokenInfo(remoteReplicaInfo));
    }
    PartitionInfo partitionInfo = new PartitionInfo(remoteReplicas, partitionId, null, cloudReplicaId);
    mountPathToPartitionInfoList.computeIfAbsent(cloudReplicaId.getMountPath(), key -> ConcurrentHashMap.newKeySet()).add(partitionInfo);
    LatchBasedInMemoryCloudDestination cloudDestination = new LatchBasedInMemoryCloudDestination(Collections.emptyList(), AzureCloudDestinationFactory.getReplicationFeedType(new VerifiableProperties(props)), clusterMap);
    ReplicationConfig replicationConfig = new ReplicationConfig(new VerifiableProperties(props));
    CloudTokenPersistor cloudTokenPersistor = new CloudTokenPersistor("replicaTokens", mountPathToPartitionInfoList, new ReplicationMetrics(new MetricRegistry(), Collections.emptyList()), clusterMap, new FindTokenHelper(blobIdFactory, replicationConfig), cloudDestination);
    cloudTokenPersistor.persist(cloudReplicaId.getMountPath(), replicaTokenInfos);
    List<RemoteReplicaInfo.ReplicaTokenInfo> retrievedReplicaTokenInfos = cloudTokenPersistor.retrieve(cloudReplicaId.getMountPath());
    Assert.assertEquals("Number of tokens doesn't match.", replicaTokenInfos.size(), retrievedReplicaTokenInfos.size());
    for (int i = 0; i < replicaTokenInfos.size(); i++) {
        Assert.assertArrayEquals("Token is not correct.", replicaTokenInfos.get(i).getReplicaToken().toBytes(), retrievedReplicaTokenInfos.get(i).getReplicaToken().toBytes());
    }
}