Examples with ReplicationManager com.github.ambry.replication.ReplicationManager used on opensource projects

Example 1 with ReplicationManager

use of com.github.ambry.replication.ReplicationManager in project ambry by linkedin.

the class AmbryServer method startup.

public void startup() throws InstantiationException {
    try {
        logger.info("starting");
        clusterParticipants = clusterAgentsFactory.getClusterParticipants();
        logger.info("Setting up JMX.");
        long startTime = SystemTime.getInstance().milliseconds();
        reporter = reporterFactory != null ? reporterFactory.apply(registry) : JmxReporter.forRegistry(registry).build();
        reporter.start();
        logger.info("creating configs");
        NetworkConfig networkConfig = new NetworkConfig(properties);
        StoreConfig storeConfig = new StoreConfig(properties);
        DiskManagerConfig diskManagerConfig = new DiskManagerConfig(properties);
        ServerConfig serverConfig = new ServerConfig(properties);
        ReplicationConfig replicationConfig = new ReplicationConfig(properties);
        ConnectionPoolConfig connectionPoolConfig = new ConnectionPoolConfig(properties);
        SSLConfig sslConfig = new SSLConfig(properties);
        ClusterMapConfig clusterMapConfig = new ClusterMapConfig(properties);
        StatsManagerConfig statsConfig = new StatsManagerConfig(properties);
        CloudConfig cloudConfig = new CloudConfig(properties);
        // verify the configs
        properties.verify();
        scheduler = Utils.newScheduler(serverConfig.serverSchedulerNumOfthreads, false);
        // mismatch in sealed/stopped replica lists that maintained by each participant.
        if (clusterParticipants != null && clusterParticipants.size() > 1 && serverConfig.serverParticipantsConsistencyCheckerPeriodSec > 0) {
            consistencyChecker = new ParticipantsConsistencyChecker(clusterParticipants, metrics);
            logger.info("Scheduling participants consistency checker with a period of {} secs", serverConfig.serverParticipantsConsistencyCheckerPeriodSec);
            consistencyCheckerScheduler = Utils.newScheduler(1, "consistency-checker-", false);
            consistencyCheckerTask = consistencyCheckerScheduler.scheduleAtFixedRate(consistencyChecker, 0, serverConfig.serverParticipantsConsistencyCheckerPeriodSec, TimeUnit.SECONDS);
        }
        logger.info("checking if node exists in clustermap host {} port {}", networkConfig.hostName, networkConfig.port);
        DataNodeId nodeId = clusterMap.getDataNodeId(networkConfig.hostName, networkConfig.port);
        if (nodeId == null) {
            throw new IllegalArgumentException("The node " + networkConfig.hostName + ":" + networkConfig.port + "is not present in the clustermap. Failing to start the datanode");
        }
        AccountServiceFactory accountServiceFactory = Utils.getObj(serverConfig.serverAccountServiceFactory, properties, registry);
        AccountService accountService = accountServiceFactory.getAccountService();
        StoreKeyFactory storeKeyFactory = Utils.getObj(storeConfig.storeKeyFactory, clusterMap);
        // In most cases, there should be only one participant in the clusterParticipants list. If there are more than one
        // and some components require sole participant, the first one in the list will be primary participant.
        storageManager = new StorageManager(storeConfig, diskManagerConfig, scheduler, registry, storeKeyFactory, clusterMap, nodeId, new BlobStoreHardDelete(), clusterParticipants, time, new BlobStoreRecovery(), accountService);
        storageManager.start();
        SSLFactory sslFactory = new NettySslHttp2Factory(sslConfig);
        if (clusterMapConfig.clusterMapEnableHttp2Replication) {
            connectionPool = new Http2BlockingChannelPool(sslFactory, new Http2ClientConfig(properties), new Http2ClientMetrics(registry));
        } else {
            connectionPool = new BlockingChannelConnectionPool(connectionPoolConfig, sslConfig, clusterMapConfig, registry);
        }
        connectionPool.start();
        StoreKeyConverterFactory storeKeyConverterFactory = Utils.getObj(serverConfig.serverStoreKeyConverterFactory, properties, registry);
        Predicate<MessageInfo> skipPredicate = new ReplicationSkipPredicate(accountService, replicationConfig);
        replicationManager = new ReplicationManager(replicationConfig, clusterMapConfig, storeConfig, storageManager, storeKeyFactory, clusterMap, scheduler, nodeId, connectionPool, registry, notificationSystem, storeKeyConverterFactory, serverConfig.serverMessageTransformer, clusterParticipants.get(0), skipPredicate);
        replicationManager.start();
        if (replicationConfig.replicationEnabledWithVcrCluster) {
            logger.info("Creating Helix cluster spectator for cloud to store replication.");
            vcrClusterSpectator = _vcrClusterAgentsFactory.getVcrClusterSpectator(cloudConfig, clusterMapConfig);
            cloudToStoreReplicationManager = new CloudToStoreReplicationManager(replicationConfig, clusterMapConfig, storeConfig, storageManager, storeKeyFactory, clusterMap, scheduler, nodeId, connectionPool, registry, notificationSystem, storeKeyConverterFactory, serverConfig.serverMessageTransformer, vcrClusterSpectator, clusterParticipants.get(0));
            cloudToStoreReplicationManager.start();
        }
        logger.info("Creating StatsManager to publish stats");
        accountStatsMySqlStore = statsConfig.enableMysqlReport ? (AccountStatsMySqlStore) new AccountStatsMySqlStoreFactory(properties, clusterMapConfig, registry).getAccountStatsStore() : null;
        statsManager = new StatsManager(storageManager, clusterMap.getReplicaIds(nodeId), registry, statsConfig, time, clusterParticipants.get(0), accountStatsMySqlStore, accountService);
        if (serverConfig.serverStatsPublishLocalEnabled) {
            statsManager.start();
        }
        ArrayList<Port> ports = new ArrayList<Port>();
        ports.add(new Port(networkConfig.port, PortType.PLAINTEXT));
        if (nodeId.hasSSLPort()) {
            ports.add(new Port(nodeId.getSSLPort(), PortType.SSL));
        }
        networkServer = new SocketServer(networkConfig, sslConfig, registry, ports);
        FindTokenHelper findTokenHelper = new FindTokenHelper(storeKeyFactory, replicationConfig);
        requests = new AmbryServerRequests(storageManager, networkServer.getRequestResponseChannel(), clusterMap, nodeId, registry, metrics, findTokenHelper, notificationSystem, replicationManager, storeKeyFactory, serverConfig, storeKeyConverterFactory, statsManager, clusterParticipants.get(0));
        requestHandlerPool = new RequestHandlerPool(serverConfig.serverRequestHandlerNumOfThreads, networkServer.getRequestResponseChannel(), requests);
        networkServer.start();
        // Start netty http2 server
        if (nodeId.hasHttp2Port()) {
            NettyConfig nettyConfig = new NettyConfig(properties);
            NettyMetrics nettyMetrics = new NettyMetrics(registry);
            Http2ServerMetrics http2ServerMetrics = new Http2ServerMetrics(registry);
            Http2ClientConfig http2ClientConfig = new Http2ClientConfig(properties);
            logger.info("Http2 port {} is enabled. Starting HTTP/2 service.", nodeId.getHttp2Port());
            NettyServerRequestResponseChannel requestResponseChannel = new NettyServerRequestResponseChannel(networkConfig.queuedMaxRequests, http2ServerMetrics);
            AmbryServerRequests ambryServerRequestsForHttp2 = new AmbryServerRequests(storageManager, requestResponseChannel, clusterMap, nodeId, registry, metrics, findTokenHelper, notificationSystem, replicationManager, storeKeyFactory, serverConfig, storeKeyConverterFactory, statsManager, clusterParticipants.get(0));
            requestHandlerPoolForHttp2 = new RequestHandlerPool(serverConfig.serverRequestHandlerNumOfThreads, requestResponseChannel, ambryServerRequestsForHttp2);
            NioServerFactory nioServerFactory = new StorageServerNettyFactory(nodeId.getHttp2Port(), requestResponseChannel, sslFactory, nettyConfig, http2ClientConfig, metrics, nettyMetrics, http2ServerMetrics, serverSecurityService);
            nettyHttp2Server = nioServerFactory.getNioServer();
            nettyHttp2Server.start();
        }
        // Other code
        List<AmbryStatsReport> ambryStatsReports = new ArrayList<>();
        Set<String> validStatsTypes = new HashSet<>();
        for (StatsReportType type : StatsReportType.values()) {
            validStatsTypes.add(type.toString());
        }
        if (serverConfig.serverStatsPublishReportEnabled) {
            serverConfig.serverStatsReportsToPublish.forEach(e -> {
                if (validStatsTypes.contains(e)) {
                    ambryStatsReports.add(new AmbryStatsReportImpl(serverConfig.serverQuotaStatsAggregateIntervalInMinutes, StatsReportType.valueOf(e)));
                }
            });
        }
        if (vcrClusterSpectator != null) {
            vcrClusterSpectator.spectate();
        }
        Callback<StatsSnapshot> accountServiceCallback = new AccountServiceCallback(accountService);
        for (ClusterParticipant clusterParticipant : clusterParticipants) {
            clusterParticipant.participate(ambryStatsReports, accountStatsMySqlStore, accountServiceCallback);
        }
        if (nettyInternalMetrics != null) {
            nettyInternalMetrics.start();
            logger.info("NettyInternalMetric starts");
        }
        logger.info("started");
        long processingTime = SystemTime.getInstance().milliseconds() - startTime;
        metrics.serverStartTimeInMs.update(processingTime);
        logger.info("Server startup time in Ms {}", processingTime);
    } catch (Exception e) {
        logger.error("Error during startup", e);
        throw new InstantiationException("failure during startup " + e);
    }
}

Example 2 with ReplicationManager

use of com.github.ambry.replication.ReplicationManager in project ambry by linkedin.

the class AmbryServerRequestsTest method addBlobStoreFailureTest.

@Test
public void addBlobStoreFailureTest() throws Exception {
    // create newPartition1 that no replica sits on current node.
    List<MockDataNodeId> dataNodes = clusterMap.getDataNodes().stream().filter(node -> !node.getHostname().equals(dataNodeId.getHostname()) || node.getPort() != dataNodeId.getPort()).collect(Collectors.toList());
    PartitionId newPartition1 = clusterMap.createNewPartition(dataNodes);
    // test that getting new replica from cluster map fails
    sendAndVerifyStoreControlRequest(newPartition1, BlobStoreControlAction.AddStore, (short) 0, ServerErrorCode.Replica_Unavailable);
    // create newPartition2 that has one replica on current node
    PartitionId newPartition2 = clusterMap.createNewPartition(clusterMap.getDataNodes());
    // test that adding store into StorageManager fails
    storageManager.returnValueOfAddBlobStore = false;
    sendAndVerifyStoreControlRequest(newPartition2, BlobStoreControlAction.AddStore, (short) 0, ServerErrorCode.Unknown_Error);
    storageManager.returnValueOfAddBlobStore = true;
    // test that adding replica into ReplicationManager fails (we first add replica into ReplicationManager to trigger failure)
    ReplicaId replicaToAdd = clusterMap.getBootstrapReplica(newPartition2.toPathString(), dataNodeId);
    replicationManager.addReplica(replicaToAdd);
    sendAndVerifyStoreControlRequest(newPartition2, BlobStoreControlAction.AddStore, (short) 0, ServerErrorCode.Unknown_Error);
    assertTrue("Remove replica from replication manager should succeed.", replicationManager.removeReplica(replicaToAdd));
    // test that adding replica into StatsManager fails
    statsManager.returnValOfAddReplica = false;
    sendAndVerifyStoreControlRequest(newPartition2, BlobStoreControlAction.AddStore, (short) 0, ServerErrorCode.Unknown_Error);
    statsManager.returnValOfAddReplica = true;
}

Example 3 with ReplicationManager

use of com.github.ambry.replication.ReplicationManager in project ambry by linkedin.

the class AmbryServerRequests method handleRemoveStoreRequest.

/**
 * Handles admin request that removes a BlobStore from current node
 * @param partitionId the {@link PartitionId} associated with BlobStore
 * @return {@link ServerErrorCode} represents result of handling admin request.
 */
private ServerErrorCode handleRemoveStoreRequest(PartitionId partitionId) throws StoreException, IOException {
    ServerErrorCode errorCode = ServerErrorCode.No_Error;
    ReplicaId replicaId = storeManager.getReplica(partitionId.toPathString());
    if (replicaId == null) {
        logger.error("{} doesn't exist on current node", partitionId);
        return ServerErrorCode.Partition_Unknown;
    }
    // Attempt to remove replica from stats manager. If replica doesn't exist, log info but don't fail the request
    statsManager.removeReplica(replicaId);
    // Attempt to remove replica from replication manager. If replica doesn't exist, log info but don't fail the request
    ((ReplicationManager) replicationEngine).removeReplica(replicaId);
    Store store = ((StorageManager) storeManager).getStore(partitionId, true);
    // Attempt to remove store from storage manager.
    if (storeManager.removeBlobStore(partitionId) && store != null) {
        ((BlobStore) store).deleteStoreFiles();
        for (ReplicaStatusDelegate replicaStatusDelegate : ((BlobStore) store).getReplicaStatusDelegates()) {
            // Remove store from sealed and stopped list (if present)
            logger.info("Removing store from sealed and stopped list(if present)");
            replicaStatusDelegate.unseal(replicaId);
            replicaStatusDelegate.unmarkStopped(Collections.singletonList(replicaId));
        }
    } else {
        errorCode = ServerErrorCode.Unknown_Error;
    }
    return errorCode;
}