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

Example 56 with DataNodeId

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

the class ReplicaThread method fixMissingStoreKeys.

/**
 * Gets all the messages from the remote node for the missing keys and writes them to the local store
 * @param connectedChannel The connected channel that represents a connection to the remote replica
 * @param replicasToReplicatePerNode The information about the replicas that is being replicated
 * @param exchangeMetadataResponseList The missing keys in the local stores whose message needs to be retrieved
 *                                     from the remote stores
 * @param remoteColoGetRequestForStandby boolean which indicates if we are getting missing keys for standby or
 *                                       non-leader replica pairs during leader-based replication.
 * @throws IOException
 * @throws ReplicationException
 */
void fixMissingStoreKeys(ConnectedChannel connectedChannel, List<RemoteReplicaInfo> replicasToReplicatePerNode, List<ExchangeMetadataResponse> exchangeMetadataResponseList, boolean remoteColoGetRequestForStandby) throws IOException, ReplicationException {
    long fixMissingStoreKeysStartTimeInMs = time.milliseconds();
    GetResponse getResponse = null;
    try {
        if (exchangeMetadataResponseList.size() != replicasToReplicatePerNode.size() || replicasToReplicatePerNode.size() == 0) {
            throw new IllegalArgumentException("ExchangeMetadataResponseList size " + exchangeMetadataResponseList.size() + " and replicasToReplicatePerNode size " + replicasToReplicatePerNode.size() + " should be the same and greater than zero");
        }
        DataNodeId remoteNode = replicasToReplicatePerNode.get(0).getReplicaId().getDataNodeId();
        getResponse = getMessagesForMissingKeys(connectedChannel, exchangeMetadataResponseList, replicasToReplicatePerNode, remoteNode, remoteColoGetRequestForStandby);
        writeMessagesToLocalStoreAndAdvanceTokens(exchangeMetadataResponseList, getResponse, replicasToReplicatePerNode, remoteNode, remoteColoGetRequestForStandby);
    } finally {
        if (getResponse != null && getResponse.getInputStream() instanceof NettyByteBufDataInputStream) {
            // if the InputStream is NettyByteBufDataInputStream based, it's time to release its buffer.
            ((NettyByteBufDataInputStream) (getResponse.getInputStream())).getBuffer().release();
        }
        long fixMissingStoreKeysTime = time.milliseconds() - fixMissingStoreKeysStartTimeInMs;
        replicationMetrics.updateFixMissingStoreKeysTime(fixMissingStoreKeysTime, replicatingFromRemoteColo, replicatingOverSsl, datacenterName);
    }
}

Example 57 with DataNodeId

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

the class ReplicaThread method applyTtlUpdate.

/**
 * Applies a TTL update to the blob described by {@code messageInfo}.
 * @param messageInfo the {@link MessageInfo} that will be transformed into a TTL update
 * @param remoteReplicaInfo The remote replica that is being replicated from
 * @throws StoreException
 */
private void applyTtlUpdate(MessageInfo messageInfo, RemoteReplicaInfo remoteReplicaInfo) throws StoreException {
    DataNodeId remoteNode = remoteReplicaInfo.getReplicaId().getDataNodeId();
    try {
        // NOTE: It is possible that the key in question may have expired and this TTL update is being applied after it
        // is deemed expired. The store will accept the op (BlobStore looks at whether the op was valid to do at the time
        // of the op, not current time) but if compaction is running at the same time and has decided to clean up the
        // record before this ttl update was applied (and this didn't find the key missing because compaction has not yet
        // committed), then we have a bad situation where only a TTL update exists in the store. This problem has to be
        // addressed. This can only happen if replication is far behind (for e.g due to a server being down for a long
        // time). Won't happen if a server is being recreated.
        messageInfo = new MessageInfo.Builder(messageInfo).isTtlUpdated(true).build();
        remoteReplicaInfo.getLocalStore().updateTtl(Collections.singletonList(messageInfo));
        logger.trace("Remote node: {} Thread name: {} Remote replica: {} Key ttl updated id: {}", remoteNode, threadName, remoteReplicaInfo.getReplicaId(), messageInfo.getStoreKey());
    } catch (StoreException e) {
        // The blob may be deleted or updated which is alright
        if (e.getErrorCode() == StoreErrorCodes.ID_Deleted || e.getErrorCode() == StoreErrorCodes.Already_Updated) {
            logger.trace("Remote node: {} Thread name: {} Remote replica: {} Key already updated: {}", remoteNode, threadName, remoteReplicaInfo.getReplicaId(), messageInfo.getStoreKey());
        } else {
            throw e;
        }
    }
    // as long as the update is guaranteed to have taken effect locally.
    if (notification != null) {
        notification.onBlobReplicaUpdated(dataNodeId.getHostname(), dataNodeId.getPort(), messageInfo.getStoreKey().getID(), BlobReplicaSourceType.REPAIRED, UpdateType.TTL_UPDATE, messageInfo);
    }
}

Example 58 with DataNodeId

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

the class CloudToStoreReplicationManager method addCloudReplica.

/**
 * Add a replica of given partition and its {@link RemoteReplicaInfo}s to backup list.
 * @param partitionName name of the partition of the replica to add.
 * @throws ReplicationException if replicas initialization failed.
 */
private void addCloudReplica(String partitionName) throws ReplicationException {
    // Adding cloud replica occurs when replica becomes leader from standby. Hence, if this a new added replica, it
    // should be present in storage manager already.
    ReplicaId localReplica = storeManager.getReplica(partitionName);
    if (localReplica == null) {
        logger.warn("Got partition leader notification for partition {} that is not present on the node", partitionName);
        return;
    }
    PartitionId partitionId = localReplica.getPartitionId();
    Store store = storeManager.getStore(partitionId);
    if (store == null) {
        logger.warn("Unable to add cloud replica for partition {} as store for the partition is not present or started.", partitionName);
        return;
    }
    DataNodeId cloudDataNode = getCloudDataNode();
    CloudReplica peerCloudReplica = new CloudReplica(partitionId, cloudDataNode);
    FindTokenFactory findTokenFactory = tokenHelper.getFindTokenFactoryFromReplicaType(peerCloudReplica.getReplicaType());
    RemoteReplicaInfo remoteReplicaInfo = new RemoteReplicaInfo(peerCloudReplica, localReplica, store, findTokenFactory.getNewFindToken(), storeConfig.storeDataFlushIntervalSeconds * SystemTime.MsPerSec * Replication_Delay_Multiplier, SystemTime.getInstance(), peerCloudReplica.getDataNodeId().getPortToConnectTo());
    replicationMetrics.addMetricsForRemoteReplicaInfo(remoteReplicaInfo, trackPerDatacenterLagInMetric);
    // Note that for each replica on a Ambry server node, there is only one cloud replica that it will be replicating from.
    List<RemoteReplicaInfo> remoteReplicaInfos = Collections.singletonList(remoteReplicaInfo);
    PartitionInfo partitionInfo = new PartitionInfo(remoteReplicaInfos, partitionId, store, localReplica);
    partitionToPartitionInfo.put(partitionId, partitionInfo);
    mountPathToPartitionInfos.computeIfAbsent(localReplica.getMountPath(), key -> ConcurrentHashMap.newKeySet()).add(partitionInfo);
    logger.info("Cloud Partition {} added to {}. CloudNode {} port {}", partitionName, dataNodeId, cloudDataNode, cloudDataNode.getPortToConnectTo());
    // Reload replication token if exist.
    reloadReplicationTokenIfExists(localReplica, remoteReplicaInfos);
    // Add remoteReplicaInfos to {@link ReplicaThread}.
    addRemoteReplicaInfoToReplicaThread(remoteReplicaInfos, true);
    if (replicationConfig.replicationTrackPerPartitionLagFromRemote) {
        replicationMetrics.addLagMetricForPartition(partitionId, true);
    }
    replicationMetrics.addCatchUpPointMetricForPartition(partitionId);
}

Example 59 with DataNodeId

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

the class ReplicaThread method addRemoteReplicaInfo.

/**
 * Add a {@link RemoteReplicaInfo} to current {@link ReplicaThread}.
 * @param remoteReplicaInfo {@link RemoteReplicaInfo} to add.
 */
public void addRemoteReplicaInfo(RemoteReplicaInfo remoteReplicaInfo) {
    lock.lock();
    try {
        allReplicatedPartitions.add(remoteReplicaInfo.getReplicaId().getPartitionId());
        DataNodeId dataNodeId = remoteReplicaInfo.getReplicaId().getDataNodeId();
        if (!replicasToReplicateGroupedByNode.computeIfAbsent(dataNodeId, key -> new HashSet<>()).add(remoteReplicaInfo)) {
            replicationMetrics.remoteReplicaInfoAddError.inc();
            // Since VCR is also listening Ambry Clustermap change, this may happen if events happens in following order:
            // 1. VCR in memory ambry-clustermap updated
            // 2. VcrClusterParticipantListener adds remote replicas for the newly added partition
            // 3. ClusterMapChangeListener adds remote replicas
            logger.warn("ReplicaThread: {}, RemoteReplicaInfo {} already exists.", threadName, remoteReplicaInfo);
        }
    } finally {
        lock.unlock();
    }
    logger.trace("RemoteReplicaInfo {} is added to ReplicaThread {}. Now working on {} dataNodeIds.", remoteReplicaInfo, threadName, replicasToReplicateGroupedByNode.keySet().size());
}

Example 60 with DataNodeId

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

the class ReplicationTest method replicaFromOfflineToBootstrapTest.

/**
 * Test that state transition in replication manager from OFFLINE to BOOTSTRAP
 * @throws Exception
 */
@Test
public void replicaFromOfflineToBootstrapTest() throws Exception {
    MockClusterMap clusterMap = new MockClusterMap();
    ClusterMapConfig clusterMapConfig = new ClusterMapConfig(verifiableProperties);
    MockHelixParticipant.metricRegistry = new MetricRegistry();
    MockHelixParticipant mockHelixParticipant = new MockHelixParticipant(clusterMapConfig);
    DataNodeId currentNode = clusterMap.getDataNodeIds().get(0);
    Pair<StorageManager, ReplicationManager> managers = createStorageManagerAndReplicationManager(clusterMap, clusterMapConfig, mockHelixParticipant);
    StorageManager storageManager = managers.getFirst();
    MockReplicationManager replicationManager = (MockReplicationManager) managers.getSecond();
    assertTrue("State change listener in cluster participant should contain replication manager listener", mockHelixParticipant.getPartitionStateChangeListeners().containsKey(StateModelListenerType.ReplicationManagerListener));
    // 1. test partition not found case (should throw exception)
    try {
        mockHelixParticipant.onPartitionBecomeBootstrapFromOffline("-1");
        fail("should fail because replica is not found");
    } catch (StateTransitionException e) {
        assertEquals("Transition error doesn't match", ReplicaNotFound, e.getErrorCode());
    }
    // 2. create a new partition and test replica addition success case
    ReplicaId newReplicaToAdd = getNewReplicaToAdd(clusterMap);
    PartitionId newPartition = newReplicaToAdd.getPartitionId();
    assertTrue("Adding new replica to Storage Manager should succeed", storageManager.addBlobStore(newReplicaToAdd));
    assertFalse("partitionToPartitionInfo should not contain new partition", replicationManager.partitionToPartitionInfo.containsKey(newPartition));
    mockHelixParticipant.onPartitionBecomeBootstrapFromOffline(newPartition.toPathString());
    assertTrue("partitionToPartitionInfo should contain new partition", replicationManager.partitionToPartitionInfo.containsKey(newPartition));
    // 3. test replica addition failure case
    replicationManager.partitionToPartitionInfo.remove(newPartition);
    replicationManager.addReplicaReturnVal = false;
    try {
        mockHelixParticipant.onPartitionBecomeBootstrapFromOffline(newPartition.toPathString());
        fail("should fail due to replica addition failure");
    } catch (StateTransitionException e) {
        assertEquals("Transition error doesn't match", ReplicaOperationFailure, e.getErrorCode());
    }
    replicationManager.addReplicaReturnVal = null;
    // 4. test OFFLINE -> BOOTSTRAP on existing replica (should be no-op)
    ReplicaId existingReplica = clusterMap.getReplicaIds(currentNode).get(0);
    assertTrue("partitionToPartitionInfo should contain existing partition", replicationManager.partitionToPartitionInfo.containsKey(existingReplica.getPartitionId()));
    mockHelixParticipant.onPartitionBecomeBootstrapFromOffline(existingReplica.getPartitionId().toPathString());
    storageManager.shutdown();
}