Examples with Replica org.apache.asterix.common.replication.Replica used on opensource projects

Example 1 with Replica

use of org.apache.asterix.common.replication.Replica in project asterixdb by apache.

the class ReplicationManager method updateReplicaState.

/**
     * Updates the state of a remote replica.
     *
     * @param replicaId
     *            The replica id to update.
     * @param newState
     *            The new state of the replica.
     * @param suspendReplication
     *            a flag indicating whether to suspend replication on state change or not.
     * @throws InterruptedException
     */
public synchronized void updateReplicaState(String replicaId, ReplicaState newState, boolean suspendReplication) throws InterruptedException {
    Replica replica = replicas.get(replicaId);
    if (replica.getState() == newState) {
        return;
    }
    if (suspendReplication) {
        //prevent new jobs/logs from coming in
        replicationSuspended.set(true);
        if (newState == ReplicaState.DEAD) {
            //assume the dead replica ACK has been received for all pending jobs
            synchronized (jobCommitAcks) {
                for (Integer jobId : jobCommitAcks.keySet()) {
                    addAckToJob(jobId, replicaId);
                }
            }
        }
        //force replication threads to stop in order to change the replication factor
        suspendReplication(true);
    }
    replica.setState(newState);
    if (newState == ReplicaState.ACTIVE) {
        replicationFactor++;
    } else if (newState == ReplicaState.DEAD && replicationFactor > INITIAL_REPLICATION_FACTOR) {
        replicationFactor--;
    }
    LOGGER.log(Level.WARNING, "Replica " + replicaId + " state changed to: " + newState.name() + ". Replication factor changed to: " + replicationFactor);
    if (suspendReplication) {
        startReplicationThreads();
    }
}

Example 2 with Replica

use of org.apache.asterix.common.replication.Replica in project asterixdb by apache.

the class ReplicationManager method getActiveRemoteReplicasSockets.

private Map<String, SocketChannel> getActiveRemoteReplicasSockets() {
    Map<String, SocketChannel> replicaNodesSockets = new HashMap<>();
    for (Replica replica : replicas.values()) {
        if (replica.getState() == ReplicaState.ACTIVE) {
            try {
                SocketChannel sc = getReplicaSocket(replica.getId());
                replicaNodesSockets.put(replica.getId(), sc);
            } catch (IOException e) {
                if (LOGGER.isLoggable(Level.WARNING)) {
                    LOGGER.log(Level.WARNING, "Could not get replica socket", e);
                }
                reportFailedReplica(replica.getId());
            }
        }
    }
    return replicaNodesSockets;
}

Example 3 with Replica

use of org.apache.asterix.common.replication.Replica in project asterixdb by apache.

the class MetadataNodeFaultToleranceStrategy method notifyNodeFailure.

@Override
public synchronized void notifyNodeFailure(String nodeId) throws HyracksDataException {
    failedNodes.add(nodeId);
    hotStandbyMetadataReplica.remove(nodeId);
    clusterManager.updateNodePartitions(nodeId, false);
    if (nodeId.equals(metadataNodeId)) {
        clusterManager.updateMetadataNode(metadataNodeId, false);
    }
    clusterManager.refreshState();
    if (replicationStrategy.isParticipant(nodeId)) {
        // Notify impacted replica
        FaultToleranceUtil.notifyImpactedReplicas(nodeId, ClusterEventType.NODE_FAILURE, clusterManager, messageBroker, replicationStrategy);
    }
    // If the failed node is the metadata node, ask its replicas to replay any committed jobs
    if (nodeId.equals(metadataNodeId)) {
        ICcApplicationContext appCtx = (ICcApplicationContext) serviceCtx.getApplicationContext();
        int metadataPartitionId = appCtx.getMetadataProperties().getMetadataPartition().getPartitionId();
        Set<Integer> metadataPartition = new HashSet<>(Arrays.asList(metadataPartitionId));
        Set<Replica> activeRemoteReplicas = replicationStrategy.getRemoteReplicas(metadataNodeId).stream().filter(replica -> !failedNodes.contains(replica.getId())).collect(Collectors.toSet());
        //TODO Do election to identity the node with latest state
        for (Replica replica : activeRemoteReplicas) {
            ReplayPartitionLogsRequestMessage msg = new ReplayPartitionLogsRequestMessage(metadataPartition);
            try {
                messageBroker.sendApplicationMessageToNC(msg, replica.getId());
            } catch (Exception e) {
                LOGGER.log(Level.WARNING, "Failed sending an application message to an NC", e);
                continue;
            }
        }
    }
}

Example 4 with Replica

use of org.apache.asterix.common.replication.Replica in project asterixdb by apache.

the class ReplicationManager method getReplicaSocket.

/**
     * Establishes a connection with a remote replica.
     *
     * @param replicaId
     *            The replica to connect to.
     * @return The socket of the remote replica
     * @throws IOException
     */
private SocketChannel getReplicaSocket(String replicaId) throws IOException {
    Replica replica = replicationProperties.getReplicaById(replicaId);
    SocketChannel sc = SocketChannel.open();
    sc.configureBlocking(true);
    InetSocketAddress address = replica.getAddress(replicationProperties);
    sc.connect(new InetSocketAddress(address.getHostString(), address.getPort()));
    return sc;
}

Example 5 with Replica

use of org.apache.asterix.common.replication.Replica in project asterixdb by apache.

the class ReplicationManager method reportFailedReplica.

/**
     * Suspends replications and sends a remote replica failure event to ReplicasEventsMonitor.
     *
     * @param replicaId
     *            the failed replica id.
     */
public void reportFailedReplica(String replicaId) {
    Replica replica = replicas.get(replicaId);
    if (replica == null) {
        return;
    }
    if (replica.getState() == ReplicaState.DEAD) {
        return;
    }
    //need to stop processing any new logs or jobs
    terminateJobsReplication.set(true);
    ReplicaEvent event = new ReplicaEvent(replica, ClusterEventType.NODE_FAILURE);
    reportReplicaEvent(event);
}