Examples with ReplicaResourcesManager org.apache.asterix.replication.storage.ReplicaResourcesManager used on opensource projects

Example 1 with ReplicaResourcesManager

use of org.apache.asterix.replication.storage.ReplicaResourcesManager in project asterixdb by apache.

the class RemoteRecoveryManager method completeFailbackProcess.

@Override
public void completeFailbackProcess() throws IOException, InterruptedException {
    ILogManager logManager = runtimeContext.getTransactionSubsystem().getLogManager();
    ReplicaResourcesManager replicaResourcesManager = (ReplicaResourcesManager) runtimeContext.getReplicaResourcesManager();
    Map<String, ClusterPartition[]> nodePartitions = runtimeContext.getMetadataProperties().getNodePartitions();
    /**
         * for each lost partition, get the remaining files from replicas
         * to complete the failback process.
         */
    try {
        for (Entry<String, Set<String>> remoteReplica : failbackRecoveryReplicas.entrySet()) {
            String replicaId = remoteReplica.getKey();
            Set<String> NCsDataToRecover = remoteReplica.getValue();
            Set<String> existingFiles = new HashSet<>();
            Set<Integer> partitionsToRecover = new HashSet<>();
            for (String nodeId : NCsDataToRecover) {
                //get partitions that will be recovered from this node
                ClusterPartition[] replicaPartitions = nodePartitions.get(nodeId);
                for (ClusterPartition partition : replicaPartitions) {
                    existingFiles.addAll(replicaResourcesManager.getPartitionIndexesFiles(partition.getPartitionId(), true));
                    partitionsToRecover.add(partition.getPartitionId());
                }
            }
            //Request remaining indexes files
            replicationManager.requestReplicaFiles(replicaId, partitionsToRecover, existingFiles);
        }
    } catch (IOException e) {
        /**
             * in case of failure during failback completion process we need to construct a new plan
             * and get all the files from the start since the remote replicas will change in the new plan.
             */
        if (LOGGER.isLoggable(Level.WARNING)) {
            LOGGER.log(Level.WARNING, "Failed during completing failback. Restarting failback process...", e);
        }
        startFailbackProcess();
    }
    //get max LSN from selected remote replicas
    long maxRemoteLSN = replicationManager.getMaxRemoteLSN(failbackRecoveryReplicas.keySet());
    //6. force LogManager to start from a partition > maxLSN in selected remote replicas
    logManager.renewLogFilesAndStartFromLSN(maxRemoteLSN);
    //start replication service after failback completed
    runtimeContext.getReplicationChannel().start();
    runtimeContext.getReplicationManager().startReplicationThreads();
    failbackRecoveryReplicas = null;
}

Example 2 with ReplicaResourcesManager

use of org.apache.asterix.replication.storage.ReplicaResourcesManager in project asterixdb by apache.

the class NCAppRuntimeContext method initialize.

@Override
public void initialize(boolean initialRun) throws IOException, ACIDException {
    ioManager = getServiceContext().getIoManager();
    threadExecutor = new ThreadExecutor(getServiceContext().getThreadFactory());
    fileMapManager = new FileMapManager(ioManager);
    ICacheMemoryAllocator allocator = new HeapBufferAllocator();
    IPageCleanerPolicy pcp = new DelayPageCleanerPolicy(600000);
    IPageReplacementStrategy prs = new ClockPageReplacementStrategy(allocator, storageProperties.getBufferCachePageSize(), storageProperties.getBufferCacheNumPages());
    AsynchronousScheduler.INSTANCE.init(getServiceContext().getThreadFactory());
    lsmIOScheduler = AsynchronousScheduler.INSTANCE;
    metadataMergePolicyFactory = new PrefixMergePolicyFactory();
    ILocalResourceRepositoryFactory persistentLocalResourceRepositoryFactory = new PersistentLocalResourceRepositoryFactory(ioManager, getServiceContext().getNodeId(), metadataProperties);
    localResourceRepository = (PersistentLocalResourceRepository) persistentLocalResourceRepositoryFactory.createRepository();
    IAppRuntimeContextProvider asterixAppRuntimeContextProvider = new AppRuntimeContextProviderForRecovery(this);
    txnSubsystem = new TransactionSubsystem(getServiceContext(), getServiceContext().getNodeId(), asterixAppRuntimeContextProvider, txnProperties);
    IRecoveryManager recoveryMgr = txnSubsystem.getRecoveryManager();
    SystemState systemState = recoveryMgr.getSystemState();
    if (initialRun || systemState == SystemState.PERMANENT_DATA_LOSS) {
        //delete any storage data before the resource factory is initialized
        localResourceRepository.deleteStorageData(true);
    }
    datasetLifecycleManager = new DatasetLifecycleManager(storageProperties, localResourceRepository, MetadataIndexImmutableProperties.FIRST_AVAILABLE_USER_DATASET_ID, txnSubsystem.getLogManager(), ioManager.getIODevices().size());
    isShuttingdown = false;
    activeManager = new ActiveManager(threadExecutor, getServiceContext().getNodeId(), activeProperties.getMemoryComponentGlobalBudget(), compilerProperties.getFrameSize());
    if (replicationProperties.isParticipant(getServiceContext().getNodeId())) {
        String nodeId = getServiceContext().getNodeId();
        replicaResourcesManager = new ReplicaResourcesManager(localResourceRepository, metadataProperties);
        replicationManager = new ReplicationManager(nodeId, replicationProperties, replicaResourcesManager, txnSubsystem.getLogManager(), asterixAppRuntimeContextProvider);
        //pass replication manager to replication required object
        //LogManager to replicate logs
        txnSubsystem.getLogManager().setReplicationManager(replicationManager);
        //PersistentLocalResourceRepository to replicate metadata files and delete backups on drop index
        localResourceRepository.setReplicationManager(replicationManager);
        /*
             * add the partitions that will be replicated in this node as inactive partitions
             */
        //get nodes which replicate to this node
        Set<String> remotePrimaryReplicas = replicationProperties.getRemotePrimaryReplicasIds(nodeId);
        for (String clientId : remotePrimaryReplicas) {
            //get the partitions of each client
            ClusterPartition[] clientPartitions = metadataProperties.getNodePartitions().get(clientId);
            for (ClusterPartition partition : clientPartitions) {
                localResourceRepository.addInactivePartition(partition.getPartitionId());
            }
        }
        //initialize replication channel
        replicationChannel = new ReplicationChannel(nodeId, replicationProperties, txnSubsystem.getLogManager(), replicaResourcesManager, replicationManager, getServiceContext(), asterixAppRuntimeContextProvider);
        remoteRecoveryManager = new RemoteRecoveryManager(replicationManager, this, replicationProperties);
        bufferCache = new BufferCache(ioManager, prs, pcp, fileMapManager, storageProperties.getBufferCacheMaxOpenFiles(), getServiceContext().getThreadFactory(), replicationManager);
    } else {
        bufferCache = new BufferCache(ioManager, prs, pcp, fileMapManager, storageProperties.getBufferCacheMaxOpenFiles(), getServiceContext().getThreadFactory());
    }
    /*
         * The order of registration is important. The buffer cache must registered before recovery and transaction
         * managers. Notes: registered components are stopped in reversed order
         */
    ILifeCycleComponentManager lccm = getServiceContext().getLifeCycleComponentManager();
    lccm.register((ILifeCycleComponent) bufferCache);
    /*
         * LogManager must be stopped after RecoveryManager, DatasetLifeCycleManager, and ReplicationManager
         * to process any logs that might be generated during stopping these components
         */
    lccm.register((ILifeCycleComponent) txnSubsystem.getLogManager());
    /*
         * ReplicationManager must be stopped after indexLifecycleManager and recovery manager
         * so that any logs/files generated during closing datasets or checkpoints are sent to remote replicas
         */
    if (replicationManager != null) {
        lccm.register(replicationManager);
    }
    lccm.register((ILifeCycleComponent) txnSubsystem.getRecoveryManager());
    /*
         * Stopping indexLifecycleManager will flush and close all datasets.
         */
    lccm.register((ILifeCycleComponent) datasetLifecycleManager);
    lccm.register((ILifeCycleComponent) txnSubsystem.getTransactionManager());
    lccm.register((ILifeCycleComponent) txnSubsystem.getLockManager());
    lccm.register(txnSubsystem.getCheckpointManager());
}