Examples with ClusterPartition org.apache.asterix.common.cluster.ClusterPartition used on opensource projects

Example 1 with ClusterPartition

use of org.apache.asterix.common.cluster.ClusterPartition in project asterixdb by apache.

the class MetadataBootstrap method enlistMetadataDataset.

/**
     * Enlist a metadata index so it is available for metadata operations
     * should be performed upon bootstrapping
     *
     * @param index
     * @throws HyracksDataException
     */
public static void enlistMetadataDataset(INCServiceContext ncServiceCtx, IMetadataIndex index) throws HyracksDataException {
    ClusterPartition metadataPartition = appContext.getMetadataProperties().getMetadataPartition();
    int metadataDeviceId = metadataPartition.getIODeviceNum();
    String metadataPartitionPath = StoragePathUtil.prepareStoragePartitionPath(ClusterProperties.INSTANCE.getStorageDirectoryName(), metadataPartition.getPartitionId());
    String resourceName = metadataPartitionPath + File.separator + index.getFileNameRelativePath();
    FileReference file = ioManager.getFileReference(metadataDeviceId, resourceName);
    index.setFile(file);
    ITypeTraits[] typeTraits = index.getTypeTraits();
    IBinaryComparatorFactory[] cmpFactories = index.getKeyBinaryComparatorFactory();
    int[] bloomFilterKeyFields = index.getBloomFilterKeyFields();
    // opTrackerProvider and ioOpCallbackFactory should both be acquired through IStorageManager
    // We are unable to do this since IStorageManager needs a dataset to determine the appropriate
    // objects
    ILSMOperationTrackerFactory opTrackerFactory = index.isPrimaryIndex() ? new PrimaryIndexOperationTrackerFactory(index.getDatasetId().getId()) : new SecondaryIndexOperationTrackerFactory(index.getDatasetId().getId());
    ILSMIOOperationCallbackFactory ioOpCallbackFactory = LSMBTreeIOOperationCallbackFactory.INSTANCE;
    IStorageComponentProvider storageComponentProvider = appContext.getStorageComponentProvider();
    if (isNewUniverse()) {
        LSMBTreeLocalResourceFactory lsmBtreeFactory = new LSMBTreeLocalResourceFactory(storageComponentProvider.getStorageManager(), typeTraits, cmpFactories, null, null, null, opTrackerFactory, ioOpCallbackFactory, storageComponentProvider.getMetadataPageManagerFactory(), new AsterixVirtualBufferCacheProvider(index.getDatasetId().getId()), storageComponentProvider.getIoOperationSchedulerProvider(), appContext.getMetadataMergePolicyFactory(), GlobalConfig.DEFAULT_COMPACTION_POLICY_PROPERTIES, true, bloomFilterKeyFields, appContext.getBloomFilterFalsePositiveRate(), true, null);
        DatasetLocalResourceFactory dsLocalResourceFactory = new DatasetLocalResourceFactory(index.getDatasetId().getId(), lsmBtreeFactory);
        // TODO(amoudi) Creating the index should be done through the same code path as other indexes
        // This is to be done by having a metadata dataset associated with each index
        IIndexBuilder indexBuilder = new IndexBuilder(ncServiceCtx, storageComponentProvider.getStorageManager(), index::getResourceId, file, dsLocalResourceFactory, true);
        indexBuilder.build();
    } else {
        final LocalResource resource = localResourceRepository.get(file.getRelativePath());
        if (resource == null) {
            throw new HyracksDataException("Could not find required metadata indexes. Please delete " + appContext.getMetadataProperties().getTransactionLogDirs().get(appContext.getTransactionSubsystem().getId()) + " to intialize as a new instance. (WARNING: all data will be lost.)");
        }
        // Why do we care about metadata dataset's resource ids? why not assign them ids similar to other resources?
        if (index.getResourceId() != resource.getId()) {
            throw new HyracksDataException("Resource Id doesn't match expected metadata index resource id");
        }
        IndexDataflowHelper indexHelper = new IndexDataflowHelper(ncServiceCtx, storageComponentProvider.getStorageManager(), file);
        // Opening the index through the helper will ensure it gets instantiated
        indexHelper.open();
        indexHelper.close();
    }
}

Example 2 with ClusterPartition

use of org.apache.asterix.common.cluster.ClusterPartition in project asterixdb by apache.

the class ClusterStateManager method resetClusterPartitionConstraint.

private synchronized void resetClusterPartitionConstraint() {
    ArrayList<String> clusterActiveLocations = new ArrayList<>();
    for (ClusterPartition p : clusterPartitions.values()) {
        if (p.isActive()) {
            clusterActiveLocations.add(p.getActiveNodeId());
        }
    }
    clusterPartitionConstraint = new AlgebricksAbsolutePartitionConstraint(clusterActiveLocations.toArray(new String[] {}));
}

Example 3 with ClusterPartition

use of org.apache.asterix.common.cluster.ClusterPartition in project asterixdb by apache.

the class PropertiesAccessor method configureNc.

private void configureNc(ConfigManager configManager, String ncId, MutableInt uniquePartitionId) throws AsterixException {
    // Now we assign the coredump and txnlog directories for this node.
    // QQQ Default values? Should they be specified here? Or should there
    // be a default.ini? Certainly wherever they are, they should be platform-dependent.
    IApplicationConfig nodeCfg = cfg.getNCEffectiveConfig(ncId);
    coredumpConfig.put(ncId, nodeCfg.getString(NodeProperties.Option.CORE_DUMP_DIR));
    transactionLogDirs.put(ncId, nodeCfg.getString(NodeProperties.Option.TXN_LOG_DIR));
    int partitionId = nodeCfg.getInt(NodeProperties.Option.STARTING_PARTITION_ID);
    if (partitionId != -1) {
        uniquePartitionId.setValue(partitionId);
    } else {
        configManager.set(ncId, NodeProperties.Option.STARTING_PARTITION_ID, uniquePartitionId.getValue());
    }
    // Now we create an array of ClusterPartitions for all the partitions
    // on this NC.
    String[] iodevices = nodeCfg.getStringArray(IODEVICES);
    String storageSubdir = nodeCfg.getString(STORAGE_SUBDIR);
    String[] nodeStores = new String[iodevices.length];
    ClusterPartition[] nodePartitions = new ClusterPartition[iodevices.length];
    for (int i = 0; i < nodePartitions.length; i++) {
        // Construct final storage path from iodevice dir + storage subdirs
        nodeStores[i] = iodevices[i] + File.separator + storageSubdir;
        // Create ClusterPartition instances for this NC.
        ClusterPartition partition = new ClusterPartition(uniquePartitionId.getAndIncrement(), ncId, i);
        ClusterPartition orig = clusterPartitions.put(partition.getPartitionId(), partition);
        if (orig != null) {
            throw AsterixException.create(ErrorCode.DUPLICATE_PARTITION_ID, partition.getPartitionId(), ncId, orig.getNodeId());
        }
        nodePartitions[i] = partition;
    }
    stores.put(ncId, nodeStores);
    nodePartitionsMap.put(ncId, nodePartitions);
}

Example 4 with ClusterPartition

use of org.apache.asterix.common.cluster.ClusterPartition 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 5 with ClusterPartition

use of org.apache.asterix.common.cluster.ClusterPartition in project asterixdb by apache.

the class RemoteRecoveryManager method startFailbackProcess.

@Override
public void startFailbackProcess() {
    int maxRecoveryAttempts = replicationProperties.getMaxRemoteRecoveryAttempts();
    PersistentLocalResourceRepository resourceRepository = (PersistentLocalResourceRepository) runtimeContext.getLocalResourceRepository();
    IDatasetLifecycleManager datasetLifeCycleManager = runtimeContext.getDatasetLifecycleManager();
    Map<String, ClusterPartition[]> nodePartitions = runtimeContext.getMetadataProperties().getNodePartitions();
    while (true) {
        //start recovery steps
        try {
            if (maxRecoveryAttempts <= 0) {
                //to avoid infinite loop in case of unexpected behavior.
                throw new IllegalStateException("Failed to perform remote recovery.");
            }
            /*** Prepare for Recovery ***/
            //1. check remote replicas states
            replicationManager.initializeReplicasState();
            int activeReplicasCount = replicationManager.getActiveReplicasCount();
            if (activeReplicasCount == 0) {
                throw new IllegalStateException("no ACTIVE remote replica(s) exists to perform remote recovery");
            }
            //2. clean any memory data that could've existed from previous failed recovery attempt
            datasetLifeCycleManager.closeAllDatasets();
            //3. remove any existing storage data and initialize storage metadata
            resourceRepository.deleteStorageData(true);
            resourceRepository.initializeNewUniverse(ClusterProperties.INSTANCE.getStorageDirectoryName());
            //4. select remote replicas to recover from per lost replica data
            failbackRecoveryReplicas = constructRemoteRecoveryPlan();
            /*** Start Recovery Per Lost Replica ***/
            for (Entry<String, Set<String>> remoteReplica : failbackRecoveryReplicas.entrySet()) {
                String replicaId = remoteReplica.getKey();
                Set<String> ncsToRecoverFor = remoteReplica.getValue();
                Set<Integer> partitionsIds = new HashSet<>();
                for (String node : ncsToRecoverFor) {
                    partitionsIds.addAll((Arrays.asList(nodePartitions.get(node))).stream().map(ClusterPartition::getPartitionId).collect(Collectors.toList()));
                }
                //1. Request indexes metadata and LSM components
                replicationManager.requestReplicaFiles(replicaId, partitionsIds, new HashSet<String>());
            }
            break;
        } catch (IOException e) {
            if (LOGGER.isLoggable(Level.WARNING)) {
                LOGGER.log(Level.WARNING, "Failed during remote recovery. Attempting again...", e);
            }
            maxRecoveryAttempts--;
        }
    }
}