Search in sources :

Example 1 with FileChannelManager

use of org.apache.flink.runtime.io.disk.FileChannelManager in project flink by apache.

the class NettyShuffleServiceFactory method createNettyShuffleEnvironment.

@VisibleForTesting
static NettyShuffleEnvironment createNettyShuffleEnvironment(NettyShuffleEnvironmentConfiguration config, ResourceID taskExecutorResourceId, TaskEventPublisher taskEventPublisher, ResultPartitionManager resultPartitionManager, MetricGroup metricGroup, Executor ioExecutor) {
    checkNotNull(config);
    checkNotNull(taskExecutorResourceId);
    checkNotNull(taskEventPublisher);
    checkNotNull(resultPartitionManager);
    checkNotNull(metricGroup);
    NettyConfig nettyConfig = config.nettyConfig();
    FileChannelManager fileChannelManager = new FileChannelManagerImpl(config.getTempDirs(), DIR_NAME_PREFIX);
    if (LOG.isInfoEnabled()) {
        LOG.info("Created a new {} for storing result partitions of BLOCKING shuffles. Used directories:\n\t{}", FileChannelManager.class.getSimpleName(), Arrays.stream(fileChannelManager.getPaths()).map(File::getAbsolutePath).collect(Collectors.joining("\n\t")));
    }
    ConnectionManager connectionManager = nettyConfig != null ? new NettyConnectionManager(resultPartitionManager, taskEventPublisher, nettyConfig, config.getMaxNumberOfConnections(), config.isConnectionReuseEnabled()) : new LocalConnectionManager();
    NetworkBufferPool networkBufferPool = new NetworkBufferPool(config.numNetworkBuffers(), config.networkBufferSize(), config.getRequestSegmentsTimeout());
    // we create a separated buffer pool here for batch shuffle instead of reusing the network
    // buffer pool directly to avoid potential side effects of memory contention, for example,
    // dead lock or "insufficient network buffer" error
    BatchShuffleReadBufferPool batchShuffleReadBufferPool = new BatchShuffleReadBufferPool(config.batchShuffleReadMemoryBytes(), config.networkBufferSize());
    // we create a separated IO executor pool here for batch shuffle instead of reusing the
    // TaskManager IO executor pool directly to avoid the potential side effects of execution
    // contention, for example, too long IO or waiting time leading to starvation or timeout
    ExecutorService batchShuffleReadIOExecutor = Executors.newFixedThreadPool(Math.max(1, Math.min(batchShuffleReadBufferPool.getMaxConcurrentRequests(), 4 * Hardware.getNumberCPUCores())), new ExecutorThreadFactory("blocking-shuffle-io"));
    registerShuffleMetrics(metricGroup, networkBufferPool);
    ResultPartitionFactory resultPartitionFactory = new ResultPartitionFactory(resultPartitionManager, fileChannelManager, networkBufferPool, batchShuffleReadBufferPool, batchShuffleReadIOExecutor, config.getBlockingSubpartitionType(), config.networkBuffersPerChannel(), config.floatingNetworkBuffersPerGate(), config.networkBufferSize(), config.isBlockingShuffleCompressionEnabled(), config.getCompressionCodec(), config.getMaxBuffersPerChannel(), config.sortShuffleMinBuffers(), config.sortShuffleMinParallelism(), config.isSSLEnabled());
    SingleInputGateFactory singleInputGateFactory = new SingleInputGateFactory(taskExecutorResourceId, config, connectionManager, resultPartitionManager, taskEventPublisher, networkBufferPool);
    return new NettyShuffleEnvironment(taskExecutorResourceId, config, networkBufferPool, connectionManager, resultPartitionManager, fileChannelManager, resultPartitionFactory, singleInputGateFactory, ioExecutor, batchShuffleReadBufferPool, batchShuffleReadIOExecutor);
}
Also used : BatchShuffleReadBufferPool(org.apache.flink.runtime.io.disk.BatchShuffleReadBufferPool) NettyConfig(org.apache.flink.runtime.io.network.netty.NettyConfig) NetworkBufferPool(org.apache.flink.runtime.io.network.buffer.NetworkBufferPool) ExecutorThreadFactory(org.apache.flink.util.concurrent.ExecutorThreadFactory) FileChannelManager(org.apache.flink.runtime.io.disk.FileChannelManager) FileChannelManagerImpl(org.apache.flink.runtime.io.disk.FileChannelManagerImpl) NettyConnectionManager(org.apache.flink.runtime.io.network.netty.NettyConnectionManager) ResultPartitionFactory(org.apache.flink.runtime.io.network.partition.ResultPartitionFactory) ExecutorService(java.util.concurrent.ExecutorService) SingleInputGateFactory(org.apache.flink.runtime.io.network.partition.consumer.SingleInputGateFactory) NettyConnectionManager(org.apache.flink.runtime.io.network.netty.NettyConnectionManager) File(java.io.File) VisibleForTesting(org.apache.flink.annotation.VisibleForTesting)

Aggregations

File (java.io.File)1 ExecutorService (java.util.concurrent.ExecutorService)1 VisibleForTesting (org.apache.flink.annotation.VisibleForTesting)1 BatchShuffleReadBufferPool (org.apache.flink.runtime.io.disk.BatchShuffleReadBufferPool)1 FileChannelManager (org.apache.flink.runtime.io.disk.FileChannelManager)1 FileChannelManagerImpl (org.apache.flink.runtime.io.disk.FileChannelManagerImpl)1 NetworkBufferPool (org.apache.flink.runtime.io.network.buffer.NetworkBufferPool)1 NettyConfig (org.apache.flink.runtime.io.network.netty.NettyConfig)1 NettyConnectionManager (org.apache.flink.runtime.io.network.netty.NettyConnectionManager)1 ResultPartitionFactory (org.apache.flink.runtime.io.network.partition.ResultPartitionFactory)1 SingleInputGateFactory (org.apache.flink.runtime.io.network.partition.consumer.SingleInputGateFactory)1 ExecutorThreadFactory (org.apache.flink.util.concurrent.ExecutorThreadFactory)1