聊聊flink的NetworkBufferPool

本文主要研究一下flink的NetworkBufferPool

BufferPoolFactory

flink-release-1.7.2/flink-runtime/src/main/java/org/apache/flink/runtime/io/network/buffer/BufferPoolFactory.java

public interface BufferPoolFactory {

    /**
     * Tries to create a buffer pool, which is guaranteed to provide at least the number of required
     * buffers.
     *
     * 

The buffer pool is of dynamic size with at least numRequiredBuffers buffers. * * @param numRequiredBuffers * minimum number of network buffers in this pool * @param maxUsedBuffers * maximum number of network buffers this pool offers */ BufferPool createBufferPool(int numRequiredBuffers, int maxUsedBuffers) throws IOException; /** * Tries to create a buffer pool with an optional owner, which is guaranteed to provide at least the * number of required buffers. * *

The buffer pool is of dynamic size with at least numRequiredBuffers buffers. * * @param numRequiredBuffers * minimum number of network buffers in this pool * @param maxUsedBuffers * maximum number of network buffers this pool offers * @param owner * the optional owner of this buffer pool to release memory when needed */ BufferPool createBufferPool(int numRequiredBuffers, int maxUsedBuffers, Optional owner) throws IOException; /** * Destroy callback for updating factory book keeping. */ void destroyBufferPool(BufferPool bufferPool) throws IOException; }

  • BufferPoolFactory定义了createBufferPool、destroyBufferPool方法;其中createBufferPool支持numRequiredBuffers、maxUsedBuffers、owner参数

NetworkBufferPool

flink-release-1.7.2/flink-runtime/src/main/java/org/apache/flink/runtime/io/network/buffer/NetworkBufferPool.java

public class NetworkBufferPool implements BufferPoolFactory {
    //......

    private static final Logger LOG = LoggerFactory.getLogger(NetworkBufferPool.class);

    private final int totalNumberOfMemorySegments;

    private final int memorySegmentSize;

    private final ArrayBlockingQueue availableMemorySegments;

    private volatile boolean isDestroyed;

    // ---- Managed buffer pools ----------------------------------------------

    private final Object factoryLock = new Object();

    private final Set allBufferPools = new HashSet<>();

    private int numTotalRequiredBuffers;

    /**
     * Allocates all {@link MemorySegment} instances managed by this pool.
     */
    public NetworkBufferPool(int numberOfSegmentsToAllocate, int segmentSize) {

        this.totalNumberOfMemorySegments = numberOfSegmentsToAllocate;
        this.memorySegmentSize = segmentSize;

        final long sizeInLong = (long) segmentSize;

        try {
            this.availableMemorySegments = new ArrayBlockingQueue<>(numberOfSegmentsToAllocate);
        }
        catch (OutOfMemoryError err) {
            throw new OutOfMemoryError("Could not allocate buffer queue of length "
                    + numberOfSegmentsToAllocate + " - " + err.getMessage());
        }

        try {
            for (int i = 0; i < numberOfSegmentsToAllocate; i++) {
                availableMemorySegments.add(MemorySegmentFactory.allocateUnpooledOffHeapMemory(segmentSize, null));
            }
        }
        catch (OutOfMemoryError err) {
            int allocated = availableMemorySegments.size();

            // free some memory
            availableMemorySegments.clear();

            long requiredMb = (sizeInLong * numberOfSegmentsToAllocate) >> 20;
            long allocatedMb = (sizeInLong * allocated) >> 20;
            long missingMb = requiredMb - allocatedMb;

            throw new OutOfMemoryError("Could not allocate enough memory segments for NetworkBufferPool " +
                    "(required (Mb): " + requiredMb +
                    ", allocated (Mb): " + allocatedMb +
                    ", missing (Mb): " + missingMb + "). Cause: " + err.getMessage());
        }

        long allocatedMb = (sizeInLong * availableMemorySegments.size()) >> 20;

        LOG.info("Allocated {} MB for network buffer pool (number of memory segments: {}, bytes per segment: {}).",
                allocatedMb, availableMemorySegments.size(), segmentSize);
    }

    @Override
    public BufferPool createBufferPool(int numRequiredBuffers, int maxUsedBuffers) throws IOException {
        return createBufferPool(numRequiredBuffers, maxUsedBuffers, Optional.empty());
    }

    @Override
    public BufferPool createBufferPool(int numRequiredBuffers, int maxUsedBuffers, Optional owner) throws IOException {
        // It is necessary to use a separate lock from the one used for buffer
        // requests to ensure deadlock freedom for failure cases.
        synchronized (factoryLock) {
            if (isDestroyed) {
                throw new IllegalStateException("Network buffer pool has already been destroyed.");
            }

            // Ensure that the number of required buffers can be satisfied.
            // With dynamic memory management this should become obsolete.
            if (numTotalRequiredBuffers + numRequiredBuffers > totalNumberOfMemorySegments) {
                throw new IOException(String.format("Insufficient number of network buffers: " +
                                "required %d, but only %d available. The total number of network " +
                                "buffers is currently set to %d of %d bytes each. You can increase this " +
                                "number by setting the configuration keys '%s', '%s', and '%s'.",
                        numRequiredBuffers,
                        totalNumberOfMemorySegments - numTotalRequiredBuffers,
                        totalNumberOfMemorySegments,
                        memorySegmentSize,
                        TaskManagerOptions.NETWORK_BUFFERS_MEMORY_FRACTION.key(),
                        TaskManagerOptions.NETWORK_BUFFERS_MEMORY_MIN.key(),
                        TaskManagerOptions.NETWORK_BUFFERS_MEMORY_MAX.key()));
            }

            this.numTotalRequiredBuffers += numRequiredBuffers;

            // We are good to go, create a new buffer pool and redistribute
            // non-fixed size buffers.
            LocalBufferPool localBufferPool =
                new LocalBufferPool(this, numRequiredBuffers, maxUsedBuffers, owner);

            allBufferPools.add(localBufferPool);

            try {
                redistributeBuffers();
            } catch (IOException e) {
                try {
                    destroyBufferPool(localBufferPool);
                } catch (IOException inner) {
                    e.addSuppressed(inner);
                }
                ExceptionUtils.rethrowIOException(e);
            }

            return localBufferPool;
        }
    }

    @Override
    public void destroyBufferPool(BufferPool bufferPool) throws IOException {
        if (!(bufferPool instanceof LocalBufferPool)) {
            throw new IllegalArgumentException("bufferPool is no LocalBufferPool");
        }

        synchronized (factoryLock) {
            if (allBufferPools.remove(bufferPool)) {
                numTotalRequiredBuffers -= bufferPool.getNumberOfRequiredMemorySegments();

                redistributeBuffers();
            }
        }
    }

    //......
}
  • NetworkBufferPool实现了BufferPoolFactory接口,它的构造器接收numberOfSegmentsToAllocate、segmentSize两个参数;构造器里头根据numberOfSegmentsToAllocate创建了availableMemorySegments这个ArrayBlockingQueue,然后通过MemorySegmentFactory.allocateUnpooledOffHeapMemory挨个创建MemorySegment添加到availableMemorySegments
  • createBufferPool方法创建的是LocalBufferPool(传递了自身NetworkBufferPool实例进去),然后添加到allBufferPools这个set中,同时增加numTotalRequiredBuffers;destroyBufferPool方法则从allBufferPools移除该bufferPool,同时减少numTotalRequiredBuffers
  • createBufferPool方法及destroyBufferPool方法会调用到redistributeBuffers方法,通过调用LocalBufferPool的setNumBuffers方法来调整buffer pool的大小

LocalBufferPool

flink-release-1.7.2/flink-runtime/src/main/java/org/apache/flink/runtime/io/network/buffer/LocalBufferPool.java

class LocalBufferPool implements BufferPool {
    private static final Logger LOG = LoggerFactory.getLogger(LocalBufferPool.class);

    /** Global network buffer pool to get buffers from. */
    private final NetworkBufferPool networkBufferPool;

    /** The minimum number of required segments for this pool. */
    private final int numberOfRequiredMemorySegments;

    /**
     * The currently available memory segments. These are segments, which have been requested from
     * the network buffer pool and are currently not handed out as Buffer instances.
     *
     * 

BEWARE: Take special care with the interactions between this lock and * locks acquired before entering this class vs. locks being acquired during calls to external * code inside this class, e.g. with * {@link org.apache.flink.runtime.io.network.partition.consumer.RemoteInputChannel#bufferQueue} * via the {@link #registeredListeners} callback. */ private final ArrayDeque availableMemorySegments = new ArrayDeque(); /** * Buffer availability listeners, which need to be notified when a Buffer becomes available. * Listeners can only be registered at a time/state where no Buffer instance was available. */ private final ArrayDeque registeredListeners = new ArrayDeque<>(); /** Maximum number of network buffers to allocate. */ private final int maxNumberOfMemorySegments; /** The current size of this pool. */ private int currentPoolSize; /** * Number of all memory segments, which have been requested from the network buffer pool and are * somehow referenced through this pool (e.g. wrapped in Buffer instances or as available segments). */ private int numberOfRequestedMemorySegments; private boolean isDestroyed; private final Optional owner; /** * Local buffer pool based on the given networkBufferPool with a minimal number of * network buffers being available. * * @param networkBufferPool * global network buffer pool to get buffers from * @param numberOfRequiredMemorySegments * minimum number of network buffers */ LocalBufferPool(NetworkBufferPool networkBufferPool, int numberOfRequiredMemorySegments) { this(networkBufferPool, numberOfRequiredMemorySegments, Integer.MAX_VALUE, Optional.empty()); } /** * Local buffer pool based on the given networkBufferPool with a minimal and maximal * number of network buffers being available. * * @param networkBufferPool * global network buffer pool to get buffers from * @param numberOfRequiredMemorySegments * minimum number of network buffers * @param maxNumberOfMemorySegments * maximum number of network buffers to allocate */ LocalBufferPool(NetworkBufferPool networkBufferPool, int numberOfRequiredMemorySegments, int maxNumberOfMemorySegments) { this(networkBufferPool, numberOfRequiredMemorySegments, maxNumberOfMemorySegments, Optional.empty()); } /** * Local buffer pool based on the given networkBufferPool and bufferPoolOwner * with a minimal and maximal number of network buffers being available. * * @param networkBufferPool * global network buffer pool to get buffers from * @param numberOfRequiredMemorySegments * minimum number of network buffers * @param maxNumberOfMemorySegments * maximum number of network buffers to allocate * @param owner * the optional owner of this buffer pool to release memory when needed */ LocalBufferPool( NetworkBufferPool networkBufferPool, int numberOfRequiredMemorySegments, int maxNumberOfMemorySegments, Optional owner) { checkArgument(maxNumberOfMemorySegments >= numberOfRequiredMemorySegments, "Maximum number of memory segments (%s) should not be smaller than minimum (%s).", maxNumberOfMemorySegments, numberOfRequiredMemorySegments); checkArgument(maxNumberOfMemorySegments > 0, "Maximum number of memory segments (%s) should be larger than 0.", maxNumberOfMemorySegments); LOG.debug("Using a local buffer pool with {}-{} buffers", numberOfRequiredMemorySegments, maxNumberOfMemorySegments); this.networkBufferPool = networkBufferPool; this.numberOfRequiredMemorySegments = numberOfRequiredMemorySegments; this.currentPoolSize = numberOfRequiredMemorySegments; this.maxNumberOfMemorySegments = maxNumberOfMemorySegments; this.owner = owner; } //...... @Override public void recycle(MemorySegment segment) { BufferListener listener; NotificationResult notificationResult = NotificationResult.BUFFER_NOT_USED; while (!notificationResult.isBufferUsed()) { synchronized (availableMemorySegments) { if (isDestroyed || numberOfRequestedMemorySegments > currentPoolSize) { returnMemorySegment(segment); return; } else { listener = registeredListeners.poll(); if (listener == null) { availableMemorySegments.add(segment); availableMemorySegments.notify(); return; } } } notificationResult = fireBufferAvailableNotification(listener, segment); } } private MemorySegment requestMemorySegment(boolean isBlocking) throws InterruptedException, IOException { synchronized (availableMemorySegments) { returnExcessMemorySegments(); boolean askToRecycle = owner.isPresent(); // fill availableMemorySegments with at least one element, wait if required while (availableMemorySegments.isEmpty()) { if (isDestroyed) { throw new IllegalStateException("Buffer pool is destroyed."); } if (numberOfRequestedMemorySegments < currentPoolSize) { final MemorySegment segment = networkBufferPool.requestMemorySegment(); if (segment != null) { numberOfRequestedMemorySegments++; return segment; } } if (askToRecycle) { owner.get().releaseMemory(1); } if (isBlocking) { availableMemorySegments.wait(2000); } else { return null; } } return availableMemorySegments.poll(); } } //...... }

  • LocalBufferPool的构造器要求传入NetworkBufferPool,而其内部的requestMemorySegment方法,在availableMemorySegments为空且numberOfRequestedMemorySegments < currentPoolSize时,会调用networkBufferPool.requestMemorySegment()来申请MemorySegment;而recycle方法在numberOfRequestedMemorySegments > currentPoolSize时会归还MemorySegment到networkBufferPool,否则在BufferListener为null的时候会归还到availableMemorySegments

小结

  • BufferPoolFactory定义了createBufferPool、destroyBufferPool方法;其中createBufferPool支持numRequiredBuffers、maxUsedBuffers、owner参数;NetworkBufferPool实现了BufferPoolFactory接口,它的构造器接收numberOfSegmentsToAllocate、segmentSize两个参数;构造器里头根据numberOfSegmentsToAllocate创建了availableMemorySegments这个ArrayBlockingQueue,然后通过MemorySegmentFactory.allocateUnpooledOffHeapMemory挨个创建MemorySegment添加到availableMemorySegments
  • NetworkBufferPool的createBufferPool方法创建的是LocalBufferPool(传递了自身NetworkBufferPool实例进去),然后添加到allBufferPools这个set中,同时增加numTotalRequiredBuffers;destroyBufferPool方法则从allBufferPools移除该bufferPool,同时减少numTotalRequiredBuffers;createBufferPool方法及destroyBufferPool方法会调用到redistributeBuffers方法,通过调用LocalBufferPool的setNumBuffers方法来调整buffer pool的大小
  • LocalBufferPool的构造器要求传入NetworkBufferPool,而其内部的requestMemorySegment方法,在availableMemorySegments为空且numberOfRequestedMemorySegments < currentPoolSize时,会调用networkBufferPool.requestMemorySegment()来申请MemorySegment;而recycle方法在numberOfRequestedMemorySegments > currentPoolSize时会归还MemorySegment到networkBufferPool,否则在BufferListener为null的时候会归还到availableMemorySegments

doc

  • NetworkBufferPool

你可能感兴趣的:(聊聊flink的NetworkBufferPool)