Search in sources :

Example 1 with FileChannelInputView

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

the class LargeRecordHandler method finishWriteAndSortKeys.

public MutableObjectIterator<T> finishWriteAndSortKeys(List<MemorySegment> memory) throws IOException {
    if (recordsOutFile == null || keysOutFile == null) {
        throw new IllegalStateException("The LargeRecordHandler has not spilled any records");
    }
    // close the writers and
    final int lastBlockBytesKeys;
    final int lastBlockBytesRecords;
    recordsOutFile.close();
    keysOutFile.close();
    lastBlockBytesKeys = keysOutFile.getBytesInLatestSegment();
    lastBlockBytesRecords = recordsOutFile.getBytesInLatestSegment();
    recordsOutFile = null;
    keysOutFile = null;
    final int pagesForReaders = Math.max(3 * MIN_SEGMENTS_FOR_KEY_SPILLING, Math.min(2 * MAX_SEGMENTS_FOR_KEY_SPILLING, memory.size() / 50));
    final int pagesForKeyReader = Math.min(pagesForReaders - MIN_SEGMENTS_FOR_KEY_SPILLING, MAX_SEGMENTS_FOR_KEY_SPILLING);
    final int pagesForRecordReader = pagesForReaders - pagesForKeyReader;
    // grab memory for the record reader
    ArrayList<MemorySegment> memForRecordReader = new ArrayList<MemorySegment>();
    ArrayList<MemorySegment> memForKeysReader = new ArrayList<MemorySegment>();
    for (int i = 0; i < pagesForRecordReader; i++) {
        memForRecordReader.add(memory.remove(memory.size() - 1));
    }
    for (int i = 0; i < pagesForKeyReader; i++) {
        memForKeysReader.add(memory.remove(memory.size() - 1));
    }
    keysReader = new FileChannelInputView(ioManager.createBlockChannelReader(keysChannel), memManager, memForKeysReader, lastBlockBytesKeys);
    InputViewIterator<Tuple> keyIterator = new InputViewIterator<Tuple>(keysReader, keySerializer);
    try {
        keySorter = ExternalSorter.newBuilder(memManager, memoryOwner, keySerializer, keyComparator, executionConfig).maxNumFileHandles(maxFilehandles).sortBuffers(1).enableSpilling(ioManager, 1.0f).memory(memory).objectReuse(this.executionConfig.isObjectReuseEnabled()).largeRecords(false).build(keyIterator);
    } catch (MemoryAllocationException e) {
        throw new IllegalStateException("We should not try allocating memory. Instead the sorter should use the provided memory.", e);
    }
    // wait for the sorter to sort the keys
    MutableObjectIterator<Tuple> result;
    try {
        result = keySorter.getIterator();
    } catch (InterruptedException e) {
        throw new IOException(e);
    }
    recordsReader = new SeekableFileChannelInputView(ioManager, recordsChannel, memManager, memForRecordReader, lastBlockBytesRecords);
    return new FetchingIterator<T>(serializer, result, recordsReader, keySerializer, numKeyFields);
}
Also used : SeekableFileChannelInputView(org.apache.flink.runtime.io.disk.SeekableFileChannelInputView) MemoryAllocationException(org.apache.flink.runtime.memory.MemoryAllocationException) ArrayList(java.util.ArrayList) IOException(java.io.IOException) MemorySegment(org.apache.flink.core.memory.MemorySegment) InputViewIterator(org.apache.flink.runtime.io.disk.InputViewIterator) SeekableFileChannelInputView(org.apache.flink.runtime.io.disk.SeekableFileChannelInputView) FileChannelInputView(org.apache.flink.runtime.io.disk.FileChannelInputView) Tuple(org.apache.flink.api.java.tuple.Tuple)

Aggregations

IOException (java.io.IOException)1 ArrayList (java.util.ArrayList)1 Tuple (org.apache.flink.api.java.tuple.Tuple)1 MemorySegment (org.apache.flink.core.memory.MemorySegment)1 FileChannelInputView (org.apache.flink.runtime.io.disk.FileChannelInputView)1 InputViewIterator (org.apache.flink.runtime.io.disk.InputViewIterator)1 SeekableFileChannelInputView (org.apache.flink.runtime.io.disk.SeekableFileChannelInputView)1 MemoryAllocationException (org.apache.flink.runtime.memory.MemoryAllocationException)1