use of org.apache.tez.runtime.library.api.IOInterruptedException in project tez by apache.
the class DefaultSorter method interruptSpillThread.
void interruptSpillThread() throws IOException {
assert !spillLock.isHeldByCurrentThread();
// sufficient motivation for the latter approach.
try {
spillThread.interrupt();
spillThread.join();
} catch (InterruptedException e) {
LOG.info(outputContext.getDestinationVertexName() + ": " + "Spill thread interrupted");
// Reset status
Thread.currentThread().interrupt();
throw new IOInterruptedException("Spill failed", e);
}
}
use of org.apache.tez.runtime.library.api.IOInterruptedException in project tez by apache.
the class PipelinedSorter method flush.
@Override
public void flush() throws IOException {
final String uniqueIdentifier = outputContext.getUniqueIdentifier();
outputContext.notifyProgress();
/**
* Possible that the thread got interrupted when flush was happening or when the flush was
* never invoked. As a part of cleanup activity in TezTaskRunner, it would invoke close()
* on all I/O. At that time, this is safe to cleanup
*/
if (isThreadInterrupted()) {
return;
}
try {
LOG.info(outputContext.getDestinationVertexName() + ": Starting flush of map output");
span.end();
merger.add(span.sort(sorter));
// force a spill in flush()
// case 1: we want to force because of following scenarios:
// we have no keys written, and flush got called
// we want atleast one spill(be it empty)
// case 2: in pipeline shuffle case, we have no way of
// knowing the last key being written until flush is called
// so for flush()->spill() we want to force spill so that
// we can send pipeline shuffle event with last event true.
spill(false);
sortmaster.shutdown();
// safe to clean up
buffers.clear();
if (indexCacheList.isEmpty()) {
/*
* If we do not have this check, and if the task gets killed in the middle, it can throw
* NPE leading to distraction when debugging.
*/
if (LOG.isDebugEnabled()) {
LOG.debug(outputContext.getDestinationVertexName() + ": Index list is empty... returning");
}
return;
}
if (!isFinalMergeEnabled()) {
// Generate events for all spills
List<Event> events = Lists.newLinkedList();
// For pipelined shuffle, previous events are already sent. Just generate the last event alone
int startIndex = (pipelinedShuffle) ? (numSpills - 1) : 0;
int endIndex = numSpills;
for (int i = startIndex; i < endIndex; i++) {
boolean isLastEvent = (i == numSpills - 1);
String pathComponent = (outputContext.getUniqueIdentifier() + "_" + i);
ShuffleUtils.generateEventOnSpill(events, isFinalMergeEnabled(), isLastEvent, outputContext, i, indexCacheList.get(i), partitions, sendEmptyPartitionDetails, pathComponent, partitionStats, reportDetailedPartitionStats(), auxiliaryService, deflater);
LOG.info(outputContext.getDestinationVertexName() + ": Adding spill event for spill (final update=" + isLastEvent + "), spillId=" + i);
}
outputContext.sendEvents(events);
return;
}
numAdditionalSpills.increment(numSpills - 1);
// In case final merge is required, the following code path is executed.
if (numSpills == 1) {
// someday be able to pass this directly to shuffle
// without writing to disk
final Path filename = spillFilePaths.get(0);
final Path indexFilename = spillFileIndexPaths.get(0);
finalOutputFile = mapOutputFile.getOutputFileForWriteInVolume(filename);
finalIndexFile = mapOutputFile.getOutputIndexFileForWriteInVolume(indexFilename);
sameVolRename(filename, finalOutputFile);
sameVolRename(indexFilename, finalIndexFile);
if (LOG.isDebugEnabled()) {
LOG.debug(outputContext.getDestinationVertexName() + ": numSpills=" + numSpills + ", finalOutputFile=" + finalOutputFile + ", " + "finalIndexFile=" + finalIndexFile + ", filename=" + filename + ", indexFilename=" + indexFilename);
}
TezSpillRecord spillRecord = new TezSpillRecord(finalIndexFile, conf);
if (reportPartitionStats()) {
for (int i = 0; i < spillRecord.size(); i++) {
partitionStats[i] += spillRecord.getIndex(i).getPartLength();
}
}
numShuffleChunks.setValue(numSpills);
fileOutputByteCounter.increment(rfs.getFileStatus(finalOutputFile).getLen());
// ??? why are events not being sent here?
return;
}
finalOutputFile = // TODO
mapOutputFile.getOutputFileForWrite(0);
finalIndexFile = // TODO
mapOutputFile.getOutputIndexFileForWrite(0);
if (LOG.isDebugEnabled()) {
LOG.debug(outputContext.getDestinationVertexName() + ": " + "numSpills: " + numSpills + ", finalOutputFile:" + finalOutputFile + ", finalIndexFile:" + finalIndexFile);
}
// The output stream for the final single output file
FSDataOutputStream finalOut = rfs.create(finalOutputFile, true, 4096);
if (!SPILL_FILE_PERMS.equals(SPILL_FILE_PERMS.applyUMask(FsPermission.getUMask(conf)))) {
rfs.setPermission(finalOutputFile, SPILL_FILE_PERMS);
}
final TezSpillRecord spillRec = new TezSpillRecord(partitions);
for (int parts = 0; parts < partitions; parts++) {
boolean shouldWrite = false;
// create the segments to be merged
List<Segment> segmentList = new ArrayList<Segment>(numSpills);
for (int i = 0; i < numSpills; i++) {
Path spillFilename = spillFilePaths.get(i);
TezIndexRecord indexRecord = indexCacheList.get(i).getIndex(parts);
if (indexRecord.hasData() || !sendEmptyPartitionDetails) {
shouldWrite = true;
DiskSegment s = new DiskSegment(rfs, spillFilename, indexRecord.getStartOffset(), indexRecord.getPartLength(), codec, ifileReadAhead, ifileReadAheadLength, ifileBufferSize, true);
segmentList.add(s);
}
}
int mergeFactor = this.conf.getInt(TezRuntimeConfiguration.TEZ_RUNTIME_IO_SORT_FACTOR, TezRuntimeConfiguration.TEZ_RUNTIME_IO_SORT_FACTOR_DEFAULT);
// sort the segments only if there are intermediate merges
boolean sortSegments = segmentList.size() > mergeFactor;
// merge
TezRawKeyValueIterator kvIter = TezMerger.merge(conf, rfs, keyClass, valClass, codec, segmentList, mergeFactor, new Path(uniqueIdentifier), (RawComparator) ConfigUtils.getIntermediateOutputKeyComparator(conf), progressable, sortSegments, true, null, spilledRecordsCounter, additionalSpillBytesRead, null, // Not using any Progress in TezMerger. Should just work.
merger.needsRLE());
// write merged output to disk
long segmentStart = finalOut.getPos();
long rawLength = 0;
long partLength = 0;
if (shouldWrite) {
Writer writer = new Writer(conf, finalOut, keyClass, valClass, codec, spilledRecordsCounter, null, merger.needsRLE());
if (combiner == null || numSpills < minSpillsForCombine) {
TezMerger.writeFile(kvIter, writer, progressable, TezRuntimeConfiguration.TEZ_RUNTIME_RECORDS_BEFORE_PROGRESS_DEFAULT);
} else {
runCombineProcessor(kvIter, writer);
}
// close
writer.close();
rawLength = writer.getRawLength();
partLength = writer.getCompressedLength();
}
outputBytesWithOverheadCounter.increment(rawLength);
// record offsets
final TezIndexRecord rec = new TezIndexRecord(segmentStart, rawLength, partLength);
spillRec.putIndex(rec, parts);
if (reportPartitionStats()) {
partitionStats[parts] += partLength;
}
}
// final merge has happened.
numShuffleChunks.setValue(1);
fileOutputByteCounter.increment(rfs.getFileStatus(finalOutputFile).getLen());
spillRec.writeToFile(finalIndexFile, conf);
finalOut.close();
for (int i = 0; i < numSpills; i++) {
Path indexFilename = spillFileIndexPaths.get(i);
Path spillFilename = spillFilePaths.get(i);
rfs.delete(indexFilename, true);
rfs.delete(spillFilename, true);
}
spillFileIndexPaths.clear();
spillFilePaths.clear();
} catch (InterruptedException ie) {
if (cleanup) {
cleanup();
}
Thread.currentThread().interrupt();
throw new IOInterruptedException("Interrupted while closing Output", ie);
}
}
use of org.apache.tez.runtime.library.api.IOInterruptedException in project tez by apache.
the class UnorderedPartitionedKVWriter method getNextAvailableBuffer.
private WrappedBuffer getNextAvailableBuffer() throws IOException {
if (availableBuffers.peek() == null) {
if (numInitializedBuffers < numBuffers) {
buffers[numInitializedBuffers] = new WrappedBuffer(numPartitions, numInitializedBuffers == numBuffers - 1 ? lastBufferSize : sizePerBuffer);
numInitializedBuffers++;
return buffers[numInitializedBuffers - 1];
} else {
// All buffers initialized, and none available right now. Wait
try {
// Ensure that spills are triggered so that buffers can be released.
mayBeSpill(true);
return availableBuffers.take();
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw new IOInterruptedException("Interrupted while waiting for next buffer", e);
}
}
} else {
return availableBuffers.poll();
}
}
use of org.apache.tez.runtime.library.api.IOInterruptedException in project tez by apache.
the class TestUnorderedKVReader method testInterruptOnNext.
@Test(timeout = 5000)
public void testInterruptOnNext() throws IOException, InterruptedException {
ShuffleManager shuffleManager = mock(ShuffleManager.class);
// Simulate an interrupt while waiting for the next fetched input.
doThrow(new InterruptedException()).when(shuffleManager).getNextInput();
TezCounters counters = new TezCounters();
TezCounter inputRecords = counters.findCounter(TaskCounter.INPUT_RECORDS_PROCESSED);
UnorderedKVReader<Text, Text> reader = new UnorderedKVReader<Text, Text>(shuffleManager, defaultConf, null, false, -1, -1, inputRecords, mock(InputContext.class));
try {
reader.next();
fail("No data available to reader. Should not be able to access any record");
} catch (IOInterruptedException e) {
// Expected exception. Any other should fail the test.
}
}
use of org.apache.tez.runtime.library.api.IOInterruptedException in project tez by apache.
the class OrderedGroupedKVInput method getReader.
/**
* Get a KVReader for the Input.</p> This method will block until the input is
* ready - i.e. the copy and merge stages are complete. Users can use the
* isInputReady method to check if the input is ready, which gives an
* indication of whether this method will block or not.
*
* NOTE: All values for the current K-V pair must be read prior to invoking
* moveToNext. Once moveToNext() is called, the valueIterator from the
* previous K-V pair will throw an Exception
*
* @return a KVReader over the sorted input.
* @throws {@link IOInterruptedException} if IO was performing a blocking operation and was interrupted
*/
@Override
public KeyValuesReader getReader() throws IOException, TezException {
// Cannot synchronize entire method since this is called form user code and can block.
TezRawKeyValueIterator rawIterLocal;
synchronized (this) {
rawIterLocal = rawIter;
if (getNumPhysicalInputs() == 0) {
return new KeyValuesReader() {
@Override
public boolean next() throws IOException {
getContext().notifyProgress();
hasCompletedProcessing();
completedProcessing = true;
return false;
}
@Override
public Object getCurrentKey() throws IOException {
throw new RuntimeException("No data available in Input");
}
@Override
public Iterable<Object> getCurrentValues() throws IOException {
throw new RuntimeException("No data available in Input");
}
};
}
}
if (rawIterLocal == null) {
try {
waitForInputReady();
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw new IOInterruptedException("Interrupted while waiting for input ready", e);
}
}
@SuppressWarnings("rawtypes") ValuesIterator valuesIter = null;
synchronized (this) {
valuesIter = vIter;
}
return new OrderedGroupedKeyValuesReader(valuesIter, getContext());
}
Aggregations