private void mergeParts() throws IOException {
// get the approximate size of the final output/index files
long finalOutFileSize = 0;
long finalIndexFileSize = 0;
Path[] filename = new Path[numSpills];
Path[] indexFileName = new Path[numSpills];
FileSystem localFs = FileSystem.getLocal(job);
for (int i = 0; i < numSpills; i++) {
filename[i] = mapOutputFile.getSpillFile(getTaskID(), i);
indexFileName[i] = mapOutputFile.getSpillIndexFile(getTaskID(), i);
finalOutFileSize += localFs.getFileStatus(filename[i]).getLen();
}
if (numSpills == 1) { // the spill is the final output
localFs.rename(filename[0], new Path(filename[0].getParent(), "file.out"));
localFs.rename(indexFileName[0], new Path(indexFileName[0].getParent(), "file.out.index"));
return;
}
// make correction in the length to include the sequence file header
// lengths for each partition
finalOutFileSize += partitions * APPROX_HEADER_LENGTH;
finalIndexFileSize = partitions * MAP_OUTPUT_INDEX_RECORD_LENGTH;
Path finalOutputFile = mapOutputFile.getOutputFileForWrite(getTaskID(), finalOutFileSize);
Path finalIndexFile =
mapOutputFile.getOutputIndexFileForWrite(getTaskID(), finalIndexFileSize);
// The output stream for the final single output file
FSDataOutputStream finalOut = localFs.create(finalOutputFile, true, 4096);
// The final index file output stream
FSDataOutputStream finalIndexOut = localFs.create(finalIndexFile, true, 4096);
if (numSpills == 0) {
// create dummy files
for (int i = 0; i < partitions; i++) {
long segmentStart = finalOut.getPos();
Writer<K, V> writer = new Writer<K, V>(job, finalOut, keyClass, valClass, codec);
writer.close();
writeIndexRecord(finalIndexOut, finalOut, segmentStart, writer);
}
finalOut.close();
finalIndexOut.close();
return;
}
{
for (int parts = 0; parts < partitions; parts++) {
// create the segments to be merged
List<Segment<K, V>> segmentList = new ArrayList<Segment<K, V>>(numSpills);
for (int i = 0; i < numSpills; i++) {
FSDataInputStream indexIn = localFs.open(indexFileName[i]);
indexIn.seek(parts * MAP_OUTPUT_INDEX_RECORD_LENGTH);
long segmentOffset = indexIn.readLong();
long rawSegmentLength = indexIn.readLong();
long segmentLength = indexIn.readLong();
indexIn.close();
Segment<K, V> s =
new Segment<K, V>(
job, localFs, filename[i], segmentOffset, segmentLength, codec, true);
segmentList.add(i, s);
if (LOG.isDebugEnabled()) {
LOG.debug(
"Index: ("
+ indexFileName[i]
+ ", "
+ segmentOffset
+ rawSegmentLength
+ ", "
+ segmentLength
+ ")");
}
}
// merge
@SuppressWarnings("unchecked")
RawKeyValueIterator kvIter =
Merger.merge(
job,
localFs,
keyClass,
valClass,
segmentList,
job.getInt("io.sort.factor", 100),
new Path(getTaskID().toString()),
job.getOutputKeyComparator(),
reporter);
// write merged output to disk
long segmentStart = finalOut.getPos();
Writer<K, V> writer = new Writer<K, V>(job, finalOut, keyClass, valClass, codec);
if (null == combinerClass
|| job.getCombineOnceOnly()
|| numSpills < minSpillsForCombine) {
Merger.writeFile(kvIter, writer, reporter, job);
} else {
combineCollector.setWriter(writer);
combineAndSpill(kvIter, combineInputCounter);
}
// close
writer.close();
// write index record
writeIndexRecord(finalIndexOut, finalOut, segmentStart, writer);
}
finalOut.close();
finalIndexOut.close();
// cleanup
for (int i = 0; i < numSpills; i++) {
localFs.delete(filename[i], true);
localFs.delete(indexFileName[i], true);
}
}
}
/**
* Handles the degenerate case where serialization fails to fit in the in-memory buffer, so we
* must spill the record from collect directly to a spill file. Consider this "losing".
*/
@SuppressWarnings("unchecked")
private void spillSingleRecord(final K key, final V value) throws IOException {
long size = kvbuffer.length + partitions * APPROX_HEADER_LENGTH;
FSDataOutputStream out = null;
FSDataOutputStream indexOut = null;
final int partition = partitioner.getPartition(key, value, partitions);
try {
// create spill file
Path filename = mapOutputFile.getSpillFileForWrite(getTaskID(), numSpills, size);
out = localFs.create(filename);
// create spill index
Path indexFilename =
mapOutputFile.getSpillIndexFileForWrite(
getTaskID(), numSpills, partitions * MAP_OUTPUT_INDEX_RECORD_LENGTH);
indexOut = localFs.create(indexFilename);
// we don't run the combiner for a single record
for (int i = 0; i < partitions; ++i) {
IFile.Writer writer = null;
try {
long segmentStart = out.getPos();
// Create a new codec, don't care!
writer = new IFile.Writer(job, out, keyClass, valClass, codec);
if (i == partition) {
if (job.getCombineOnceOnly()) {
Reducer combiner = (Reducer) ReflectionUtils.newInstance(combinerClass, job);
combineCollector.setWriter(writer);
combiner.reduce(
key,
new Iterator<V>() {
private boolean done = false;
public boolean hasNext() {
return !done;
}
public V next() {
if (done) throw new NoSuchElementException();
done = true;
return value;
}
public void remove() {
throw new UnsupportedOperationException();
}
},
combineCollector,
reporter);
} else {
final long recordStart = out.getPos();
writer.append(key, value);
// Note that our map byte count will not be accurate with
// compression
mapOutputByteCounter.increment(out.getPos() - recordStart);
}
}
writer.close();
// index record
writeIndexRecord(indexOut, out, segmentStart, writer);
} catch (IOException e) {
if (null != writer) writer.close();
throw e;
}
}
++numSpills;
} finally {
if (out != null) out.close();
if (indexOut != null) indexOut.close();
}
}
