/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.tools.mapred.lib;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.hadoop.mapreduce.*;
import org.apache.hadoop.mapreduce.lib.input.FileSplit;
import org.apache.hadoop.tools.DistCpConstants;
import org.apache.hadoop.tools.util.DistCpUtils;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.io.SequenceFile;
import org.apache.hadoop.io.IOUtils;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.fs.FileStatus;
import java.util.List;
import java.util.ArrayList;
import java.io.IOException;
/**
* DynamicInputFormat implements the "Worker pattern" for DistCp.
* Rather than to split up the copy-list into a set of static splits,
* the DynamicInputFormat does the following:
* 1. Splits the copy-list into small chunks on the DFS.
* 2. Creates a set of empty "dynamic" splits, that each consume as many chunks
* as it can.
* This arrangement ensures that a single slow mapper won't slow down the entire
* job (since the slack will be picked up by other mappers, who consume more
* chunks.)
* By varying the split-ratio, one can vary chunk sizes to achieve different
* performance characteristics.
*/
public class DynamicInputFormat<K, V> extends InputFormat<K, V> {
private static final Log LOG = LogFactory.getLog(DynamicInputFormat.class);
private static final String CONF_LABEL_LISTING_SPLIT_RATIO
= "mapred.listing.split.ratio";
private static final String CONF_LABEL_NUM_SPLITS
= "mapred.num.splits";
private static final String CONF_LABEL_NUM_ENTRIES_PER_CHUNK
= "mapred.num.entries.per.chunk";
/**
* Implementation of InputFormat::getSplits(). This method splits up the
* copy-listing file into chunks, and assigns the first batch to different
* tasks.
* @param jobContext JobContext for the map job.
* @return The list of (empty) dynamic input-splits.
* @throws IOException, on failure.
* @throws InterruptedException
*/
@Override
public List<InputSplit> getSplits(JobContext jobContext)
throws IOException, InterruptedException {
LOG.info("DynamicInputFormat: Getting splits for job:"
+ jobContext.getJobID());
return createSplits(jobContext,
splitCopyListingIntoChunksWithShuffle(jobContext));
}
private List<InputSplit> createSplits(JobContext jobContext,
List<DynamicInputChunk> chunks)
throws IOException {
int numMaps = getNumMapTasks(jobContext.getConfiguration());
final int nSplits = Math.min(numMaps, chunks.size());
List<InputSplit> splits = new ArrayList<InputSplit>(nSplits);
for (int i=0; i< nSplits; ++i) {
TaskID taskId = new TaskID(jobContext.getJobID(), TaskType.MAP, i);
chunks.get(i).assignTo(taskId);
splits.add(new FileSplit(chunks.get(i).getPath(), 0,
// Setting non-zero length for FileSplit size, to avoid a possible
// future when 0-sized file-splits are considered "empty" and skipped
// over.
MIN_RECORDS_PER_CHUNK,
null));
}
DistCpUtils.publish(jobContext.getConfiguration(),
CONF_LABEL_NUM_SPLITS, splits.size());
return splits;
}
private static int N_CHUNKS_OPEN_AT_ONCE_DEFAULT = 16;
private List<DynamicInputChunk> splitCopyListingIntoChunksWithShuffle
(JobContext context) throws IOException {
final Configuration configuration = context.getConfiguration();
int numRecords = getNumberOfRecords(configuration);
int numMaps = getNumMapTasks(configuration);
// Number of chunks each map will process, on average.
int splitRatio = getListingSplitRatio(configuration, numMaps, numRecords);
validateNumChunksUsing(splitRatio, numMaps);
int numEntriesPerChunk = (int)Math.ceil((float)numRecords
/(splitRatio * numMaps));
DistCpUtils.publish(context.getConfiguration(),
CONF_LABEL_NUM_ENTRIES_PER_CHUNK,
numEntriesPerChunk);
final int nChunksTotal = (int)Math.ceil((float)numRecords/numEntriesPerChunk);
int nChunksOpenAtOnce
= Math.min(N_CHUNKS_OPEN_AT_ONCE_DEFAULT, nChunksTotal);
Path listingPath = getListingFilePath(configuration);
SequenceFile.Reader reader
= new SequenceFile.Reader(configuration,
SequenceFile.Reader.file(listingPath));
List<DynamicInputChunk> openChunks
= new ArrayList<DynamicInputChunk>();
List<DynamicInputChunk> chunksFinal = new ArrayList<DynamicInputChunk>();
FileStatus fileStatus = new FileStatus();
Text relPath = new Text();
int recordCounter = 0;
int chunkCount = 0;
try {
while (reader.next(relPath, fileStatus)) {
if (recordCounter % (nChunksOpenAtOnce*numEntriesPerChunk) == 0) {
// All chunks full. Create new chunk-set.
closeAll(openChunks);
chunksFinal.addAll(openChunks);
openChunks = createChunks(
configuration, chunkCount, nChunksTotal, nChunksOpenAtOnce);
chunkCount += openChunks.size();
nChunksOpenAtOnce = openChunks.size();
recordCounter = 0;
}
// Shuffle into open chunks.
openChunks.get(recordCounter%nChunksOpenAtOnce).write(relPath, fileStatus);
++recordCounter;
}
} finally {
closeAll(openChunks);
chunksFinal.addAll(openChunks);
IOUtils.closeStream(reader);
}
LOG.info("Number of dynamic-chunk-files created: " + chunksFinal.size());
return chunksFinal;
}
private static void validateNumChunksUsing(int splitRatio, int numMaps)
throws IOException {
if (splitRatio * numMaps > MAX_CHUNKS_TOLERABLE)
throw new IOException("Too many chunks created with splitRatio:"
+ splitRatio + ", numMaps:" + numMaps
+ ". Reduce numMaps or decrease split-ratio to proceed.");
}
private static void closeAll(List<DynamicInputChunk> chunks) {
for (DynamicInputChunk chunk: chunks)
chunk.close();
}
private static List<DynamicInputChunk> createChunks(Configuration config,
int chunkCount, int nChunksTotal, int nChunksOpenAtOnce)
throws IOException {
List<DynamicInputChunk> chunks = new ArrayList<DynamicInputChunk>();
int chunkIdUpperBound
= Math.min(nChunksTotal, chunkCount + nChunksOpenAtOnce);
// If there will be fewer than nChunksOpenAtOnce chunks left after
// the current batch of chunks, fold the remaining chunks into
// the current batch.
if (nChunksTotal - chunkIdUpperBound < nChunksOpenAtOnce)
chunkIdUpperBound = nChunksTotal;
for (int i=chunkCount; i < chunkIdUpperBound; ++i)
chunks.add(createChunk(i, config));
return chunks;
}
private static DynamicInputChunk createChunk(int chunkId, Configuration config)
throws IOException {
return DynamicInputChunk.createChunkForWrite(String.format("%05d", chunkId),
config);
}
private static Path getListingFilePath(Configuration configuration) {
String listingFilePathString = configuration.get(
DistCpConstants.CONF_LABEL_LISTING_FILE_PATH, "");
assert !listingFilePathString.equals("") : "Listing file not found.";
Path listingFilePath = new Path(listingFilePathString);
try {
assert listingFilePath.getFileSystem(configuration)
.exists(listingFilePath) : "Listing file: " + listingFilePath +
" not found.";
} catch (IOException e) {
assert false : "Listing file: " + listingFilePath
+ " couldn't be accessed. " + e.getMessage();
}
return listingFilePath;
}
private static int getNumberOfRecords(Configuration configuration) {
return DistCpUtils.getInt(configuration,
DistCpConstants.CONF_LABEL_TOTAL_NUMBER_OF_RECORDS);
}
private static int getNumMapTasks(Configuration configuration) {
return DistCpUtils.getInt(configuration,
JobContext.NUM_MAPS);
}
private static int getListingSplitRatio(Configuration configuration,
int numMaps, int numPaths) {
return configuration.getInt(
CONF_LABEL_LISTING_SPLIT_RATIO,
getSplitRatio(numMaps, numPaths));
}
private static final int MAX_CHUNKS_TOLERABLE = 400;
private static final int MAX_CHUNKS_IDEAL = 100;
private static final int MIN_RECORDS_PER_CHUNK = 5;
private static final int SPLIT_RATIO_DEFAULT = 2;
/**
* Package private, for testability.
* @param nMaps The number of maps requested for.
* @param nRecords The number of records to be copied.
* @return The number of splits each map should handle, ideally.
*/
static int getSplitRatio(int nMaps, int nRecords) {
if (nMaps == 1) {
LOG.warn("nMaps == 1. Why use DynamicInputFormat?");
return 1;
}
if (nMaps > MAX_CHUNKS_IDEAL)
return SPLIT_RATIO_DEFAULT;
int nPickups = (int)Math.ceil((float)MAX_CHUNKS_IDEAL/nMaps);
int nRecordsPerChunk = (int)Math.ceil((float)nRecords/(nMaps*nPickups));
return nRecordsPerChunk < MIN_RECORDS_PER_CHUNK ?
SPLIT_RATIO_DEFAULT : nPickups;
}
static int getNumEntriesPerChunk(Configuration configuration) {
return DistCpUtils.getInt(configuration,
CONF_LABEL_NUM_ENTRIES_PER_CHUNK);
}
/**
* Implementation of Inputformat::createRecordReader().
* @param inputSplit The split for which the RecordReader is required.
* @param taskAttemptContext TaskAttemptContext for the current attempt.
* @return DynamicRecordReader instance.
* @throws IOException, on failure.
* @throws InterruptedException
*/
@Override
public RecordReader<K, V> createRecordReader(
InputSplit inputSplit,
TaskAttemptContext taskAttemptContext)
throws IOException, InterruptedException {
return new DynamicRecordReader<K, V>();
}
}