@Override
public List<InputSplit> getSplits(JobContext context) throws IOException, InterruptedException {
JobConf jobConf = (JobConf) HadoopCompat.getConfiguration(context);
initInputFormat(jobConf);
org.apache.hadoop.mapred.InputSplit[] splits =
realInputFormat.getSplits(jobConf, jobConf.getNumMapTasks());
if (splits == null) {
return null;
}
List<InputSplit> resultSplits = new ArrayList<InputSplit>(splits.length);
for (org.apache.hadoop.mapred.InputSplit split : splits) {
if (split.getClass() == org.apache.hadoop.mapred.FileSplit.class) {
org.apache.hadoop.mapred.FileSplit mapredFileSplit =
((org.apache.hadoop.mapred.FileSplit) split);
resultSplits.add(
new FileSplit(
mapredFileSplit.getPath(),
mapredFileSplit.getStart(),
mapredFileSplit.getLength(),
mapredFileSplit.getLocations()));
} else {
resultSplits.add(new InputSplitWrapper(split));
}
}
return resultSplits;
}
/**
* set up input file which has the list of input files.
*
* @return boolean
* @throws IOException
*/
private boolean setup() throws IOException {
estimateSavings();
final String randomId = getRandomId();
JobClient jClient = new JobClient(jobconf);
Path jobdir = new Path(jClient.getSystemDir(), NAME + "_" + randomId);
LOG.info(JOB_DIR_LABEL + "=" + jobdir);
jobconf.set(JOB_DIR_LABEL, jobdir.toString());
Path log = new Path(jobdir, "_logs");
// The control file should have small size blocks. This helps
// in spreading out the load from mappers that will be spawned.
jobconf.setInt("dfs.blocks.size", OP_LIST_BLOCK_SIZE);
FileOutputFormat.setOutputPath(jobconf, log);
LOG.info("log=" + log);
// create operation list
FileSystem fs = jobdir.getFileSystem(jobconf);
Path opList = new Path(jobdir, "_" + OP_LIST_LABEL);
jobconf.set(OP_LIST_LABEL, opList.toString());
int opCount = 0, synCount = 0;
SequenceFile.Writer opWriter = null;
try {
opWriter =
SequenceFile.createWriter(
fs, jobconf, opList, Text.class, PolicyInfo.class, SequenceFile.CompressionType.NONE);
for (RaidPolicyPathPair p : raidPolicyPathPairList) {
// If a large set of files are Raided for the first time, files
// in the same directory that tend to have the same size will end up
// with the same map. This shuffle mixes things up, allowing a better
// mix of files.
java.util.Collections.shuffle(p.srcPaths);
for (FileStatus st : p.srcPaths) {
opWriter.append(new Text(st.getPath().toString()), p.policy);
opCount++;
if (++synCount > SYNC_FILE_MAX) {
opWriter.sync();
synCount = 0;
}
}
}
} finally {
if (opWriter != null) {
opWriter.close();
}
fs.setReplication(opList, OP_LIST_REPLICATION); // increase replication for control file
}
raidPolicyPathPairList.clear();
jobconf.setInt(OP_COUNT_LABEL, opCount);
LOG.info("Number of files=" + opCount);
jobconf.setNumMapTasks(
getMapCount(opCount, new JobClient(jobconf).getClusterStatus().getTaskTrackers()));
LOG.info("jobName= " + jobName + " numMapTasks=" + jobconf.getNumMapTasks());
return opCount != 0;
}
// method to write splits for old api mapper.
private int writeOldSplits(JobConf job, Path jobSubmitDir) throws IOException {
org.apache.hadoop.mapred.InputSplit[] splits =
job.getInputFormat().getSplits(job, job.getNumMapTasks());
// sort the splits into order based on size, so that the biggest
// go first
Arrays.sort(
splits,
new Comparator<org.apache.hadoop.mapred.InputSplit>() {
public int compare(
org.apache.hadoop.mapred.InputSplit a, org.apache.hadoop.mapred.InputSplit b) {
try {
long left = a.getLength();
long right = b.getLength();
if (left == right) {
return 0;
} else if (left < right) {
return 1;
} else {
return -1;
}
} catch (IOException ie) {
throw new RuntimeException("Problem getting input split size", ie);
}
}
});
JobSplitWriter.createSplitFiles(jobSubmitDir, job, jobSubmitDir.getFileSystem(job), splits);
return splits.length;
}
/**
* Use the input splits to take samples of the input and generate sample keys. By default reads
* 100,000 keys from 10 locations in the input, sorts them and picks N-1 keys to generate N
* equally sized partitions.
*
* @param conf the job to sample
* @param partFile where to write the output file to
* @throws IOException if something goes wrong
*/
public static void writePartitionFile(JobConf conf, Path partFile) throws IOException {
TeraInputFormat inFormat = new TeraInputFormat();
TextSampler sampler = new TextSampler();
Text key = new Text();
Text value = new Text();
int partitions = conf.getNumReduceTasks();
long sampleSize = conf.getLong(SAMPLE_SIZE, 100000);
InputSplit[] splits = inFormat.getSplits(conf, conf.getNumMapTasks());
int samples = Math.min(10, splits.length);
long recordsPerSample = sampleSize / samples;
int sampleStep = splits.length / samples;
long records = 0;
// take N samples from different parts of the input
for (int i = 0; i < samples; ++i) {
RecordReader<Text, Text> reader =
inFormat.getRecordReader(splits[sampleStep * i], conf, null);
while (reader.next(key, value)) {
sampler.addKey(key);
records += 1;
if ((i + 1) * recordsPerSample <= records) {
break;
}
}
}
FileSystem outFs = partFile.getFileSystem(conf);
if (outFs.exists(partFile)) {
outFs.delete(partFile, false);
}
SequenceFile.Writer writer =
SequenceFile.createWriter(outFs, conf, partFile, Text.class, NullWritable.class);
NullWritable nullValue = NullWritable.get();
for (Text split : sampler.createPartitions(partitions)) {
writer.append(split, nullValue);
}
writer.close();
}
//
// new API - just delegates to the Old API
//
@Override
public List<InputSplit> getSplits(JobContext context) {
JobConf conf = (JobConf) context.getConfiguration();
return Arrays.asList((InputSplit[]) getSplits(conf, conf.getNumMapTasks()));
}
