@Override
public InputSplit[] getSplits(JobConf entries, int i) throws IOException {
Path baseDir = null;
if (entries.get(BASE_DIR) != null) baseDir = new Path(entries.get(BASE_DIR));
StringableList tmpDeltaDirs = new StringableList(entries.get(DELTA_DIRS));
Path[] deltaDirs = tmpDeltaDirs.toArray(new Path[tmpDeltaDirs.size()]);
StringableList dirsToSearch = new StringableList(entries.get(DIRS_TO_SEARCH));
Map<Integer, BucketTracker> splitToBucketMap = new HashMap<Integer, BucketTracker>();
for (Path dir : dirsToSearch) {
FileSystem fs = dir.getFileSystem(entries);
// If this is a base or delta directory, then we need to be looking for the bucket files.
// But if it's a legacy file then we need to add it directly.
if (dir.getName().startsWith(AcidUtils.BASE_PREFIX)
|| dir.getName().startsWith(AcidUtils.DELTA_PREFIX)) {
boolean sawBase = dir.getName().startsWith(AcidUtils.BASE_PREFIX);
FileStatus[] files = fs.listStatus(dir, AcidUtils.bucketFileFilter);
for (FileStatus f : files) {
// For each file, figure out which bucket it is.
Matcher matcher = AcidUtils.BUCKET_DIGIT_PATTERN.matcher(f.getPath().getName());
addFileToMap(matcher, f.getPath(), sawBase, splitToBucketMap);
}
} else {
// Legacy file, see if it's a bucket file
Matcher matcher = AcidUtils.LEGACY_BUCKET_DIGIT_PATTERN.matcher(dir.getName());
addFileToMap(matcher, dir, true, splitToBucketMap);
}
}
List<InputSplit> splits = new ArrayList<InputSplit>(splitToBucketMap.size());
for (Map.Entry<Integer, BucketTracker> e : splitToBucketMap.entrySet()) {
BucketTracker bt = e.getValue();
splits.add(
new CompactorInputSplit(
entries, e.getKey(), bt.buckets, bt.sawBase ? baseDir : null, deltaDirs));
}
LOG.debug("Returning " + splits.size() + " splits");
return splits.toArray(new InputSplit[splits.size()]);
}
/**
* Run a compactor job.
*
* @param conf Hive configuration file
* @param jobName name to run this job with
* @param t metastore table
* @param sd metastore storage descriptor
* @param txns list of valid transactions
* @param isMajor is this a major compaction?
* @throws java.io.IOException if the job fails
*/
void run(
HiveConf conf,
String jobName,
Table t,
StorageDescriptor sd,
ValidTxnList txns,
boolean isMajor,
Worker.StatsUpdater su)
throws IOException {
JobConf job = new JobConf(conf);
job.setJobName(jobName);
job.setOutputKeyClass(NullWritable.class);
job.setOutputValueClass(NullWritable.class);
job.setJarByClass(CompactorMR.class);
LOG.debug("User jar set to " + job.getJar());
job.setMapperClass(CompactorMap.class);
job.setNumReduceTasks(0);
job.setInputFormat(CompactorInputFormat.class);
job.setOutputFormat(NullOutputFormat.class);
job.setOutputCommitter(CompactorOutputCommitter.class);
String queueName = conf.getVar(HiveConf.ConfVars.COMPACTOR_JOB_QUEUE);
if (queueName != null && queueName.length() > 0) {
job.setQueueName(queueName);
}
job.set(FINAL_LOCATION, sd.getLocation());
job.set(TMP_LOCATION, sd.getLocation() + "/" + TMPDIR + "_" + UUID.randomUUID().toString());
job.set(INPUT_FORMAT_CLASS_NAME, sd.getInputFormat());
job.set(OUTPUT_FORMAT_CLASS_NAME, sd.getOutputFormat());
job.setBoolean(IS_MAJOR, isMajor);
job.setBoolean(IS_COMPRESSED, sd.isCompressed());
job.set(TABLE_PROPS, new StringableMap(t.getParameters()).toString());
job.setInt(NUM_BUCKETS, sd.getNumBuckets());
job.set(ValidTxnList.VALID_TXNS_KEY, txns.toString());
setColumnTypes(job, sd.getCols());
// Figure out and encode what files we need to read. We do this here (rather than in
// getSplits below) because as part of this we discover our minimum and maximum transactions,
// and discovering that in getSplits is too late as we then have no way to pass it to our
// mapper.
AcidUtils.Directory dir = AcidUtils.getAcidState(new Path(sd.getLocation()), conf, txns, false);
StringableList dirsToSearch = new StringableList();
Path baseDir = null;
if (isMajor) {
// There may not be a base dir if the partition was empty before inserts or if this
// partition is just now being converted to ACID.
baseDir = dir.getBaseDirectory();
if (baseDir == null) {
List<HdfsFileStatusWithId> originalFiles = dir.getOriginalFiles();
if (!(originalFiles == null) && !(originalFiles.size() == 0)) {
// There are original format files
for (HdfsFileStatusWithId stat : originalFiles) {
Path path = stat.getFileStatus().getPath();
dirsToSearch.add(path);
LOG.debug("Adding original file " + path + " to dirs to search");
}
// Set base to the location so that the input format reads the original files.
baseDir = new Path(sd.getLocation());
}
} else {
// add our base to the list of directories to search for files in.
LOG.debug("Adding base directory " + baseDir + " to dirs to search");
dirsToSearch.add(baseDir);
}
}
List<AcidUtils.ParsedDelta> parsedDeltas = dir.getCurrentDirectories();
if (parsedDeltas == null || parsedDeltas.size() == 0) {
// Seriously, no deltas? Can't compact that.
LOG.error("No delta files found to compact in " + sd.getLocation());
return;
}
StringableList deltaDirs = new StringableList();
long minTxn = Long.MAX_VALUE;
long maxTxn = Long.MIN_VALUE;
for (AcidUtils.ParsedDelta delta : parsedDeltas) {
LOG.debug("Adding delta " + delta.getPath() + " to directories to search");
dirsToSearch.add(delta.getPath());
deltaDirs.add(delta.getPath());
minTxn = Math.min(minTxn, delta.getMinTransaction());
maxTxn = Math.max(maxTxn, delta.getMaxTransaction());
}
if (baseDir != null) job.set(BASE_DIR, baseDir.toString());
job.set(DELTA_DIRS, deltaDirs.toString());
job.set(DIRS_TO_SEARCH, dirsToSearch.toString());
job.setLong(MIN_TXN, minTxn);
job.setLong(MAX_TXN, maxTxn);
LOG.debug("Setting minimum transaction to " + minTxn);
LOG.debug("Setting maximume transaction to " + maxTxn);
RunningJob rj = JobClient.runJob(job);
LOG.info(
"Submitted "
+ (isMajor ? CompactionType.MAJOR : CompactionType.MINOR)
+ " compaction job '"
+ jobName
+ "' with jobID="
+ rj.getID()
+ " to "
+ job.getQueueName()
+ " queue. "
+ "(current delta dirs count="
+ dir.getCurrentDirectories().size()
+ ", obsolete delta dirs count="
+ dir.getObsolete());
rj.waitForCompletion();
su.gatherStats();
}
