@Override
public void configure(JobConf conf) {
this.threshold = conf.getFloat(PARAM_APS_THRESHOLD, DEFAULT_THRESHOLD);
int reducerID = conf.getInt("mapred.task.partition", -1);
int max = conf.getInt(PARAM_APS_MAXKEY, 0);
int nstripes = conf.getInt(PARAM_APS_STRIPES, 1);
int spread = conf.getInt(PARAM_APS_REDUCER_PER_STRIPE, 1);
if (reducerID < 0 || max == 0) {
LOG.error("Could not find stripe ID, reverting to whole rest file loading");
LOG.debug("reducer = " + reducerID + "\t max = " + max + "\t nstripes = " + nstripes);
// open the pruned part file in the DistrubutedCache
haspruned = FileUtils.readRestFile(conf, pruned);
} else {
int stripe = GenericKey.StripePartitioner.findStripe(reducerID, spread);
int from = GenericKey.StripePartitioner.minKeyInStripe(stripe, nstripes, max);
int to = from + GenericKey.StripePartitioner.numKeysInStripe(stripe, nstripes, max);
// read from 'from' included, to 'to' excluded
LOG.info(
"Reducer "
+ reducerID
+ " loading stripe "
+ stripe
+ " of "
+ nstripes
+ " ("
+ from
+ ","
+ (to - 1)
+ ")");
haspruned = FileUtils.readRestFile(conf, pruned, from, to);
}
if (!haspruned) LOG.warn("No pruned file provided in DistributedCache");
else LOG.info("Read " + pruned.size() + " entries from pruned file");
}
/** Save the values out of the configuaration that we need to write the data. */
@Override
public void configure(JobConf job) {
numBytesToWrite = job.getLong("test.randomwrite.bytes_per_map", 1 * 1024 * 1024 * 20);
minKeySize = job.getInt("test.randomwrite.min_key", 100);
keySizeRange = job.getInt("test.randomwrite.max_key", 1000) - minKeySize;
minValueSize = job.getInt("test.randomwrite.min_value", 0);
valueSizeRange = job.getInt("test.randomwrite.max_value", 20000) - minValueSize;
}
@VisibleForTesting
Fetcher(
JobConf job,
TaskAttemptID reduceId,
ShuffleSchedulerImpl<K, V> scheduler,
MergeManager<K, V> merger,
Reporter reporter,
ShuffleClientMetrics metrics,
ExceptionReporter exceptionReporter,
SecretKey shuffleKey,
int id) {
this.jobConf = job;
this.reporter = reporter;
this.scheduler = scheduler;
this.merger = merger;
this.metrics = metrics;
this.exceptionReporter = exceptionReporter;
this.id = id;
this.reduce = reduceId.getTaskID().getId();
this.shuffleSecretKey = shuffleKey;
ioErrs = reporter.getCounter(SHUFFLE_ERR_GRP_NAME, ShuffleErrors.IO_ERROR.toString());
wrongLengthErrs =
reporter.getCounter(SHUFFLE_ERR_GRP_NAME, ShuffleErrors.WRONG_LENGTH.toString());
badIdErrs = reporter.getCounter(SHUFFLE_ERR_GRP_NAME, ShuffleErrors.BAD_ID.toString());
wrongMapErrs = reporter.getCounter(SHUFFLE_ERR_GRP_NAME, ShuffleErrors.WRONG_MAP.toString());
connectionErrs = reporter.getCounter(SHUFFLE_ERR_GRP_NAME, ShuffleErrors.CONNECTION.toString());
wrongReduceErrs =
reporter.getCounter(SHUFFLE_ERR_GRP_NAME, ShuffleErrors.WRONG_REDUCE.toString());
this.connectionTimeout =
job.getInt(MRJobConfig.SHUFFLE_CONNECT_TIMEOUT, DEFAULT_STALLED_COPY_TIMEOUT);
this.readTimeout = job.getInt(MRJobConfig.SHUFFLE_READ_TIMEOUT, DEFAULT_READ_TIMEOUT);
setName("fetcher#" + id);
setDaemon(true);
synchronized (Fetcher.class) {
sslShuffle =
job.getBoolean(MRConfig.SHUFFLE_SSL_ENABLED_KEY, MRConfig.SHUFFLE_SSL_ENABLED_DEFAULT);
if (sslShuffle && sslFactory == null) {
sslFactory = new SSLFactory(SSLFactory.Mode.CLIENT, job);
try {
sslFactory.init();
} catch (Exception ex) {
sslFactory.destroy();
throw new RuntimeException(ex);
}
}
}
}
/**
* Generate the requested number of file splits, with the filename set to the filename of the
* output file.
*/
public InputSplit[] getSplits(JobConf job, int numSplits) throws IOException {
/** 设置输入分片的个数* */
JobClient client = new JobClient(job);
ClusterStatus cluster = client.getClusterStatus();
/** 如果属性不存在 则返回默认的值 * */
int numMapsPerHost = job.getInt("test.randomwriter.maps_per_host", 10);
long numBytesToWritePerMap =
job.getLong("test.randomwrite.bytes_per_map", 1 * 1024 * 1024 * 1024);
if (numBytesToWritePerMap == 0) {
System.err.println("Cannot have test.randomwrite.bytes_per_map set to 0");
}
long totalBytesToWrite =
job.getLong(
"test.randomwrite.total_bytes",
numMapsPerHost * numBytesToWritePerMap * cluster.getTaskTrackers());
int numMaps = (int) (totalBytesToWrite / numBytesToWritePerMap);
if (numMaps == 0 && totalBytesToWrite > 0) {
numMaps = 1;
}
System.out.println("numMaps-------" + numMaps);
InputSplit[] result = new InputSplit[numMaps];
Path outDir = FileOutputFormat.getOutputPath(job);
for (int i = 0; i < result.length; ++i) {
result[i] = new FileSplit(new Path(outDir, "dummy-split-" + i), 0, 1, (String[]) null);
}
return result;
}
/**
* This is the main routine for launching a distributed random write job. It runs 10 maps/node and
* each node writes 1 gig of data to a DFS file. The reduce doesn't do anything.
*
* @throws IOException
*/
public int run(String[] args) throws Exception {
if (args.length == 0) {
System.out.println("Usage: writer <out-dir>");
ToolRunner.printGenericCommandUsage(System.out);
return -1;
}
Path outDir = new Path(args[0]);
JobConf job = new JobConf(getConf());
job.setJarByClass(RandomWriter.class);
job.setJobName("random-writer");
FileOutputFormat.setOutputPath(job, outDir);
job.setOutputKeyClass(BytesWritable.class);
job.setOutputValueClass(BytesWritable.class);
job.setInputFormat(RandomInputFormat.class);
job.setMapperClass(Map.class);
job.setReducerClass(IdentityReducer.class);
job.setOutputFormat(SequenceFileOutputFormat.class);
JobClient client = new JobClient(job);
ClusterStatus cluster = client.getClusterStatus();
/** 如果属性不存在 则返回默认的值 * */
int numMapsPerHost = job.getInt("test.randomwriter.maps_per_host", 10);
long numBytesToWritePerMap =
job.getLong("test.randomwrite.bytes_per_map", 1 * 1024 * 1024 * 1024);
if (numBytesToWritePerMap == 0) {
System.err.println("Cannot have test.randomwrite.bytes_per_map set to 0");
return -2;
}
long totalBytesToWrite =
job.getLong(
"test.randomwrite.total_bytes",
numMapsPerHost * numBytesToWritePerMap * cluster.getTaskTrackers());
int numMaps = (int) (totalBytesToWrite / numBytesToWritePerMap);
if (numMaps == 0 && totalBytesToWrite > 0) {
numMaps = 1;
job.setLong("test.randomwrite.bytes_per_map", totalBytesToWrite);
}
job.setNumMapTasks(numMaps);
/** 建议型的 * */
System.out.println("Running " + numMaps + " maps.");
// reducer NONE
job.setNumReduceTasks(0);
Date startTime = new Date();
System.out.println("Job started: " + startTime);
JobClient.runJob(job);
Date endTime = new Date();
System.out.println("Job ended: " + endTime);
System.out.println(
"The job took " + (endTime.getTime() - startTime.getTime()) / 1000 + " seconds.");
return 0;
}
/**
* passing data and popsize so that they would be available before a call to the mapper
*
* @param job this mapreduce job
*/
@Override
public void configure(JobConf job) {
popsize = job.getInt("popsize", popsize);
data = job.get("dataset", data);
new Population(data, popsize);
}
@Override
public void configure(JobConf job) {
numberOfNodes = job.getInt("numberOfNodes", 0);
if (numberOfNodes == 0) {
System.exit(0);
}
}
public void configure(JobConf job) {
this.jobConf = job;
urlNormalizers = new URLNormalizers(job, URLNormalizers.SCOPE_INJECT);
interval = jobConf.getInt("db.fetch.interval.default", 2592000);
filters = new URLFilters(jobConf);
scfilters = new ScoringFilters(jobConf);
scoreInjected = jobConf.getFloat("db.score.injected", 1.0f);
curTime = job.getLong("injector.current.time", System.currentTimeMillis());
}
/**
* Calculate how many maps to run. Number of maps is bounded by a minimum of the cumulative size
* of the copy / (distcp.bytes.per.map, default BYTES_PER_MAP or -m on the command line) and at
* most (distcp.max.map.tasks, default MAX_MAPS_PER_NODE * nodes in the cluster).
*
* @param totalBytes Count of total bytes for job
* @param job The job to configure
* @return Count of maps to run.
*/
private static void setMapCount(long totalBytes, JobConf job) throws IOException {
int numMaps = (int) (totalBytes / job.getLong(BYTES_PER_MAP_LABEL, BYTES_PER_MAP));
numMaps =
Math.min(
numMaps,
job.getInt(
MAX_MAPS_LABEL,
MAX_MAPS_PER_NODE * new JobClient(job).getClusterStatus().getTaskTrackers()));
job.setNumMapTasks(Math.max(numMaps, 1));
}
public void configure(JobConf job) {
_jobConf = job;
crawlerCount =
job.getInt(CrawlEnvironment.PROPERTY_NUM_CRAWLERS, CrawlEnvironment.CRAWLERS.length);
partitionNumber = job.getInt("mapred.task.partition", -1);
try {
FileSystem fs = FileSystem.get(job);
Path workPath = FileOutputFormat.getOutputPath(job);
debugURLStream =
fs.create(new Path(workPath, "debugURLS-" + NUMBER_FORMAT.format(partitionNumber)));
urlDebugURLWriter = new OutputStreamWriter(debugURLStream, Charset.forName("UTF-8"));
_emittedURLSFilter = new URLFPBloomFilter(NUM_ELEMENTS, NUM_HASH_FUNCTIONS, NUM_BITS);
} catch (IOException e) {
LOG.error(StringUtils.stringifyException(e));
throw new RuntimeException(e);
}
}
/**
* Produce splits such that each is no greater than the quotient of the total size and the
* number of splits requested.
*
* @param job The handle to the JobConf object
* @param numSplits Number of splits requested
*/
public InputSplit[] getSplits(JobConf job, int numSplits) throws IOException {
int cnfiles = job.getInt(SRC_COUNT_LABEL, -1);
long cbsize = job.getLong(TOTAL_SIZE_LABEL, -1);
String srcfilelist = job.get(SRC_LIST_LABEL, "");
if (cnfiles < 0 || cbsize < 0 || "".equals(srcfilelist)) {
throw new RuntimeException(
"Invalid metadata: #files("
+ cnfiles
+ ") total_size("
+ cbsize
+ ") listuri("
+ srcfilelist
+ ")");
}
Path src = new Path(srcfilelist);
FileSystem fs = src.getFileSystem(job);
FileStatus srcst = fs.getFileStatus(src);
ArrayList<FileSplit> splits = new ArrayList<FileSplit>(numSplits);
LongWritable key = new LongWritable();
FilePair value = new FilePair();
final long targetsize = cbsize / numSplits;
long pos = 0L;
long last = 0L;
long acc = 0L;
long cbrem = srcst.getLen();
SequenceFile.Reader sl = null;
try {
sl = new SequenceFile.Reader(fs, src, job);
for (; sl.next(key, value); last = sl.getPosition()) {
// if adding this split would put this split past the target size,
// cut the last split and put this next file in the next split.
if (acc + key.get() > targetsize && acc != 0) {
long splitsize = last - pos;
splits.add(new FileSplit(src, pos, splitsize, (String[]) null));
cbrem -= splitsize;
pos = last;
acc = 0L;
}
acc += key.get();
}
} finally {
checkAndClose(sl);
}
if (cbrem != 0) {
splits.add(new FileSplit(src, pos, cbrem, (String[]) null));
}
return splits.toArray(new FileSplit[splits.size()]);
}
public JobBuilder setAffinityNoBalancing(Path affinityPath, Set<String> exclusionSet)
throws IOException {
// set node affinity ...
String affinityMask =
NodeAffinityMaskBuilder.buildNodeAffinityMask(
FileSystem.get(_jobConf),
affinityPath,
null,
exclusionSet,
_jobConf.getInt("mapred.tasktracker.reduce.tasks.maximum", -1),
true);
NodeAffinityMaskBuilder.setNodeAffinityMask(_jobConf, affinityMask);
return this;
}
/**
* Mapper configuration. Extracts source and destination file system, as well as top-level paths
* on source and destination directories. Gets the named file systems, to be used later in map.
*/
public void configure(JobConf job) {
destPath = new Path(job.get(DST_DIR_LABEL, "/"));
try {
destFileSys = destPath.getFileSystem(job);
} catch (IOException ex) {
throw new RuntimeException("Unable to get the named file system.", ex);
}
sizeBuf = job.getInt("copy.buf.size", 128 * 1024);
buffer = new byte[sizeBuf];
ignoreReadFailures = job.getBoolean(Options.IGNORE_READ_FAILURES.propertyname, false);
preserve_status = job.getBoolean(Options.PRESERVE_STATUS.propertyname, false);
if (preserve_status) {
preseved = FileAttribute.parse(job.get(PRESERVE_STATUS_LABEL));
}
update = job.getBoolean(Options.UPDATE.propertyname, false);
overwrite = !update && job.getBoolean(Options.OVERWRITE.propertyname, false);
this.job = job;
}
@Override
public void configure(JobConf job) {
super.configure(job);
this.totalBytes = 0;
this.chunkId =
job.getInt(
"mapred.task.partition",
-1); // http://www.mail-archive.com/[email protected]/msg05749.html
if (this.chunkId < 0) {
throw new RuntimeException("Incorrect chunk id ");
}
this.nodeId = this.chunkId / getNumChunks();
this.client = new AdminClient(getCluster(), new AdminClientConfig());
logger.info("Reducer for Node id - " + this.nodeId + " and chunk id - " + this.chunkId);
}
public void configure(JobConf job) {
// 'key' == sortInput for sort-input; key == sortOutput for sort-output
key = deduceInputFile(job);
if (key == sortOutput) {
partitioner = new HashPartitioner<WritableComparable, Writable>();
// Figure the 'current' partition and no. of reduces of the 'sort'
try {
URI inputURI = new URI(job.get("map.input.file"));
String inputFile = inputURI.getPath();
partition =
Integer.valueOf(inputFile.substring(inputFile.lastIndexOf("part") + 5)).intValue();
noSortReducers = job.getInt("sortvalidate.sort.reduce.tasks", -1);
} catch (Exception e) {
System.err.println("Caught: " + e);
System.exit(-1);
}
}
}
/**
* Produce splits such that each is no greater than the quotient of the total size and the
* number of splits requested.
*
* @param job The handle to the JobConf object
* @param numSplits Number of splits requested
*/
public InputSplit[] getSplits(JobConf job, int numSplits) throws IOException {
final int srcCount = job.getInt(OP_COUNT_LABEL, -1);
final int targetcount = srcCount / numSplits;
String srclist = job.get(OP_LIST_LABEL, "");
if (srcCount < 0 || "".equals(srclist)) {
throw new RuntimeException(
"Invalid metadata: #files(" + srcCount + ") listuri(" + srclist + ")");
}
Path srcs = new Path(srclist);
FileSystem fs = srcs.getFileSystem(job);
List<FileSplit> splits = new ArrayList<FileSplit>(numSplits);
Text key = new Text();
PolicyInfo value = new PolicyInfo();
SequenceFile.Reader in = null;
long prev = 0L;
int count = 0; // count src
try {
for (in = new SequenceFile.Reader(fs, srcs, job); in.next(key, value); ) {
long curr = in.getPosition();
long delta = curr - prev;
if (++count > targetcount) {
count = 0;
splits.add(new FileSplit(srcs, prev, delta, (String[]) null));
prev = curr;
}
}
} finally {
in.close();
}
long remaining = fs.getFileStatus(srcs).getLen() - prev;
if (remaining != 0) {
splits.add(new FileSplit(srcs, prev, remaining, (String[]) null));
}
LOG.info(
"jobname= " + jobName + " numSplits=" + numSplits + ", splits.size()=" + splits.size());
return splits.toArray(new FileSplit[splits.size()]);
}
public void configure(JobConf conf) {
try {
Path vInput;
FileSystem fs;
URI[] fvector;
nsize = conf.getInt("DIMENTION", 0);
sumVec = new double[nsize];
resVec = new double[nsize];
diaVec = new double[nsize];
Arrays.fill(sumVec, 0);
Arrays.fill(resVec, 0);
Arrays.fill(diaVec, 0);
fvector = DistributedCache.getCacheFiles(conf);
vInput = new Path(fvector[0].getPath());
fs = FileSystem.get(URI.create("hdfs://node17.cs.rochester.edu:9000"), conf);
FSDataInputStream fdis = fs.open(vInput);
String line;
while ((line = fdis.readLine()) != null) {
StringTokenizer tokenizer = new StringTokenizer(line);
int rowIdx = Integer.parseInt(tokenizer.nextToken());
int colIdx = Integer.parseInt(tokenizer.nextToken());
double matVar = Double.parseDouble(tokenizer.nextToken());
if (rowIdx == colIdx) {
diaVec[rowIdx] = matVar;
} else if (colIdx == nsize) {
resVec[rowIdx] = matVar;
} else {
sumVec[rowIdx] += matVar;
}
}
} catch (IOException e) {
e.printStackTrace();
}
}
public void configure(JobConf job) {
_attemptID = TaskAttemptID.forName(job.get("mapred.task.id"));
_maxAttemptsPerTask = job.getInt("mapred.max.tracker.failures", 4);
_splitDetails = job.get(ARCSplitReader.SPLIT_DETAILS, "Spit Details Unknown");
}
public void configure(JobConf job) {
nSamples = job.getInt("nSamples", 100);
ONE.set(1);
}
public void configure(JobConf job) {
mNodeCnt = job.getInt("NodeCount", 0);
}
public int run(String[] args) throws Exception {
if (args.length != 3) {
System.err.println("Usage: " + getClass().getName() + " <input> <output> <nPopulation>");
// ToolRunner.printGenericCommandUsage(System.err);
return -1;
}
// Create a JobConf using the processed <code>conf</code>
final JobConf jobConf = new JobConf(getConf(), getClass());
// Specify various job-specific parameters
jobConf.setJobName(MapreduceStringFinder.class.getSimpleName());
// setting the input format
jobConf.setInputFormat(Individuals.class);
// setting the output ke and value class
jobConf.setOutputKeyClass(Text.class);
jobConf.setOutputValueClass(BooleanWritable.class);
// setting the mapper class
jobConf.setMapperClass(CIMapper.class);
jobConf.setNumMapTasks(3); // setting number of maptasks
// setting the reducer class
jobConf.setReducerClass(CIReducer.class);
// setup input/output directories
final String dataset = args[0];
FileInputFormat.setInputPaths(jobConf, new Path(dataset));
FileOutputFormat.setOutputPath(jobConf, new Path(args[1]));
final int pop = Integer.parseInt(args[2]);
// based on the configuration, make this job threadable
if (jobConf.getInt("mapred.tasktracker.map.tasks.maximum", 2) == 1) {
jobConf.setMapRunnerClass(MultithreadedMapRunner.class);
jobConf.setInt("mapred.map.multithreadedrunner.threads", 100);
}
jobConf.setInt("mapred.map.multithreadedrunner.threads", 100);
// for computation intensive data, do not allow the job to fail if the task tracker does not
// respond
// with a heatbeat message before the timeout value
final int timeout = 9000000;
jobConf.setInt("mapred.task.timeout", timeout);
// set the parameters to be available before a call to the mapper
jobConf.setInt("popsize", pop);
jobConf.setStrings("dataset", dataset);
// int map = jobConf.getNumMapTasks();
// System.out.println("Number of Maps"+ map);
// start the map/reduce job
System.out.println("Starting Job");
// get the start time for this job
final long startTime = System.currentTimeMillis();
// Submit the job, then poll for progress until the job is complete
JobClient.runJob(jobConf);
// get the end time for this job
final long endTime = System.currentTimeMillis();
// get the duration of this job
final double duration = (endTime - startTime) / 1000.0;
// System.out.println("Job Finished in " + duration + " seconds");
// getElapsedTime(startTime - endTime);
return 0;
}
public void configure(JobConf conf) {
delim = conf.get("delim", ",");
keyFieldIndexTwo = conf.getInt("keyFieldIndexTwo", 0);
}
/** Execute a query plan using Hadoop. */
@SuppressWarnings({"deprecation", "unchecked"})
@Override
public int execute(DriverContext driverContext) {
IOPrepareCache ioPrepareCache = IOPrepareCache.get();
ioPrepareCache.clear();
boolean success = true;
Context ctx = driverContext.getCtx();
boolean ctxCreated = false;
Path emptyScratchDir;
MapWork mWork = work.getMapWork();
ReduceWork rWork = work.getReduceWork();
try {
if (ctx == null) {
ctx = new Context(job);
ctxCreated = true;
}
emptyScratchDir = ctx.getMRTmpPath();
FileSystem fs = emptyScratchDir.getFileSystem(job);
fs.mkdirs(emptyScratchDir);
} catch (IOException e) {
e.printStackTrace();
console.printError(
"Error launching map-reduce job",
"\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
return 5;
}
HiveFileFormatUtils.prepareJobOutput(job);
// See the javadoc on HiveOutputFormatImpl and HadoopShims.prepareJobOutput()
job.setOutputFormat(HiveOutputFormatImpl.class);
job.setMapperClass(ExecMapper.class);
job.setMapOutputKeyClass(HiveKey.class);
job.setMapOutputValueClass(BytesWritable.class);
try {
String partitioner = HiveConf.getVar(job, ConfVars.HIVEPARTITIONER);
job.setPartitionerClass(JavaUtils.loadClass(partitioner));
} catch (ClassNotFoundException e) {
throw new RuntimeException(e.getMessage(), e);
}
if (mWork.getNumMapTasks() != null) {
job.setNumMapTasks(mWork.getNumMapTasks().intValue());
}
if (mWork.getMaxSplitSize() != null) {
HiveConf.setLongVar(
job, HiveConf.ConfVars.MAPREDMAXSPLITSIZE, mWork.getMaxSplitSize().longValue());
}
if (mWork.getMinSplitSize() != null) {
HiveConf.setLongVar(
job, HiveConf.ConfVars.MAPREDMINSPLITSIZE, mWork.getMinSplitSize().longValue());
}
if (mWork.getMinSplitSizePerNode() != null) {
HiveConf.setLongVar(
job,
HiveConf.ConfVars.MAPREDMINSPLITSIZEPERNODE,
mWork.getMinSplitSizePerNode().longValue());
}
if (mWork.getMinSplitSizePerRack() != null) {
HiveConf.setLongVar(
job,
HiveConf.ConfVars.MAPREDMINSPLITSIZEPERRACK,
mWork.getMinSplitSizePerRack().longValue());
}
job.setNumReduceTasks(rWork != null ? rWork.getNumReduceTasks().intValue() : 0);
job.setReducerClass(ExecReducer.class);
// set input format information if necessary
setInputAttributes(job);
// Turn on speculative execution for reducers
boolean useSpeculativeExecReducers =
HiveConf.getBoolVar(job, HiveConf.ConfVars.HIVESPECULATIVEEXECREDUCERS);
HiveConf.setBoolVar(
job, HiveConf.ConfVars.HADOOPSPECULATIVEEXECREDUCERS, useSpeculativeExecReducers);
String inpFormat = HiveConf.getVar(job, HiveConf.ConfVars.HIVEINPUTFORMAT);
if (mWork.isUseBucketizedHiveInputFormat()) {
inpFormat = BucketizedHiveInputFormat.class.getName();
}
LOG.info("Using " + inpFormat);
try {
job.setInputFormat(JavaUtils.loadClass(inpFormat));
} catch (ClassNotFoundException e) {
throw new RuntimeException(e.getMessage(), e);
}
// No-Op - we don't really write anything here ..
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(Text.class);
// Transfer HIVEAUXJARS and HIVEADDEDJARS to "tmpjars" so hadoop understands
// it
String auxJars = HiveConf.getVar(job, HiveConf.ConfVars.HIVEAUXJARS);
String addedJars = HiveConf.getVar(job, HiveConf.ConfVars.HIVEADDEDJARS);
if (StringUtils.isNotBlank(auxJars) || StringUtils.isNotBlank(addedJars)) {
String allJars =
StringUtils.isNotBlank(auxJars)
? (StringUtils.isNotBlank(addedJars) ? addedJars + "," + auxJars : auxJars)
: addedJars;
LOG.info("adding libjars: " + allJars);
initializeFiles("tmpjars", allJars);
}
// Transfer HIVEADDEDFILES to "tmpfiles" so hadoop understands it
String addedFiles = HiveConf.getVar(job, HiveConf.ConfVars.HIVEADDEDFILES);
if (StringUtils.isNotBlank(addedFiles)) {
initializeFiles("tmpfiles", addedFiles);
}
int returnVal = 0;
boolean noName = StringUtils.isEmpty(HiveConf.getVar(job, HiveConf.ConfVars.HADOOPJOBNAME));
if (noName) {
// This is for a special case to ensure unit tests pass
HiveConf.setVar(job, HiveConf.ConfVars.HADOOPJOBNAME, "JOB" + Utilities.randGen.nextInt());
}
String addedArchives = HiveConf.getVar(job, HiveConf.ConfVars.HIVEADDEDARCHIVES);
// Transfer HIVEADDEDARCHIVES to "tmparchives" so hadoop understands it
if (StringUtils.isNotBlank(addedArchives)) {
initializeFiles("tmparchives", addedArchives);
}
try {
MapredLocalWork localwork = mWork.getMapRedLocalWork();
if (localwork != null && localwork.hasStagedAlias()) {
if (!ShimLoader.getHadoopShims().isLocalMode(job)) {
Path localPath = localwork.getTmpPath();
Path hdfsPath = mWork.getTmpHDFSPath();
FileSystem hdfs = hdfsPath.getFileSystem(job);
FileSystem localFS = localPath.getFileSystem(job);
FileStatus[] hashtableFiles = localFS.listStatus(localPath);
int fileNumber = hashtableFiles.length;
String[] fileNames = new String[fileNumber];
for (int i = 0; i < fileNumber; i++) {
fileNames[i] = hashtableFiles[i].getPath().getName();
}
// package and compress all the hashtable files to an archive file
String stageId = this.getId();
String archiveFileName = Utilities.generateTarFileName(stageId);
localwork.setStageID(stageId);
CompressionUtils.tar(localPath.toUri().getPath(), fileNames, archiveFileName);
Path archivePath = Utilities.generateTarPath(localPath, stageId);
LOG.info("Archive " + hashtableFiles.length + " hash table files to " + archivePath);
// upload archive file to hdfs
Path hdfsFilePath = Utilities.generateTarPath(hdfsPath, stageId);
short replication = (short) job.getInt("mapred.submit.replication", 10);
hdfs.copyFromLocalFile(archivePath, hdfsFilePath);
hdfs.setReplication(hdfsFilePath, replication);
LOG.info("Upload 1 archive file from" + archivePath + " to: " + hdfsFilePath);
// add the archive file to distributed cache
DistributedCache.createSymlink(job);
DistributedCache.addCacheArchive(hdfsFilePath.toUri(), job);
LOG.info(
"Add 1 archive file to distributed cache. Archive file: " + hdfsFilePath.toUri());
}
}
work.configureJobConf(job);
List<Path> inputPaths = Utilities.getInputPaths(job, mWork, emptyScratchDir, ctx, false);
Utilities.setInputPaths(job, inputPaths);
Utilities.setMapRedWork(job, work, ctx.getMRTmpPath());
if (mWork.getSamplingType() > 0 && rWork != null && job.getNumReduceTasks() > 1) {
try {
handleSampling(ctx, mWork, job);
job.setPartitionerClass(HiveTotalOrderPartitioner.class);
} catch (IllegalStateException e) {
console.printInfo("Not enough sampling data.. Rolling back to single reducer task");
rWork.setNumReduceTasks(1);
job.setNumReduceTasks(1);
} catch (Exception e) {
LOG.error("Sampling error", e);
console.printError(
e.toString(), "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
rWork.setNumReduceTasks(1);
job.setNumReduceTasks(1);
}
}
// remove the pwd from conf file so that job tracker doesn't show this
// logs
String pwd = HiveConf.getVar(job, HiveConf.ConfVars.METASTOREPWD);
if (pwd != null) {
HiveConf.setVar(job, HiveConf.ConfVars.METASTOREPWD, "HIVE");
}
JobClient jc = new JobClient(job);
// make this client wait if job tracker is not behaving well.
Throttle.checkJobTracker(job, LOG);
if (mWork.isGatheringStats() || (rWork != null && rWork.isGatheringStats())) {
// initialize stats publishing table
StatsPublisher statsPublisher;
StatsFactory factory = StatsFactory.newFactory(job);
if (factory != null) {
statsPublisher = factory.getStatsPublisher();
List<String> statsTmpDir = Utilities.getStatsTmpDirs(mWork, job);
if (rWork != null) {
statsTmpDir.addAll(Utilities.getStatsTmpDirs(rWork, job));
}
StatsCollectionContext sc = new StatsCollectionContext(job);
sc.setStatsTmpDirs(statsTmpDir);
if (!statsPublisher.init(sc)) { // creating stats table if not exists
if (HiveConf.getBoolVar(job, HiveConf.ConfVars.HIVE_STATS_RELIABLE)) {
throw new HiveException(
ErrorMsg.STATSPUBLISHER_INITIALIZATION_ERROR.getErrorCodedMsg());
}
}
}
}
Utilities.createTmpDirs(job, mWork);
Utilities.createTmpDirs(job, rWork);
SessionState ss = SessionState.get();
if (HiveConf.getVar(job, HiveConf.ConfVars.HIVE_EXECUTION_ENGINE).equals("tez")
&& ss != null) {
TezSessionState session = ss.getTezSession();
TezSessionPoolManager.getInstance().close(session, true);
}
// Finally SUBMIT the JOB!
rj = jc.submitJob(job);
// replace it back
if (pwd != null) {
HiveConf.setVar(job, HiveConf.ConfVars.METASTOREPWD, pwd);
}
returnVal = jobExecHelper.progress(rj, jc, ctx.getHiveTxnManager());
success = (returnVal == 0);
} catch (Exception e) {
e.printStackTrace();
String mesg = " with exception '" + Utilities.getNameMessage(e) + "'";
if (rj != null) {
mesg = "Ended Job = " + rj.getJobID() + mesg;
} else {
mesg = "Job Submission failed" + mesg;
}
// Has to use full name to make sure it does not conflict with
// org.apache.commons.lang.StringUtils
console.printError(mesg, "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
success = false;
returnVal = 1;
} finally {
Utilities.clearWork(job);
try {
if (ctxCreated) {
ctx.clear();
}
if (rj != null) {
if (returnVal != 0) {
rj.killJob();
}
jobID = rj.getID().toString();
}
} catch (Exception e) {
LOG.warn("Failed while cleaning up ", e);
} finally {
HadoopJobExecHelper.runningJobs.remove(rj);
}
}
// get the list of Dynamic partition paths
try {
if (rj != null) {
if (mWork.getAliasToWork() != null) {
for (Operator<? extends OperatorDesc> op : mWork.getAliasToWork().values()) {
op.jobClose(job, success);
}
}
if (rWork != null) {
rWork.getReducer().jobClose(job, success);
}
}
} catch (Exception e) {
// jobClose needs to execute successfully otherwise fail task
if (success) {
success = false;
returnVal = 3;
String mesg = "Job Commit failed with exception '" + Utilities.getNameMessage(e) + "'";
console.printError(mesg, "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
}
}
return (returnVal);
}
public void configure(JobConf job) {
int n = job.getInt("NodeCnt", 0);
node.setType(PageRankNode.TYPE_COMPLETE);
node.setPageRank((float) -StrictMath.log(n));
}
static int getMaxConcurrentSteps(JobConf jobConf) {
return jobConf.getInt(MAX_CONCURRENT_STEPS, 0);
}
public void configure(JobConf job) {
memoryMXBean = ManagementFactory.getMemoryMXBean();
l4j.info("maximum memory = " + memoryMXBean.getHeapMemoryUsage().getMax());
try {
l4j.info(
"conf classpath = " + Arrays.asList(((URLClassLoader) job.getClassLoader()).getURLs()));
l4j.info(
"thread classpath = "
+ Arrays.asList(
((URLClassLoader) Thread.currentThread().getContextClassLoader()).getURLs()));
} catch (Exception e) {
l4j.info("cannot get classpath: " + e.getMessage());
}
try {
jc = job;
int estCountBucketSize = jc.getInt("hive.exec.estdistinct.bucketsize.log", 15);
int estCountBufferSize = jc.getInt("hive.exec.estdistinct.buffsize.log", 8);
GenericUDAFCardinalityEstimation.initParams(estCountBucketSize, estCountBufferSize);
boolean isChangSizeZero2Null =
jc.getBoolean(
HiveConf.ConfVars.HIVE_UDTF_EXPLODE_CHANGE_ZERO_SIZE_2_NULL.varname,
HiveConf.ConfVars.HIVE_UDTF_EXPLODE_CHANGE_ZERO_SIZE_2_NULL.defaultBoolVal);
boolean isChangeNull2Null =
jc.getBoolean(
HiveConf.ConfVars.HIVE_UDTF_EXPLODE_CHANGE_NULL_2_NULL.varname,
HiveConf.ConfVars.HIVE_UDTF_EXPLODE_CHANGE_NULL_2_NULL.defaultBoolVal);
GenericUDTFExplode.isChangSizeZero2Null = isChangSizeZero2Null;
GenericUDTFExplode.isChangNull2Null = isChangeNull2Null;
GenericUDTFPosExplode.isChangSizeZero2Null = isChangSizeZero2Null;
GenericUDTFPosExplode.isChangNull2Null = isChangeNull2Null;
mapredWork mrwork = Utilities.getMapRedWork(job);
mo = new MapOperator();
mo.setConf(mrwork);
mo.setChildren(job);
l4j.info(mo.dump(0));
mo.initialize(jc, null);
localWork = mrwork.getMapLocalWork();
if (localWork == null) {
return;
}
fetchOperators = new HashMap<String, FetchOperator>();
for (Map.Entry<String, fetchWork> entry : localWork.getAliasToFetchWork().entrySet()) {
fetchOperators.put(entry.getKey(), new FetchOperator(entry.getValue(), job));
l4j.info("fetchoperator for " + entry.getKey() + " created");
}
for (Map.Entry<String, FetchOperator> entry : fetchOperators.entrySet()) {
Operator<? extends Serializable> forwardOp = localWork.getAliasToWork().get(entry.getKey());
forwardOp.initialize(
jc, new ObjectInspector[] {entry.getValue().getOutputObjectInspector()});
l4j.info("fetchoperator for " + entry.getKey() + " initialized");
}
} catch (Throwable e) {
abort = true;
if (e instanceof OutOfMemoryError) {
throw (OutOfMemoryError) e;
} else {
throw new RuntimeException("Map operator initialization failed", e);
}
}
}
public MergeManagerImpl(
TaskAttemptID reduceId,
JobConf jobConf,
FileSystem localFS,
LocalDirAllocator localDirAllocator,
Reporter reporter,
CompressionCodec codec,
Class<? extends Reducer> combinerClass,
CombineOutputCollector<K, V> combineCollector,
Counters.Counter spilledRecordsCounter,
Counters.Counter reduceCombineInputCounter,
Counters.Counter mergedMapOutputsCounter,
ExceptionReporter exceptionReporter,
Progress mergePhase,
MapOutputFile mapOutputFile) {
this.reduceId = reduceId;
this.jobConf = jobConf;
this.localDirAllocator = localDirAllocator;
this.exceptionReporter = exceptionReporter;
this.reporter = reporter;
this.codec = codec;
this.combinerClass = combinerClass;
this.combineCollector = combineCollector;
this.reduceCombineInputCounter = reduceCombineInputCounter;
this.spilledRecordsCounter = spilledRecordsCounter;
this.mergedMapOutputsCounter = mergedMapOutputsCounter;
this.mapOutputFile = mapOutputFile;
this.mapOutputFile.setConf(jobConf);
this.localFS = localFS;
this.rfs = ((LocalFileSystem) localFS).getRaw();
final float maxInMemCopyUse = jobConf.getFloat(MRJobConfig.SHUFFLE_INPUT_BUFFER_PERCENT, 0.90f);
if (maxInMemCopyUse > 1.0 || maxInMemCopyUse < 0.0) {
throw new IllegalArgumentException(
"Invalid value for " + MRJobConfig.SHUFFLE_INPUT_BUFFER_PERCENT + ": " + maxInMemCopyUse);
}
// Allow unit tests to fix Runtime memory
this.memoryLimit =
(long)
(jobConf.getLong(
MRJobConfig.REDUCE_MEMORY_TOTAL_BYTES,
Math.min(Runtime.getRuntime().maxMemory(), Integer.MAX_VALUE))
* maxInMemCopyUse);
this.ioSortFactor = jobConf.getInt(MRJobConfig.IO_SORT_FACTOR, 100);
final float singleShuffleMemoryLimitPercent =
jobConf.getFloat(
MRJobConfig.SHUFFLE_MEMORY_LIMIT_PERCENT, DEFAULT_SHUFFLE_MEMORY_LIMIT_PERCENT);
if (singleShuffleMemoryLimitPercent <= 0.0f || singleShuffleMemoryLimitPercent > 1.0f) {
throw new IllegalArgumentException(
"Invalid value for "
+ MRJobConfig.SHUFFLE_MEMORY_LIMIT_PERCENT
+ ": "
+ singleShuffleMemoryLimitPercent);
}
usedMemory = 0L;
commitMemory = 0L;
this.maxSingleShuffleLimit = (long) (memoryLimit * singleShuffleMemoryLimitPercent);
this.memToMemMergeOutputsThreshold =
jobConf.getInt(MRJobConfig.REDUCE_MEMTOMEM_THRESHOLD, ioSortFactor);
this.mergeThreshold =
(long) (this.memoryLimit * jobConf.getFloat(MRJobConfig.SHUFFLE_MERGE_PERCENT, 0.90f));
LOG.info(
"MergerManager: memoryLimit="
+ memoryLimit
+ ", "
+ "maxSingleShuffleLimit="
+ maxSingleShuffleLimit
+ ", "
+ "mergeThreshold="
+ mergeThreshold
+ ", "
+ "ioSortFactor="
+ ioSortFactor
+ ", "
+ "memToMemMergeOutputsThreshold="
+ memToMemMergeOutputsThreshold);
if (this.maxSingleShuffleLimit >= this.mergeThreshold) {
throw new RuntimeException(
"Invlaid configuration: "
+ "maxSingleShuffleLimit should be less than mergeThreshold"
+ "maxSingleShuffleLimit: "
+ this.maxSingleShuffleLimit
+ "mergeThreshold: "
+ this.mergeThreshold);
}
boolean allowMemToMemMerge = jobConf.getBoolean(MRJobConfig.REDUCE_MEMTOMEM_ENABLED, false);
if (allowMemToMemMerge) {
this.memToMemMerger =
new IntermediateMemoryToMemoryMerger(this, memToMemMergeOutputsThreshold);
this.memToMemMerger.start();
} else {
this.memToMemMerger = null;
}
this.inMemoryMerger = createInMemoryMerger();
this.inMemoryMerger.start();
this.onDiskMerger = new OnDiskMerger(this);
this.onDiskMerger.start();
this.mergePhase = mergePhase;
}
@Override
public void configure(JobConf conf) {
nstripes = conf.getInt(PARAM_APS_STRIPES, 1);
}
public void configure(JobConf job) {
// get buckets per crawler ...
bucketsPerCrawler = job.getInt(CrawlEnvironment.PROPERTY_NUM_BUCKETS_PER_CRAWLER, 8);
}
/**
* Copy a file to a destination.
*
* @param srcstat src path and metadata
* @param dstpath dst path
* @param reporter
*/
private void copy(
FileStatus srcstat,
Path relativedst,
OutputCollector<WritableComparable<?>, Text> outc,
Reporter reporter)
throws IOException {
Path absdst = new Path(destPath, relativedst);
int totfiles = job.getInt(SRC_COUNT_LABEL, -1);
assert totfiles >= 0 : "Invalid file count " + totfiles;
// if a directory, ensure created even if empty
if (srcstat.isDir()) {
if (destFileSys.exists(absdst)) {
if (!destFileSys.getFileStatus(absdst).isDir()) {
throw new IOException("Failed to mkdirs: " + absdst + " is a file.");
}
} else if (!destFileSys.mkdirs(absdst)) {
throw new IOException("Failed to mkdirs " + absdst);
}
// TODO: when modification times can be set, directories should be
// emitted to reducers so they might be preserved. Also, mkdirs does
// not currently return an error when the directory already exists;
// if this changes, all directory work might as well be done in reduce
return;
}
if (destFileSys.exists(absdst) && !overwrite && !needsUpdate(srcstat, destFileSys, absdst)) {
outc.collect(null, new Text("SKIP: " + srcstat.getPath()));
++skipcount;
reporter.incrCounter(Counter.SKIP, 1);
updateStatus(reporter);
return;
}
Path tmpfile = new Path(job.get(TMP_DIR_LABEL), relativedst);
long cbcopied = 0L;
FSDataInputStream in = null;
FSDataOutputStream out = null;
try {
// open src file
try {
in = srcstat.getPath().getFileSystem(job).open(srcstat.getPath());
} catch (IOException e) {
LOG.error("Failed to open src file " + srcstat.getPath() + ", ignore and return");
in = null;
return;
}
reporter.incrCounter(Counter.BYTESEXPECTED, srcstat.getLen());
// open tmp file
out = create(tmpfile, reporter, srcstat);
// copy file
for (int cbread; (cbread = in.read(buffer)) >= 0; ) {
out.write(buffer, 0, cbread);
cbcopied += cbread;
reporter.setStatus(
String.format("%.2f ", cbcopied * 100.0 / srcstat.getLen())
+ absdst
+ " [ "
+ StringUtils.humanReadableInt(cbcopied)
+ " / "
+ StringUtils.humanReadableInt(srcstat.getLen())
+ " ]");
}
} finally {
checkAndClose(in);
checkAndClose(out);
}
if (cbcopied != srcstat.getLen()) {
if (srcstat.getLen() == 0 && cbcopied > 0) {
LOG.info("most likely see a WAL file corruption: " + srcstat.getPath());
} else {
throw new IOException(
"File size not matched: copied "
+ bytesString(cbcopied)
+ " to tmpfile (="
+ tmpfile
+ ") but expected "
+ bytesString(srcstat.getLen())
+ " from "
+ srcstat.getPath());
}
} else {
if (totfiles == 1) {
// Copying a single file; use dst path provided by user as destination
// rather than destination directory, if a file
Path dstparent = absdst.getParent();
if (!(destFileSys.exists(dstparent) && destFileSys.getFileStatus(dstparent).isDir())) {
absdst = dstparent;
}
}
if (destFileSys.exists(absdst) && destFileSys.getFileStatus(absdst).isDir()) {
throw new IOException(absdst + " is a directory");
}
if (!destFileSys.mkdirs(absdst.getParent())) {
throw new IOException("Failed to craete parent dir: " + absdst.getParent());
}
rename(tmpfile, absdst);
FileStatus dststat = destFileSys.getFileStatus(absdst);
if (dststat.getLen() != srcstat.getLen()) {
destFileSys.delete(absdst, false);
throw new IOException(
"File size not matched: copied "
+ bytesString(dststat.getLen())
+ " to dst (="
+ absdst
+ ") but expected "
+ bytesString(srcstat.getLen())
+ " from "
+ srcstat.getPath());
}
updatePermissions(srcstat, dststat);
}
// report at least once for each file
++copycount;
reporter.incrCounter(Counter.BYTESCOPIED, cbcopied);
reporter.incrCounter(Counter.COPY, 1);
updateStatus(reporter);
}
