/*
* 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.pig.backend.hadoop.executionengine.mapReduceLayer;
import java.io.File;
import java.io.FileOutputStream;
import java.io.IOException;
import java.net.URI;
import java.net.URISyntaxException;
import java.util.ArrayList;
import java.util.Collections;
import java.util.HashMap;
import java.util.LinkedList;
import java.util.List;
import java.util.Map;
import java.util.Properties;
import org.apache.commons.logging.Log;
import org.apache.commons.logging.LogFactory;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.filecache.DistributedCache;
import org.apache.hadoop.fs.FileStatus;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.WritableComparable;
import org.apache.hadoop.io.WritableComparator;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapred.JobPriority;
import org.apache.hadoop.mapred.jobcontrol.Job;
import org.apache.hadoop.mapred.jobcontrol.JobControl;
import org.apache.hadoop.util.RunJar;
import org.apache.pig.ComparisonFunc;
import org.apache.pig.ExecType;
import org.apache.pig.FuncSpec;
import org.apache.pig.LoadFunc;
import org.apache.pig.PigException;
import org.apache.pig.StoreFuncInterface;
import org.apache.pig.backend.executionengine.ExecException;
import org.apache.pig.backend.hadoop.HDataType;
import org.apache.pig.backend.hadoop.executionengine.mapReduceLayer.partitioners.SecondaryKeyPartitioner;
import org.apache.pig.backend.hadoop.executionengine.mapReduceLayer.partitioners.SkewedPartitioner;
import org.apache.pig.backend.hadoop.executionengine.mapReduceLayer.partitioners.WeightedRangePartitioner;
import org.apache.pig.backend.hadoop.executionengine.mapReduceLayer.plans.MROperPlan;
import org.apache.pig.backend.hadoop.executionengine.physicalLayer.PhysicalOperator;
import org.apache.pig.backend.hadoop.executionengine.physicalLayer.plans.PhyPlanVisitor;
import org.apache.pig.backend.hadoop.executionengine.physicalLayer.plans.PhysicalPlan;
import org.apache.pig.backend.hadoop.executionengine.physicalLayer.relationalOperators.POFRJoin;
import org.apache.pig.backend.hadoop.executionengine.physicalLayer.relationalOperators.POLoad;
import org.apache.pig.backend.hadoop.executionengine.physicalLayer.relationalOperators.POMergeCogroup;
import org.apache.pig.backend.hadoop.executionengine.physicalLayer.relationalOperators.POMergeJoin;
import org.apache.pig.backend.hadoop.executionengine.physicalLayer.relationalOperators.POPackage;
import org.apache.pig.backend.hadoop.executionengine.physicalLayer.relationalOperators.POStore;
import org.apache.pig.backend.hadoop.executionengine.physicalLayer.util.PlanHelper;
import org.apache.pig.data.BagFactory;
import org.apache.pig.data.DataType;
import org.apache.pig.data.Tuple;
import org.apache.pig.data.TupleFactory;
import org.apache.pig.impl.PigContext;
import org.apache.pig.impl.io.FileLocalizer;
import org.apache.pig.impl.io.FileSpec;
import org.apache.pig.impl.io.NullableBytesWritable;
import org.apache.pig.impl.io.NullableDoubleWritable;
import org.apache.pig.impl.io.NullableFloatWritable;
import org.apache.pig.impl.io.NullableIntWritable;
import org.apache.pig.impl.io.NullableLongWritable;
import org.apache.pig.impl.io.NullablePartitionWritable;
import org.apache.pig.impl.io.NullableText;
import org.apache.pig.impl.io.NullableTuple;
import org.apache.pig.impl.io.PigNullableWritable;
import org.apache.pig.impl.plan.DepthFirstWalker;
import org.apache.pig.impl.plan.OperatorKey;
import org.apache.pig.impl.plan.VisitorException;
import org.apache.pig.impl.util.JarManager;
import org.apache.pig.impl.util.ObjectSerializer;
import org.apache.pig.impl.util.Pair;
import org.apache.pig.impl.util.UDFContext;
import org.apache.pig.impl.util.UriUtil;
import org.apache.pig.impl.util.Utils;
import org.apache.pig.tools.pigstats.ScriptState;
/**
* This is compiler class that takes an MROperPlan and converts
* it into a JobControl object with the relevant dependency info
* maintained. The JobControl Object is made up of Jobs each of
* which has a JobConf. The MapReduceOper corresponds to a Job
* and the getJobCong method returns the JobConf that is configured
* as per the MapReduceOper
*
* <h2>Comparator Design</h2>
* <p>
* A few words on how comparators are chosen. In almost all cases we use raw
* comparators (the one exception being when the user provides a comparison
* function for order by). For order by queries the PigTYPERawComparator
* functions are used, where TYPE is Int, Long, etc. These comparators are
* null aware and asc/desc aware. The first byte of each of the
* NullableTYPEWritable classes contains info on whether the value is null.
* Asc/desc is written as an array into the JobConf with the key pig.sortOrder
* so that it can be read by each of the comparators as part of their
* setConf call.
* <p>
* For non-order by queries, PigTYPEWritableComparator classes are used.
* These are all just type specific instances of WritableComparator.
*
*/
public class JobControlCompiler{
MROperPlan plan;
Configuration conf;
PigContext pigContext;
private static final Log log = LogFactory.getLog(JobControlCompiler.class);
public static final String LOG_DIR = "_logs";
public static final String END_OF_INP_IN_MAP = "pig.invoke.close.in.map";
/**
* We will serialize the POStore(s) present in map and reduce in lists in
* the Hadoop Conf. In the case of Multi stores, we could deduce these from
* the map plan and reduce plan but in the case of single store, we remove
* the POStore from the plan - in either case, we serialize the POStore(s)
* so that PigOutputFormat and PigOutputCommiter can get the POStore(s) in
* the same way irrespective of whether it is multi store or single store.
*/
public static final String PIG_MAP_STORES = "pig.map.stores";
public static final String PIG_REDUCE_STORES = "pig.reduce.stores";
// A mapping of job to pair of store locations and tmp locations for that job
private Map<Job, Pair<List<POStore>, Path>> jobStoreMap;
private Map<Job, MapReduceOper> jobMroMap;
public JobControlCompiler(PigContext pigContext, Configuration conf) throws IOException {
this.pigContext = pigContext;
this.conf = conf;
jobStoreMap = new HashMap<Job, Pair<List<POStore>, Path>>();
jobMroMap = new HashMap<Job, MapReduceOper>();
}
/**
* Returns all store locations of a previously compiled job
*/
public List<POStore> getStores(Job job) {
Pair<List<POStore>, Path> pair = jobStoreMap.get(job);
if (pair != null && pair.first != null) {
return pair.first;
} else {
return new ArrayList<POStore>();
}
}
/**
* Resets the state
*/
public void reset() {
jobStoreMap = new HashMap<Job, Pair<List<POStore>, Path>>();
jobMroMap = new HashMap<Job, MapReduceOper>();
UDFContext.getUDFContext().reset();
}
Map<Job, MapReduceOper> getJobMroMap() {
return Collections.unmodifiableMap(jobMroMap);
}
/**
* Moves all the results of a collection of MR jobs to the final
* output directory. Some of the results may have been put into a
* temp location to work around restrictions with multiple output
* from a single map reduce job.
*
* This method should always be called after the job execution
* completes.
*/
public void moveResults(List<Job> completedJobs) throws IOException {
for (Job job: completedJobs) {
Pair<List<POStore>, Path> pair = jobStoreMap.get(job);
if (pair != null && pair.second != null) {
Path tmp = pair.second;
Path abs = new Path(tmp, "abs");
Path rel = new Path(tmp, "rel");
FileSystem fs = tmp.getFileSystem(conf);
if (fs.exists(abs)) {
moveResults(abs, abs.toUri().getPath(), fs);
}
if (fs.exists(rel)) {
moveResults(rel, rel.toUri().getPath()+"/", fs);
}
}
}
}
/**
* Walks the temporary directory structure to move (rename) files
* to their final location.
*/
private void moveResults(Path p, String rem, FileSystem fs) throws IOException {
for (FileStatus fstat: fs.listStatus(p)) {
Path src = fstat.getPath();
if (fstat.isDir()) {
log.info("mkdir: "+src);
fs.mkdirs(removePart(src, rem));
moveResults(fstat.getPath(), rem, fs);
} else {
Path dst = removePart(src, rem);
log.info("mv: "+src+" "+dst);
fs.rename(src,dst);
}
}
}
private Path removePart(Path src, String part) {
URI uri = src.toUri();
String pathStr = uri.getPath().replace(part, "");
return new Path(pathStr);
}
/**
* Compiles all jobs that have no dependencies removes them from
* the plan and returns. Should be called with the same plan until
* exhausted.
* @param plan - The MROperPlan to be compiled
* @param grpName - The name given to the JobControl
* @return JobControl object - null if no more jobs in plan
* @throws JobCreationException
*/
public JobControl compile(MROperPlan plan, String grpName) throws JobCreationException{
// Assert plan.size() != 0
this.plan = plan;
JobControl jobCtrl = new JobControl(grpName);
try {
List<MapReduceOper> roots = new LinkedList<MapReduceOper>();
roots.addAll(plan.getRoots());
for (MapReduceOper mro: roots) {
if(mro instanceof NativeMapReduceOper) {
return null;
}
Job job = getJob(mro, conf, pigContext);
jobMroMap.put(job, mro);
jobCtrl.addJob(job);
}
} catch (JobCreationException jce) {
throw jce;
} catch(Exception e) {
int errCode = 2017;
String msg = "Internal error creating job configuration.";
throw new JobCreationException(msg, errCode, PigException.BUG, e);
}
return jobCtrl;
}
// Update Map-Reduce plan with the execution status of the jobs. If one job
// completely fail (the job has only one store and that job fail), then we
// remove all its dependent jobs. This method will return the number of MapReduceOper
// removed from the Map-Reduce plan
public int updateMROpPlan(List<Job> completeFailedJobs)
{
int sizeBefore = plan.size();
for (Job job : completeFailedJobs) // remove all subsequent jobs
{
MapReduceOper mrOper = jobMroMap.get(job);
plan.trimBelow(mrOper);
plan.remove(mrOper);
}
// Remove successful jobs from jobMroMap
for (Job job : jobMroMap.keySet())
{
if (!completeFailedJobs.contains(job))
{
MapReduceOper mro = jobMroMap.get(job);
plan.remove(mro);
}
}
jobMroMap.clear();
int sizeAfter = plan.size();
return sizeBefore-sizeAfter;
}
/**
* The method that creates the Job corresponding to a MapReduceOper.
* The assumption is that
* every MapReduceOper will have a load and a store. The JobConf removes
* the load operator and serializes the input filespec so that PigInputFormat can
* take over the creation of splits. It also removes the store operator
* and serializes the output filespec so that PigOutputFormat can take over
* record writing. The remaining portion of the map plan and reduce plans are
* serialized and stored for the PigMapReduce or PigMapOnly objects to take over
* the actual running of the plans.
* The Mapper & Reducer classes and the required key value formats are set.
* Checks if this is a map only job and uses PigMapOnly class as the mapper
* and uses PigMapReduce otherwise.
* If it is a Map Reduce job, it is bound to have a package operator. Remove it from
* the reduce plan and serializes it so that the PigMapReduce class can use it to package
* the indexed tuples received by the reducer.
* @param mro - The MapReduceOper for which the JobConf is required
* @param conf - the Configuration object from which JobConf is built
* @param pigContext - The PigContext passed on from execution engine
* @return Job corresponding to mro
* @throws JobCreationException
*/
@SuppressWarnings({ "unchecked", "deprecation" })
private Job getJob(MapReduceOper mro, Configuration config, PigContext pigContext) throws JobCreationException{
org.apache.hadoop.mapreduce.Job nwJob = null;
try{
nwJob = new org.apache.hadoop.mapreduce.Job(config);
}catch(Exception e) {
throw new JobCreationException(e);
}
Configuration conf = nwJob.getConfiguration();
ArrayList<FileSpec> inp = new ArrayList<FileSpec>();
ArrayList<List<OperatorKey>> inpTargets = new ArrayList<List<OperatorKey>>();
ArrayList<String> inpSignatureLists = new ArrayList<String>();
ArrayList<POStore> storeLocations = new ArrayList<POStore>();
Path tmpLocation = null;
// add settings for pig statistics
String setScriptProp = conf.get(ScriptState.INSERT_ENABLED, "true");
if (setScriptProp.equalsIgnoreCase("true")) {
ScriptState ss = ScriptState.get();
ss.addSettingsToConf(mro, conf);
}
conf.set("mapred.mapper.new-api", "true");
conf.set("mapred.reducer.new-api", "true");
String buffPercent = conf.get("mapred.job.reduce.markreset.buffer.percent");
if (buffPercent == null || Double.parseDouble(buffPercent) <= 0) {
log.info("mapred.job.reduce.markreset.buffer.percent is not set, set to default 0.3");
conf.set("mapred.job.reduce.markreset.buffer.percent", "0.3");
}else{
log.info("mapred.job.reduce.markreset.buffer.percent is set to " + conf.get("mapred.job.reduce.markreset.buffer.percent"));
}
try{
//Process the POLoads
List<POLoad> lds = PlanHelper.getLoads(mro.mapPlan);
if(lds!=null && lds.size()>0){
for (POLoad ld : lds) {
LoadFunc lf = ld.getLoadFunc();
// Call setLocation as a hacky way of letting a LoadFunc fix up the Job.
// Note that setLocation will get called on the loadFuncs later, as well.
// That's ok as setLocation getting called multiple times is documented behavior.
if (lf !=null) {
lf.setLocation(ld.getLFile().getFileName(), nwJob);
}
//Store the inp filespecs
inp.add(ld.getLFile());
//Store the target operators for tuples read
//from this input
List<PhysicalOperator> ldSucs = mro.mapPlan.getSuccessors(ld);
List<OperatorKey> ldSucKeys = new ArrayList<OperatorKey>();
if(ldSucs!=null){
for (PhysicalOperator operator2 : ldSucs) {
ldSucKeys.add(operator2.getOperatorKey());
}
}
inpTargets.add(ldSucKeys);
inpSignatureLists.add(ld.getSignature());
//Remove the POLoad from the plan
mro.mapPlan.remove(ld);
}
}
//Create the jar of all functions and classes required
File submitJarFile = File.createTempFile("Job", ".jar");
// ensure the job jar is deleted on exit
submitJarFile.deleteOnExit();
FileOutputStream fos = new FileOutputStream(submitJarFile);
JarManager.createJar(fos, mro.UDFs, pigContext);
//Start setting the JobConf properties
conf.set("mapred.jar", submitJarFile.getPath());
conf.set("pig.inputs", ObjectSerializer.serialize(inp));
conf.set("pig.inpTargets", ObjectSerializer.serialize(inpTargets));
conf.set("pig.inpSignatures", ObjectSerializer.serialize(inpSignatureLists));
conf.set("pig.pigContext", ObjectSerializer.serialize(pigContext));
conf.set("udf.import.list", ObjectSerializer.serialize(PigContext.getPackageImportList()));
// this is for unit tests since some don't create PigServer
// if user specified the job name using -D switch, Pig won't reset the name then.
if (System.getProperty("mapred.job.name") == null &&
pigContext.getProperties().getProperty(PigContext.JOB_NAME) != null){
nwJob.setJobName(pigContext.getProperties().getProperty(PigContext.JOB_NAME));
}
if (pigContext.getProperties().getProperty(PigContext.JOB_PRIORITY) != null) {
// If the job priority was set, attempt to get the corresponding enum value
// and set the hadoop job priority.
String jobPriority = pigContext.getProperties().getProperty(PigContext.JOB_PRIORITY).toUpperCase();
try {
// Allow arbitrary case; the Hadoop job priorities are all upper case.
conf.set("mapred.job.priority", JobPriority.valueOf(jobPriority).toString());
} catch (IllegalArgumentException e) {
StringBuffer sb = new StringBuffer("The job priority must be one of [");
JobPriority[] priorities = JobPriority.values();
for (int i = 0; i < priorities.length; ++i) {
if (i > 0) sb.append(", ");
sb.append(priorities[i]);
}
sb.append("]. You specified [" + jobPriority + "]");
throw new JobCreationException(sb.toString());
}
}
// Setup the DistributedCache for this job
setupDistributedCache(pigContext, nwJob.getConfiguration(), pigContext.getProperties(),
"pig.streaming.ship.files", true);
setupDistributedCache(pigContext, nwJob.getConfiguration(), pigContext.getProperties(),
"pig.streaming.cache.files", false);
nwJob.setInputFormatClass(PigInputFormat.class);
//Process POStore and remove it from the plan
LinkedList<POStore> mapStores = PlanHelper.getStores(mro.mapPlan);
LinkedList<POStore> reduceStores = PlanHelper.getStores(mro.reducePlan);
for (POStore st: mapStores) {
storeLocations.add(st);
StoreFuncInterface sFunc = st.getStoreFunc();
sFunc.setStoreLocation(st.getSFile().getFileName(), nwJob);
}
for (POStore st: reduceStores) {
storeLocations.add(st);
StoreFuncInterface sFunc = st.getStoreFunc();
sFunc.setStoreLocation(st.getSFile().getFileName(), nwJob);
}
// the OutputFormat we report to Hadoop is always PigOutputFormat
nwJob.setOutputFormatClass(PigOutputFormat.class);
if (mapStores.size() + reduceStores.size() == 1) { // single store case
log.info("Setting up single store job");
POStore st;
if (reduceStores.isEmpty()) {
st = mapStores.get(0);
mro.mapPlan.remove(st);
}
else {
st = reduceStores.get(0);
mro.reducePlan.remove(st);
}
// set out filespecs
String outputPath = st.getSFile().getFileName();
FuncSpec outputFuncSpec = st.getSFile().getFuncSpec();
conf.set("pig.streaming.log.dir",
new Path(outputPath, LOG_DIR).toString());
conf.set("pig.streaming.task.output.dir", outputPath);
}
else { // multi store case
log.info("Setting up multi store job");
String tmpLocationStr = FileLocalizer
.getTemporaryPath(pigContext).toString();
tmpLocation = new Path(tmpLocationStr);
nwJob.setOutputFormatClass(PigOutputFormat.class);
for (POStore sto: storeLocations) {
sto.setMultiStore(true);
}
conf.set("pig.streaming.log.dir",
new Path(tmpLocation, LOG_DIR).toString());
conf.set("pig.streaming.task.output.dir", tmpLocation.toString());
}
// store map key type
// this is needed when the key is null to create
// an appropriate NullableXXXWritable object
conf.set("pig.map.keytype", ObjectSerializer.serialize(new byte[] { mro.mapKeyType }));
// set parent plan in all operators in map and reduce plans
// currently the parent plan is really used only when POStream is present in the plan
new PhyPlanSetter(mro.mapPlan).visit();
new PhyPlanSetter(mro.reducePlan).visit();
// this call modifies the ReplFiles names of POFRJoin operators
// within the MR plans, must be called before the plans are
// serialized
setupDistributedCacheForJoin(mro, pigContext, conf);
POPackage pack = null;
if(mro.reducePlan.isEmpty()){
//MapOnly Job
nwJob.setMapperClass(PigMapOnly.Map.class);
nwJob.setNumReduceTasks(0);
conf.set("pig.mapPlan", ObjectSerializer.serialize(mro.mapPlan));
if(mro.isEndOfAllInputSetInMap()) {
// this is used in Map.close() to decide whether the
// pipeline needs to be rerun one more time in the close()
// The pipeline is rerun if there either was a stream or POMergeJoin
conf.set(END_OF_INP_IN_MAP, "true");
}
}
else{
//Map Reduce Job
//Process the POPackage operator and remove it from the reduce plan
if(!mro.combinePlan.isEmpty()){
POPackage combPack = (POPackage)mro.combinePlan.getRoots().get(0);
mro.combinePlan.remove(combPack);
nwJob.setCombinerClass(PigCombiner.Combine.class);
conf.set("pig.combinePlan", ObjectSerializer.serialize(mro.combinePlan));
conf.set("pig.combine.package", ObjectSerializer.serialize(combPack));
} else if (mro.needsDistinctCombiner()) {
nwJob.setCombinerClass(DistinctCombiner.Combine.class);
log.info("Setting identity combiner class.");
}
pack = (POPackage)mro.reducePlan.getRoots().get(0);
mro.reducePlan.remove(pack);
nwJob.setMapperClass(PigMapReduce.Map.class);
nwJob.setReducerClass(PigMapReduce.Reduce.class);
// first check the PARALLE in query, then check the defaultParallel in PigContext, and last do estimation
if (mro.requestedParallelism > 0)
nwJob.setNumReduceTasks(mro.requestedParallelism);
else if (pigContext.defaultParallel > 0)
conf.set("mapred.reduce.tasks", ""+pigContext.defaultParallel);
else
estimateNumberOfReducers(conf,lds);
if (mro.customPartitioner != null)
nwJob.setPartitionerClass(PigContext.resolveClassName(mro.customPartitioner));
conf.set("pig.mapPlan", ObjectSerializer.serialize(mro.mapPlan));
if(mro.isEndOfAllInputSetInMap()) {
// this is used in Map.close() to decide whether the
// pipeline needs to be rerun one more time in the close()
// The pipeline is rerun only if there was a stream or merge-join.
conf.set(END_OF_INP_IN_MAP, "true");
}
conf.set("pig.reducePlan", ObjectSerializer.serialize(mro.reducePlan));
if(mro.isEndOfAllInputSetInReduce()) {
// this is used in Map.close() to decide whether the
// pipeline needs to be rerun one more time in the close()
// The pipeline is rerun only if there was a stream
conf.set("pig.stream.in.reduce", "true");
}
conf.set("pig.reduce.package", ObjectSerializer.serialize(pack));
conf.set("pig.reduce.key.type", Byte.toString(pack.getKeyType()));
if (mro.getUseSecondaryKey()) {
nwJob.setGroupingComparatorClass(PigSecondaryKeyGroupComparator.class);
nwJob.setPartitionerClass(SecondaryKeyPartitioner.class);
nwJob.setSortComparatorClass(PigSecondaryKeyComparator.class);
nwJob.setOutputKeyClass(NullableTuple.class);
conf.set("pig.secondarySortOrder",
ObjectSerializer.serialize(mro.getSecondarySortOrder()));
}
else
{
Class<? extends WritableComparable> keyClass = HDataType.getWritableComparableTypes(pack.getKeyType()).getClass();
nwJob.setOutputKeyClass(keyClass);
selectComparator(mro, pack.getKeyType(), nwJob);
}
nwJob.setOutputValueClass(NullableTuple.class);
}
if(mro.isGlobalSort() || mro.isLimitAfterSort()){
// Only set the quantiles file and sort partitioner if we're a
// global sort, not for limit after sort.
if (mro.isGlobalSort()) {
String symlink = addSingleFileToDistributedCache(
pigContext, conf, mro.getQuantFile(), "pigsample");
conf.set("pig.quantilesFile", symlink);
nwJob.setPartitionerClass(WeightedRangePartitioner.class);
}
if (mro.isUDFComparatorUsed) {
boolean usercomparator = false;
for (String compFuncSpec : mro.UDFs) {
Class comparator = PigContext.resolveClassName(compFuncSpec);
if(ComparisonFunc.class.isAssignableFrom(comparator)) {
nwJob.setMapperClass(PigMapReduce.MapWithComparator.class);
nwJob.setReducerClass(PigMapReduce.ReduceWithComparator.class);
conf.set("pig.reduce.package", ObjectSerializer.serialize(pack));
conf.set("pig.usercomparator", "true");
nwJob.setOutputKeyClass(NullableTuple.class);
nwJob.setSortComparatorClass(comparator);
usercomparator = true;
break;
}
}
if (!usercomparator) {
String msg = "Internal error. Can't find the UDF comparator";
throw new IOException (msg);
}
} else {
conf.set("pig.sortOrder",
ObjectSerializer.serialize(mro.getSortOrder()));
}
}
if (mro.isSkewedJoin()) {
String symlink = addSingleFileToDistributedCache(pigContext,
conf, mro.getSkewedJoinPartitionFile(), "pigdistkey");
conf.set("pig.keyDistFile", symlink);
nwJob.setPartitionerClass(SkewedPartitioner.class);
nwJob.setMapperClass(PigMapReduce.MapWithPartitionIndex.class);
nwJob.setMapOutputKeyClass(NullablePartitionWritable.class);
nwJob.setGroupingComparatorClass(PigGroupingPartitionWritableComparator.class);
}
// unset inputs for POStore, otherwise, map/reduce plan will be unnecessarily deserialized
for (POStore st: mapStores) { st.setInputs(null); st.setParentPlan(null);}
for (POStore st: reduceStores) { st.setInputs(null); st.setParentPlan(null);}
// tmp file compression setups
if (Utils.tmpFileCompression(pigContext)) {
conf.setBoolean("pig.tmpfilecompression", true);
conf.set("pig.tmpfilecompression.codec", Utils.tmpFileCompressionCodec(pigContext));
}
conf.set(PIG_MAP_STORES, ObjectSerializer.serialize(mapStores));
conf.set(PIG_REDUCE_STORES, ObjectSerializer.serialize(reduceStores));
String tmp;
long maxCombinedSplitSize = 0;
if (!mro.combineSmallSplits() || pigContext.getProperties().getProperty("pig.splitCombination", "true").equals("false"))
conf.setBoolean("pig.noSplitCombination", true);
else if ((tmp = pigContext.getProperties().getProperty("pig.maxCombinedSplitSize", null)) != null) {
try {
maxCombinedSplitSize = Long.parseLong(tmp);
} catch (NumberFormatException e) {
log.warn("Invalid numeric format for pig.maxCombinedSplitSize; use the default maximum combined split size");
}
}
if (maxCombinedSplitSize > 0)
conf.setLong("pig.maxCombinedSplitSize", maxCombinedSplitSize);
// Serialize the UDF specific context info.
UDFContext.getUDFContext().serialize(conf);
Job cjob = new Job(new JobConf(nwJob.getConfiguration()), new ArrayList());
jobStoreMap.put(cjob,new Pair<List<POStore>, Path>(storeLocations, tmpLocation));
return cjob;
} catch (JobCreationException jce) {
throw jce;
} catch(Exception e) {
int errCode = 2017;
String msg = "Internal error creating job configuration.";
throw new JobCreationException(msg, errCode, PigException.BUG, e);
}
}
/**
* Currently the estimation of reducer number is only applied to HDFS, The estimation is based on the input size of data storage on HDFS.
* Two parameters can been configured for the estimation, one is pig.exec.reducers.max which constrain the maximum number of reducer task (default is 999). The other
* is pig.exec.reducers.bytes.per.reducer(default value is 1000*1000*1000) which means the how much data can been handled for each reducer.
* e.g. the following is your pig script
* a = load '/data/a';
* b = load '/data/b';
* c = join a by $0, b by $0;
* store c into '/tmp';
*
* The size of /data/a is 1000*1000*1000, and size of /data/b is 2*1000*1000*1000.
* Then the estimated reducer number is (1000*1000*1000+2*1000*1000*1000)/(1000*1000*1000)=3
* @param conf
* @param lds
* @throws IOException
*/
static int estimateNumberOfReducers(Configuration conf, List<POLoad> lds) throws IOException {
long bytesPerReducer = conf.getLong("pig.exec.reducers.bytes.per.reducer", (long) (1000 * 1000 * 1000));
int maxReducers = conf.getInt("pig.exec.reducers.max", 999);
long totalInputFileSize = getTotalInputFileSize(conf, lds);
log.info("BytesPerReducer=" + bytesPerReducer + " maxReducers="
+ maxReducers + " totalInputFileSize=" + totalInputFileSize);
int reducers = (int)Math.ceil((totalInputFileSize+0.0) / bytesPerReducer);
reducers = Math.max(1, reducers);
reducers = Math.min(maxReducers, reducers);
conf.setInt("mapred.reduce.tasks", reducers);
log.info("Neither PARALLEL nor default parallelism is set for this job. Setting number of reducers to " + reducers);
return reducers;
}
private static long getTotalInputFileSize(Configuration conf, List<POLoad> lds) throws IOException {
List<String> inputs = new ArrayList<String>();
if(lds!=null && lds.size()>0){
for (POLoad ld : lds) {
inputs.add(ld.getLFile().getFileName());
}
}
long size = 0;
FileSystem fs = FileSystem.get(conf);
for (String input : inputs){
//Using custom uri parsing because 'new Path(location).toUri()' fails
// for some valid uri's (eg jdbc style), and 'new Uri(location)' fails
// for valid hdfs paths that contain curly braces
if(!UriUtil.isHDFSFileOrLocal(input)){
//skip if it is not hdfs or local file
continue;
}
//the input file location might be a list of comma separeated files,
// separate them out
for(String location : LoadFunc.getPathStrings(input)){
if(! UriUtil.isHDFSFileOrLocal(location)){
continue;
}
FileStatus[] status=fs.globStatus(new Path(location));
if (status != null){
for (FileStatus s : status){
size += getPathLength(fs, s);
}
}
}
}
return size;
}
private static long getPathLength(FileSystem fs,FileStatus status) throws IOException{
if (!status.isDir()){
return status.getLen();
}else{
FileStatus[] children = fs.listStatus(status.getPath());
long size=0;
for (FileStatus child : children){
size +=getPathLength(fs, child);
}
return size;
}
}
public static class PigSecondaryKeyGroupComparator extends WritableComparator {
public PigSecondaryKeyGroupComparator() {
// super(TupleFactory.getInstance().tupleClass(), true);
super(NullableTuple.class, true);
}
@SuppressWarnings("unchecked")
@Override
public int compare(WritableComparable a, WritableComparable b)
{
PigNullableWritable wa = (PigNullableWritable)a;
PigNullableWritable wb = (PigNullableWritable)b;
if ((wa.getIndex() & PigNullableWritable.mqFlag) != 0) { // this is a multi-query index
if ((wa.getIndex() & PigNullableWritable.idxSpace) < (wb.getIndex() & PigNullableWritable.idxSpace)) return -1;
else if ((wa.getIndex() & PigNullableWritable.idxSpace) > (wb.getIndex() & PigNullableWritable.idxSpace)) return 1;
// If equal, we fall through
}
// wa and wb are guaranteed to be not null, POLocalRearrange will create a tuple anyway even if main key and secondary key
// are both null; however, main key can be null, we need to check for that using the same logic we have in PigNullableWritable
Object valuea = null;
Object valueb = null;
try {
// Get the main key from compound key
valuea = ((Tuple)wa.getValueAsPigType()).get(0);
valueb = ((Tuple)wb.getValueAsPigType()).get(0);
} catch (ExecException e) {
throw new RuntimeException("Unable to access tuple field", e);
}
if (!wa.isNull() && !wb.isNull()) {
int result = DataType.compare(valuea, valueb);
// If any of the field inside tuple is null, then we do not merge keys
// See PIG-927
if (result == 0 && valuea instanceof Tuple && valueb instanceof Tuple)
{
try {
for (int i=0;i<((Tuple)valuea).size();i++)
if (((Tuple)valueb).get(i)==null)
return (wa.getIndex()&PigNullableWritable.idxSpace) - (wb.getIndex()&PigNullableWritable.idxSpace);
} catch (ExecException e) {
throw new RuntimeException("Unable to access tuple field", e);
}
}
return result;
} else if (valuea==null && valueb==null) {
// If they're both null, compare the indicies
if ((wa.getIndex() & PigNullableWritable.idxSpace) < (wb.getIndex() & PigNullableWritable.idxSpace)) return -1;
else if ((wa.getIndex() & PigNullableWritable.idxSpace) > (wb.getIndex() & PigNullableWritable.idxSpace)) return 1;
else return 0;
}
else if (valuea==null) return -1;
else return 1;
}
}
public static class PigWritableComparator extends WritableComparator {
@SuppressWarnings("unchecked")
protected PigWritableComparator(Class c) {
super(c);
}
@Override
public int compare(byte[] b1, int s1, int l1, byte[] b2, int s2, int l2){
return WritableComparator.compareBytes(b1, s1, l1, b2, s2, l2);
}
}
public static class PigIntWritableComparator extends PigWritableComparator {
public PigIntWritableComparator() {
super(NullableIntWritable.class);
}
}
public static class PigLongWritableComparator extends PigWritableComparator {
public PigLongWritableComparator() {
super(NullableLongWritable.class);
}
}
public static class PigFloatWritableComparator extends PigWritableComparator {
public PigFloatWritableComparator() {
super(NullableFloatWritable.class);
}
}
public static class PigDoubleWritableComparator extends PigWritableComparator {
public PigDoubleWritableComparator() {
super(NullableDoubleWritable.class);
}
}
public static class PigCharArrayWritableComparator extends PigWritableComparator {
public PigCharArrayWritableComparator() {
super(NullableText.class);
}
}
public static class PigDBAWritableComparator extends PigWritableComparator {
public PigDBAWritableComparator() {
super(NullableBytesWritable.class);
}
}
public static class PigTupleWritableComparator extends PigWritableComparator {
public PigTupleWritableComparator() {
super(TupleFactory.getInstance().tupleClass());
}
}
public static class PigBagWritableComparator extends PigWritableComparator {
public PigBagWritableComparator() {
super(BagFactory.getInstance().newDefaultBag().getClass());
}
}
// XXX hadoop 20 new API integration: we need to explicitly set the Grouping
// Comparator
public static class PigGroupingIntWritableComparator extends WritableComparator {
public PigGroupingIntWritableComparator() {
super(NullableIntWritable.class, true);
}
}
public static class PigGroupingLongWritableComparator extends WritableComparator {
public PigGroupingLongWritableComparator() {
super(NullableLongWritable.class, true);
}
}
public static class PigGroupingFloatWritableComparator extends WritableComparator {
public PigGroupingFloatWritableComparator() {
super(NullableFloatWritable.class, true);
}
}
public static class PigGroupingDoubleWritableComparator extends WritableComparator {
public PigGroupingDoubleWritableComparator() {
super(NullableDoubleWritable.class, true);
}
}
public static class PigGroupingCharArrayWritableComparator extends WritableComparator {
public PigGroupingCharArrayWritableComparator() {
super(NullableText.class, true);
}
}
public static class PigGroupingDBAWritableComparator extends WritableComparator {
public PigGroupingDBAWritableComparator() {
super(NullableBytesWritable.class, true);
}
}
public static class PigGroupingTupleWritableComparator extends WritableComparator {
public PigGroupingTupleWritableComparator() {
super(NullableTuple.class, true);
}
}
public static class PigGroupingPartitionWritableComparator extends WritableComparator {
public PigGroupingPartitionWritableComparator() {
super(NullablePartitionWritable.class, true);
}
}
public static class PigGroupingBagWritableComparator extends WritableComparator {
public PigGroupingBagWritableComparator() {
super(BagFactory.getInstance().newDefaultBag().getClass(), true);
}
}
private void selectComparator(
MapReduceOper mro,
byte keyType,
org.apache.hadoop.mapreduce.Job job) throws JobCreationException {
// If this operator is involved in an order by, use the pig specific raw
// comparators. If it has a cogroup, we need to set the comparator class
// to the raw comparator and the grouping comparator class to pig specific
// raw comparators (which skip the index). Otherwise use the hadoop provided
// raw comparator.
// An operator has an order by if global sort is set or if it's successor has
// global sort set (because in that case it's the sampling job) or if
// it's a limit after a sort.
boolean hasOrderBy = false;
if (mro.isGlobalSort() || mro.isLimitAfterSort() || mro.usingTypedComparator()) {
hasOrderBy = true;
} else {
List<MapReduceOper> succs = plan.getSuccessors(mro);
if (succs != null) {
MapReduceOper succ = succs.get(0);
if (succ.isGlobalSort()) hasOrderBy = true;
}
}
if (hasOrderBy) {
switch (keyType) {
case DataType.INTEGER:
job.setSortComparatorClass(PigIntRawComparator.class);
break;
case DataType.LONG:
job.setSortComparatorClass(PigLongRawComparator.class);
break;
case DataType.FLOAT:
job.setSortComparatorClass(PigFloatRawComparator.class);
break;
case DataType.DOUBLE:
job.setSortComparatorClass(PigDoubleRawComparator.class);
break;
case DataType.CHARARRAY:
job.setSortComparatorClass(PigTextRawComparator.class);
break;
case DataType.BYTEARRAY:
job.setSortComparatorClass(PigBytesRawComparator.class);
break;
case DataType.MAP:
int errCode = 1068;
String msg = "Using Map as key not supported.";
throw new JobCreationException(msg, errCode, PigException.INPUT);
case DataType.TUPLE:
job.setSortComparatorClass(PigTupleSortComparator.class);
break;
case DataType.BAG:
errCode = 1068;
msg = "Using Bag as key not supported.";
throw new JobCreationException(msg, errCode, PigException.INPUT);
default:
break;
}
return;
}
switch (keyType) {
case DataType.INTEGER:
job.setSortComparatorClass(PigIntWritableComparator.class);
job.setGroupingComparatorClass(PigGroupingIntWritableComparator.class);
break;
case DataType.LONG:
job.setSortComparatorClass(PigLongWritableComparator.class);
job.setGroupingComparatorClass(PigGroupingLongWritableComparator.class);
break;
case DataType.FLOAT:
job.setSortComparatorClass(PigFloatWritableComparator.class);
job.setGroupingComparatorClass(PigGroupingFloatWritableComparator.class);
break;
case DataType.DOUBLE:
job.setSortComparatorClass(PigDoubleWritableComparator.class);
job.setGroupingComparatorClass(PigGroupingDoubleWritableComparator.class);
break;
case DataType.CHARARRAY:
job.setSortComparatorClass(PigCharArrayWritableComparator.class);
job.setGroupingComparatorClass(PigGroupingCharArrayWritableComparator.class);
break;
case DataType.BYTEARRAY:
job.setSortComparatorClass(PigDBAWritableComparator.class);
job.setGroupingComparatorClass(PigGroupingDBAWritableComparator.class);
break;
case DataType.MAP:
int errCode = 1068;
String msg = "Using Map as key not supported.";
throw new JobCreationException(msg, errCode, PigException.INPUT);
case DataType.TUPLE:
job.setSortComparatorClass(PigTupleWritableComparator.class);
job.setGroupingComparatorClass(PigGroupingTupleWritableComparator.class);
break;
case DataType.BAG:
errCode = 1068;
msg = "Using Bag as key not supported.";
throw new JobCreationException(msg, errCode, PigException.INPUT);
default:
errCode = 2036;
msg = "Unhandled key type " + DataType.findTypeName(keyType);
throw new JobCreationException(msg, errCode, PigException.BUG);
}
}
private void setupDistributedCacheForJoin(MapReduceOper mro,
PigContext pigContext, Configuration conf) throws IOException {
new JoinDistributedCacheVisitor(mro.mapPlan, pigContext, conf)
.visit();
new JoinDistributedCacheVisitor(mro.reducePlan, pigContext, conf)
.visit();
}
private static void setupDistributedCache(PigContext pigContext,
Configuration conf,
Properties properties, String key,
boolean shipToCluster)
throws IOException {
// Set up the DistributedCache for this job
String fileNames = properties.getProperty(key);
if (fileNames != null) {
String[] paths = fileNames.split(",");
setupDistributedCache(pigContext, conf, paths, shipToCluster);
}
}
private static void setupDistributedCache(PigContext pigContext,
Configuration conf, String[] paths, boolean shipToCluster) throws IOException {
// Turn on the symlink feature
DistributedCache.createSymlink(conf);
for (String path : paths) {
path = path.trim();
if (path.length() != 0) {
Path src = new Path(path);
// Ensure that 'src' is a valid URI
URI srcURI = null;
try {
srcURI = new URI(src.toString());
} catch (URISyntaxException ue) {
int errCode = 6003;
String msg = "Invalid cache specification. " +
"File doesn't exist: " + src;
throw new ExecException(msg, errCode, PigException.USER_ENVIRONMENT);
}
// Ship it to the cluster if necessary and add to the
// DistributedCache
if (shipToCluster) {
Path dst =
new Path(FileLocalizer.getTemporaryPath(pigContext).toString());
FileSystem fs = dst.getFileSystem(conf);
fs.copyFromLocalFile(src, dst);
// Construct the dst#srcName uri for DistributedCache
URI dstURI = null;
try {
dstURI = new URI(dst.toString() + "#" + src.getName());
} catch (URISyntaxException ue) {
byte errSrc = pigContext.getErrorSource();
int errCode = 0;
switch(errSrc) {
case PigException.REMOTE_ENVIRONMENT:
errCode = 6004;
break;
case PigException.USER_ENVIRONMENT:
errCode = 4004;
break;
default:
errCode = 2037;
break;
}
String msg = "Invalid ship specification. " +
"File doesn't exist: " + dst;
throw new ExecException(msg, errCode, errSrc);
}
DistributedCache.addCacheFile(dstURI, conf);
} else {
DistributedCache.addCacheFile(srcURI, conf);
}
}
}
}
private static String addSingleFileToDistributedCache(
PigContext pigContext, Configuration conf, String filename,
String prefix) throws IOException {
if (!FileLocalizer.fileExists(filename, pigContext)) {
throw new IOException(
"Internal error: skew join partition file "
+ filename + " does not exist");
}
String symlink = filename;
// XXX Hadoop currently doesn't support distributed cache in local mode.
// This line will be removed after the support is added by Hadoop team.
if (pigContext.getExecType() != ExecType.LOCAL) {
symlink = prefix + "_"
+ Integer.toString(System.identityHashCode(filename)) + "_"
+ Long.toString(System.currentTimeMillis());
filename = filename + "#" + symlink;
setupDistributedCache(pigContext, conf, new String[] { filename },
false);
}
return symlink;
}
private static class JoinDistributedCacheVisitor extends PhyPlanVisitor {
private PigContext pigContext = null;
private Configuration conf = null;
public JoinDistributedCacheVisitor(PhysicalPlan plan,
PigContext pigContext, Configuration conf) {
super(plan, new DepthFirstWalker<PhysicalOperator, PhysicalPlan>(
plan));
this.pigContext = pigContext;
this.conf = conf;
}
@Override
public void visitFRJoin(POFRJoin join) throws VisitorException {
// XXX Hadoop currently doesn't support distributed cache in local mode.
// This line will be removed after the support is added
if (pigContext.getExecType() == ExecType.LOCAL) return;
// set up distributed cache for the replicated files
FileSpec[] replFiles = join.getReplFiles();
ArrayList<String> replicatedPath = new ArrayList<String>();
FileSpec[] newReplFiles = new FileSpec[replFiles.length];
// the first input is not replicated
for (int i = 0; i < replFiles.length; i++) {
// ignore fragmented file
String symlink = "";
if (i != join.getFragment()) {
symlink = "pigrepl_" + join.getOperatorKey().toString() + "_"
+ Integer.toString(System.identityHashCode(replFiles[i].getFileName()))
+ "_" + Long.toString(System.currentTimeMillis())
+ "_" + i;
replicatedPath.add(replFiles[i].getFileName() + "#"
+ symlink);
}
newReplFiles[i] = new FileSpec(symlink,
(replFiles[i] == null ? null : replFiles[i].getFuncSpec()));
}
join.setReplFiles(newReplFiles);
try {
setupDistributedCache(pigContext, conf, replicatedPath
.toArray(new String[0]), false);
} catch (IOException e) {
String msg = "Internal error. Distributed cache could not " +
"be set up for the replicated files";
throw new VisitorException(msg, e);
}
}
@Override
public void visitMergeJoin(POMergeJoin join) throws VisitorException {
// XXX Hadoop currently doesn't support distributed cache in local mode.
// This line will be removed after the support is added
if (pigContext.getExecType() == ExecType.LOCAL) return;
String indexFile = join.getIndexFile();
// merge join may not use an index file
if (indexFile == null) return;
try {
String symlink = addSingleFileToDistributedCache(pigContext,
conf, indexFile, "indexfile_");
join.setIndexFile(symlink);
} catch (IOException e) {
String msg = "Internal error. Distributed cache could not " +
"be set up for merge join index file";
throw new VisitorException(msg, e);
}
}
@Override
public void visitMergeCoGroup(POMergeCogroup mergeCoGrp)
throws VisitorException {
// XXX Hadoop currently doesn't support distributed cache in local mode.
// This line will be removed after the support is added
if (pigContext.getExecType() == ExecType.LOCAL) return;
String indexFile = mergeCoGrp.getIndexFileName();
if (indexFile == null) throw new VisitorException("No index file");
try {
String symlink = addSingleFileToDistributedCache(pigContext,
conf, indexFile, "indexfile_mergecogrp_");
mergeCoGrp.setIndexFileName(symlink);
} catch (IOException e) {
String msg = "Internal error. Distributed cache could not " +
"be set up for merge cogrp index file";
throw new VisitorException(msg, e);
}
}
}
}