/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.record;
import org.apache.hadoop.mapred.*;
import org.apache.hadoop.fs.*;
import org.apache.hadoop.io.*;
import org.apache.hadoop.io.SequenceFile.CompressionType;
import org.apache.hadoop.conf.*;
import junit.framework.TestCase;
import java.io.*;
import java.util.*;
/**********************************************************
* MapredLoadTest generates a bunch of work that exercises
* a Hadoop Map-Reduce system (and DFS, too). It goes through
* the following steps:
*
* 1) Take inputs 'range' and 'counts'.
* 2) Generate 'counts' random integers between 0 and range-1.
* 3) Create a file that lists each integer between 0 and range-1,
* and lists the number of times that integer was generated.
* 4) Emit a (very large) file that contains all the integers
* in the order generated.
* 5) After the file has been generated, read it back and count
* how many times each int was generated.
* 6) Compare this big count-map against the original one. If
* they match, then SUCCESS! Otherwise, FAILURE!
*
* OK, that's how we can think about it. What are the map-reduce
* steps that get the job done?
*
* 1) In a non-mapred thread, take the inputs 'range' and 'counts'.
* 2) In a non-mapread thread, generate the answer-key and write to disk.
* 3) In a mapred job, divide the answer key into K jobs.
* 4) A mapred 'generator' task consists of K map jobs. Each reads
* an individual "sub-key", and generates integers according to
* to it (though with a random ordering).
* 5) The generator's reduce task agglomerates all of those files
* into a single one.
* 6) A mapred 'reader' task consists of M map jobs. The output
* file is cut into M pieces. Each of the M jobs counts the
* individual ints in its chunk and creates a map of all seen ints.
* 7) A mapred job integrates all the count files into a single one.
*
**********************************************************/
public class TestRecordMR extends TestCase {
/**
* Modified to make it a junit test.
* The RandomGen Job does the actual work of creating
* a huge file of assorted numbers. It receives instructions
* as to how many times each number should be counted. Then
* it emits those numbers in a crazy order.
*
* The map() function takes a key/val pair that describes
* a value-to-be-emitted (the key) and how many times it
* should be emitted (the value), aka "numtimes". map() then
* emits a series of intermediate key/val pairs. It emits
* 'numtimes' of these. The key is a random number and the
* value is the 'value-to-be-emitted'.
*
* The system collates and merges these pairs according to
* the random number. reduce() function takes in a key/value
* pair that consists of a crazy random number and a series
* of values that should be emitted. The random number key
* is now dropped, and reduce() emits a pair for every intermediate value.
* The emitted key is an intermediate value. The emitted value
* is just a blank string. Thus, we've created a huge file
* of numbers in random order, but where each number appears
* as many times as we were instructed.
*/
static public class RandomGenMapper implements Mapper<RecInt, RecInt,
RecInt, RecString> {
Random r = new Random();
public void configure(JobConf job) {
}
public void map(RecInt key,
RecInt val,
OutputCollector<RecInt, RecString> out,
Reporter reporter) throws IOException {
int randomVal = key.getData();
int randomCount = val.getData();
for (int i = 0; i < randomCount; i++) {
out.collect(new RecInt(Math.abs(r.nextInt())),
new RecString(Integer.toString(randomVal)));
}
}
public void close() {
}
}
/**
*/
static public class RandomGenReducer implements Reducer<RecInt, RecString,
RecInt, RecString> {
public void configure(JobConf job) {
}
public void reduce(RecInt key,
Iterator<RecString> it,
OutputCollector<RecInt, RecString> out,
Reporter reporter) throws IOException {
int keyint = key.getData();
while (it.hasNext()) {
String val = it.next().getData();
out.collect(new RecInt(Integer.parseInt(val)),
new RecString(""));
}
}
public void close() {
}
}
/**
* The RandomCheck Job does a lot of our work. It takes
* in a num/string keyspace, and transforms it into a
* key/count(int) keyspace.
*
* The map() function just emits a num/1 pair for every
* num/string input pair.
*
* The reduce() function sums up all the 1s that were
* emitted for a single key. It then emits the key/total
* pair.
*
* This is used to regenerate the random number "answer key".
* Each key here is a random number, and the count is the
* number of times the number was emitted.
*/
static public class RandomCheckMapper implements Mapper<RecInt, RecString,
RecInt, RecString> {
public void configure(JobConf job) {
}
public void map(RecInt key,
RecString val,
OutputCollector<RecInt, RecString> out,
Reporter reporter) throws IOException {
int pos = key.getData();
String str = val.getData();
out.collect(new RecInt(pos), new RecString("1"));
}
public void close() {
}
}
/**
*/
static public class RandomCheckReducer implements Reducer<RecInt, RecString,
RecInt, RecString> {
public void configure(JobConf job) {
}
public void reduce(RecInt key,
Iterator<RecString> it,
OutputCollector<RecInt, RecString> out,
Reporter reporter) throws IOException {
int keyint = key.getData();
int count = 0;
while (it.hasNext()) {
it.next();
count++;
}
out.collect(new RecInt(keyint), new RecString(Integer.toString(count)));
}
public void close() {
}
}
/**
* The Merge Job is a really simple one. It takes in
* an int/int key-value set, and emits the same set.
* But it merges identical keys by adding their values.
*
* Thus, the map() function is just the identity function
* and reduce() just sums. Nothing to see here!
*/
static public class MergeMapper implements Mapper<RecInt, RecString,
RecInt, RecInt> {
public void configure(JobConf job) {
}
public void map(RecInt key,
RecString val,
OutputCollector<RecInt, RecInt> out,
Reporter reporter) throws IOException {
int keyint = key.getData();
String valstr = val.getData();
out.collect(new RecInt(keyint), new RecInt(Integer.parseInt(valstr)));
}
public void close() {
}
}
static public class MergeReducer implements Reducer<RecInt, RecInt,
RecInt, RecInt> {
public void configure(JobConf job) {
}
public void reduce(RecInt key,
Iterator<RecInt> it,
OutputCollector<RecInt, RecInt> out,
Reporter reporter) throws IOException {
int keyint = key.getData();
int total = 0;
while (it.hasNext()) {
total += it.next().getData();
}
out.collect(new RecInt(keyint), new RecInt(total));
}
public void close() {
}
}
private static int range = 10;
private static int counts = 100;
private static Random r = new Random();
private static Configuration conf = new Configuration();
public void testMapred() throws Exception {
launch();
}
/**
*
*/
public static void launch() throws Exception {
//
// Generate distribution of ints. This is the answer key.
//
int countsToGo = counts;
int dist[] = new int[range];
for (int i = 0; i < range; i++) {
double avgInts = (1.0 * countsToGo) / (range - i);
dist[i] = (int) Math.max(0, Math.round(avgInts + (Math.sqrt(avgInts) * r.nextGaussian())));
countsToGo -= dist[i];
}
if (countsToGo > 0) {
dist[dist.length-1] += countsToGo;
}
//
// Write the answer key to a file.
//
FileSystem fs = FileSystem.get(conf);
Path testdir = new Path("mapred.loadtest");
if (!fs.mkdirs(testdir)) {
throw new IOException("Mkdirs failed to create directory " + testdir.toString());
}
Path randomIns = new Path(testdir, "genins");
if (!fs.mkdirs(randomIns)) {
throw new IOException("Mkdirs failed to create directory " + randomIns.toString());
}
Path answerkey = new Path(randomIns, "answer.key");
SequenceFile.Writer out = SequenceFile.createWriter(fs, conf,
answerkey, RecInt.class, RecInt.class,
CompressionType.NONE);
try {
for (int i = 0; i < range; i++) {
RecInt k = new RecInt();
RecInt v = new RecInt();
k.setData(i);
v.setData(dist[i]);
out.append(k, v);
}
} finally {
out.close();
}
//
// Now we need to generate the random numbers according to
// the above distribution.
//
// We create a lot of map tasks, each of which takes at least
// one "line" of the distribution. (That is, a certain number
// X is to be generated Y number of times.)
//
// A map task emits Y key/val pairs. The val is X. The key
// is a randomly-generated number.
//
// The reduce task gets its input sorted by key. That is, sorted
// in random order. It then emits a single line of text that
// for the given values. It does not emit the key.
//
// Because there's just one reduce task, we emit a single big
// file of random numbers.
//
Path randomOuts = new Path(testdir, "genouts");
fs.delete(randomOuts, true);
JobConf genJob = new JobConf(conf, TestRecordMR.class);
FileInputFormat.setInputPaths(genJob, randomIns);
genJob.setInputFormat(SequenceFileInputFormat.class);
genJob.setMapperClass(RandomGenMapper.class);
FileOutputFormat.setOutputPath(genJob, randomOuts);
genJob.setOutputKeyClass(RecInt.class);
genJob.setOutputValueClass(RecString.class);
genJob.setOutputFormat(SequenceFileOutputFormat.class);
genJob.setReducerClass(RandomGenReducer.class);
genJob.setNumReduceTasks(1);
JobClient.runJob(genJob);
//
// Next, we read the big file in and regenerate the
// original map. It's split into a number of parts.
// (That number is 'intermediateReduces'.)
//
// We have many map tasks, each of which read at least one
// of the output numbers. For each number read in, the
// map task emits a key/value pair where the key is the
// number and the value is "1".
//
// We have a single reduce task, which receives its input
// sorted by the key emitted above. For each key, there will
// be a certain number of "1" values. The reduce task sums
// these values to compute how many times the given key was
// emitted.
//
// The reduce task then emits a key/val pair where the key
// is the number in question, and the value is the number of
// times the key was emitted. This is the same format as the
// original answer key (except that numbers emitted zero times
// will not appear in the regenerated key.) The answer set
// is split into a number of pieces. A final MapReduce job
// will merge them.
//
// There's not really a need to go to 10 reduces here
// instead of 1. But we want to test what happens when
// you have multiple reduces at once.
//
int intermediateReduces = 10;
Path intermediateOuts = new Path(testdir, "intermediateouts");
fs.delete(intermediateOuts, true);
JobConf checkJob = new JobConf(conf, TestRecordMR.class);
FileInputFormat.setInputPaths(checkJob, randomOuts);
checkJob.setInputFormat(SequenceFileInputFormat.class);
checkJob.setMapperClass(RandomCheckMapper.class);
FileOutputFormat.setOutputPath(checkJob, intermediateOuts);
checkJob.setOutputKeyClass(RecInt.class);
checkJob.setOutputValueClass(RecString.class);
checkJob.setOutputFormat(SequenceFileOutputFormat.class);
checkJob.setReducerClass(RandomCheckReducer.class);
checkJob.setNumReduceTasks(intermediateReduces);
JobClient.runJob(checkJob);
//
// OK, now we take the output from the last job and
// merge it down to a single file. The map() and reduce()
// functions don't really do anything except reemit tuples.
// But by having a single reduce task here, we end up merging
// all the files.
//
Path finalOuts = new Path(testdir, "finalouts");
fs.delete(finalOuts, true);
JobConf mergeJob = new JobConf(conf, TestRecordMR.class);
FileInputFormat.setInputPaths(mergeJob, intermediateOuts);
mergeJob.setInputFormat(SequenceFileInputFormat.class);
mergeJob.setMapperClass(MergeMapper.class);
FileOutputFormat.setOutputPath(mergeJob, finalOuts);
mergeJob.setOutputKeyClass(RecInt.class);
mergeJob.setOutputValueClass(RecInt.class);
mergeJob.setOutputFormat(SequenceFileOutputFormat.class);
mergeJob.setReducerClass(MergeReducer.class);
mergeJob.setNumReduceTasks(1);
JobClient.runJob(mergeJob);
//
// Finally, we compare the reconstructed answer key with the
// original one. Remember, we need to ignore zero-count items
// in the original key.
//
boolean success = true;
Path recomputedkey = new Path(finalOuts, "part-00000");
SequenceFile.Reader in = new SequenceFile.Reader(fs, recomputedkey, conf);
int totalseen = 0;
try {
RecInt key = new RecInt();
RecInt val = new RecInt();
for (int i = 0; i < range; i++) {
if (dist[i] == 0) {
continue;
}
if (!in.next(key, val)) {
System.err.println("Cannot read entry " + i);
success = false;
break;
} else {
if (!((key.getData() == i) && (val.getData() == dist[i]))) {
System.err.println("Mismatch! Pos=" + key.getData() + ", i=" + i + ", val=" + val.getData() + ", dist[i]=" + dist[i]);
success = false;
}
totalseen += val.getData();
}
}
if (success) {
if (in.next(key, val)) {
System.err.println("Unnecessary lines in recomputed key!");
success = false;
}
}
} finally {
in.close();
}
int originalTotal = 0;
for (int i = 0; i < dist.length; i++) {
originalTotal += dist[i];
}
System.out.println("Original sum: " + originalTotal);
System.out.println("Recomputed sum: " + totalseen);
//
// Write to "results" whether the test succeeded or not.
//
Path resultFile = new Path(testdir, "results");
BufferedWriter bw = new BufferedWriter(new OutputStreamWriter(fs.create(resultFile)));
try {
bw.write("Success=" + success + "\n");
System.out.println("Success=" + success);
} finally {
bw.close();
}
fs.delete(testdir, true);
}
/**
* Launches all the tasks in order.
*/
public static void main(String[] argv) throws Exception {
if (argv.length < 2) {
System.err.println("Usage: TestRecordMR <range> <counts>");
System.err.println();
System.err.println("Note: a good test will have a <counts> value that is substantially larger than the <range>");
return;
}
int i = 0;
int range = Integer.parseInt(argv[i++]);
int counts = Integer.parseInt(argv[i++]);
launch();
}
}