/**
* Reads back the evaluations.
*
* @param fs File System
* @param conf Job configuration
* @param outpath output <code>Path</code>
* @param evaluations <code>List<Fitness></code> that contains the evaluated fitness for
* each candidate from the input population, sorted in the same order as the candidates.
* @throws IOException
*/
private static void importEvaluations(
FileSystem fs, JobConf conf, Path outpath, List<CDFitness> evaluations) throws IOException {
Sorter sorter = new Sorter(fs, LongWritable.class, CDFitness.class, conf);
// merge and sort the outputs
Path[] outfiles = OutputUtils.listOutputFiles(fs, outpath);
Path output = new Path(outpath, "output.sorted");
sorter.merge(outfiles, output);
// import the evaluations
LongWritable key = new LongWritable();
CDFitness value = new CDFitness();
Reader reader = new Reader(fs, output, conf);
while (reader.next(key, value)) {
evaluations.add(new CDFitness(value));
}
reader.close();
}
private void cloneOutput() throws IOException {
List<FileStatus> listStatus = getOutputMappings();
/*
* Initialize to empty list, in which case swap() will be a no-op. The reference is then replaced with a real list, which is
* used in the subsequent iterations.
*/
List<Path> crushInput = emptyList();
Text srcFile = new Text();
Text crushOut = new Text();
Text prevCrushOut = new Text();
for (FileStatus partFile : listStatus) {
Path path = partFile.getPath();
Reader reader = new Reader(fs, path, fs.getConf());
try {
while (reader.next(srcFile, crushOut)) {
if (!crushOut.equals(prevCrushOut)) {
swap(crushInput, prevCrushOut.toString());
prevCrushOut.set(crushOut);
crushInput = new LinkedList<Path>();
}
crushInput.add(new Path(srcFile.toString()));
}
} finally {
try {
reader.close();
} catch (IOException e) {
LOG.warn("Trapped exception when closing " + path, e);
}
}
swap(crushInput, prevCrushOut.toString());
}
}
@Test
public void testSequenceFile() throws Exception {
populateFile();
Pipeline p = Pipeline.create(pipelineOptions.getOptions());
@SuppressWarnings("unchecked")
Class<? extends FileInputFormat<IntWritable, Text>> inputFormatClass =
(Class<? extends FileInputFormat<IntWritable, Text>>)
(Class<?>) SequenceFileInputFormat.class;
HadoopIO.Read.Bound<IntWritable, Text> read =
HadoopIO.Read.from(
inputFile.getAbsolutePath(), inputFormatClass, IntWritable.class, Text.class);
PCollection<KV<IntWritable, Text>> input =
p.apply(read)
.setCoder(
KvCoder.of(WritableCoder.of(IntWritable.class), WritableCoder.of(Text.class)));
@SuppressWarnings("unchecked")
Class<? extends FileOutputFormat<IntWritable, Text>> outputFormatClass =
(Class<? extends FileOutputFormat<IntWritable, Text>>)
(Class<?>) TemplatedSequenceFileOutputFormat.class;
@SuppressWarnings("unchecked")
HadoopIO.Write.Bound<IntWritable, Text> write =
HadoopIO.Write.to(
outputFile.getAbsolutePath(), outputFormatClass, IntWritable.class, Text.class);
input.apply(write.withoutSharding());
p.run();
IntWritable key = new IntWritable();
Text value = new Text();
try (Reader reader =
new Reader(new Configuration(), Reader.file(new Path(outputFile.toURI())))) {
int i = 0;
while (reader.next(key, value)) {
assertEquals(i, key.get());
assertEquals("value-" + i, value.toString());
i++;
}
}
}
/**
* Moves the skipped files to the output directory. Called when operation in normal (non-clone)
* mode.
*/
private void moveOutput() throws IOException {
List<FileStatus> listStatus = getOutputMappings();
Text srcFile = new Text();
Text crushOut = new Text();
Set<String> crushOutputFiles = new HashSet<String>(nBuckets);
for (FileStatus partFile : listStatus) {
Path path = partFile.getPath();
Reader reader = new Reader(fs, path, fs.getConf());
try {
while (reader.next(srcFile, crushOut)) {
crushOutputFiles.add(new Path(crushOut.toString()).toUri().getPath());
}
} finally {
try {
reader.close();
} catch (IOException e) {
LOG.warn("Trapped exception when closing " + path, e);
}
}
}
assert crushOutputFiles.size() == nBuckets;
/*
* The crushoutput files will appear in a subdirectory of the output directory. The subdirectory will be the full path of the
* input directory that was crushed. E.g.
*
* Crush input:
* /user/me/input/dir1/file1
* /user/me/input/dir1/file2
* /user/me/input/dir2/file3
* /user/me/input/dir2/file4
* /user/me/input/dir3/dir4/file5
* /user/me/input/dir3/dir4/file6
*
* Crush output:
* /user/me/output/user/me/input/dir1/crushed_file ...
* /user/me/output/user/me/input/dir2/crushed_file ...
* /user/me/output/user/me/input/dir2/dir3/dir4/crushed_file ...
*
* We need to collapse this down to:
* /user/me/output/dir1/crushed_file ...
* /user/me/output/dir2/crushed_file ...
* /user/me/output/dir2/dir3/dir4/crushed_file ...
*/
String srcDirName = fs.makeQualified(srcDir).toUri().getPath();
String destName = fs.makeQualified(dest).toUri().getPath();
String partToReplace = fs.makeQualified(outDir).toUri().getPath() + "/crush" + srcDirName;
print(Verbosity.INFO, "\n\nCopying crush files to " + destName);
for (String crushOutputFile : crushOutputFiles) {
Path srcPath = new Path(crushOutputFile);
Path destPath =
new Path(destName + crushOutputFile.substring(partToReplace.length())).getParent();
rename(srcPath, destPath, null);
}
print(Verbosity.INFO, "\n\nMoving skipped files to " + destName);
/*
* Don't forget to move the files that were not crushed to the output dir so that the output dir has all the data that was in
* the input dir, the difference being there are fewer files in the output dir.
*/
for (String name : skippedFiles) {
Path srcPath = new Path(name);
Path destPath = new Path(destName + name.substring(srcDirName.length())).getParent();
rename(srcPath, destPath, null);
}
}
/** Runs this tool. */
public int run(String[] args) throws Exception {
long startWholeProgram = System.currentTimeMillis();
boolean hasConverged = false;
int counter = 0;
Reader sequenceFilereader;
// Must have four arguments
if (args.length != 4) {
MapReduce.printUsage();
return -1;
}
// Set input and output file paths
String inputPathToAdjacencyTextFile = args[0];
String outputPathToNodeSequenceFileFormat = args[1];
// Configure Job 1:
int mapTasks = Integer.parseInt(args[2]);
int reduceTasks = Integer.parseInt(args[3]);
int reduceTasksSetup = 1;
// Configure Job Setup
JobConf conf1 = new JobConf(MapReduce.class);
conf1.setInt("numberOfNodes", numberNodes);
conf1.setJobName("Setup Job");
conf1.setNumMapTasks(mapTasks);
conf1.setNumReduceTasks(reduceTasksSetup);
FileInputFormat.setInputPaths(conf1, new Path(inputPathToAdjacencyTextFile));
FileOutputFormat.setOutputPath(conf1, new Path(outputPathToNodeSequenceFileFormat));
FileOutputFormat.setCompressOutput(conf1, false);
conf1.setOutputKeyClass(Text.class);
conf1.setOutputValueClass(MapReduceNode.class);
conf1.setOutputFormat(SequenceFileOutputFormat.class);
conf1.setInputFormat(TextInputFormat.class);
conf1.setMapperClass(ConfigureMapper.class);
conf1.setReducerClass(ConfigureReducer.class);
// Delete the output directory if it exists already
Path tempDir = new Path(outputPathToNodeSequenceFileFormat);
FileSystem.get(tempDir.toUri(), conf1).delete(tempDir, true);
long startTime = System.currentTimeMillis();
// Run Configure Job
RunningJob job = JobClient.runJob(conf1);
sLogger.info(
"Config Job Finished in " + (System.currentTimeMillis() - startTime) / 1000.0 + " seconds");
String inputPath = args[1];
String outputPath = null;
String outputPathForNormalisedPagerank = null;
String outputPathforFinalSortedPagerank = null;
// Page Rank Calculation Job 2 (Iterative until convergence has been reached)
while (!hasConverged) {
System.out.println("*** ITERATION " + counter + ", number of nodes: " + numberNodes);
counter++;
sLogger.info("***** ITERATION " + counter);
outputPath = args[1] + counter;
// Pure Page Rank Calculation Job Setup
JobConf pageRankJob = new JobConf(getConf(), MapReduce.class);
pageRankJob.setInt("numberOfNodes", numberNodes);
FileInputFormat.setInputPaths(pageRankJob, new Path(inputPath));
FileOutputFormat.setOutputPath(pageRankJob, new Path(outputPath));
FileOutputFormat.setCompressOutput(pageRankJob, false);
pageRankJob.setJobName("PP Iteration " + counter);
pageRankJob.setNumMapTasks(mapTasks);
pageRankJob.setNumReduceTasks(reduceTasks);
pageRankJob.setOutputKeyClass(Text.class);
pageRankJob.setOutputValueClass(MapReduceNode.class);
pageRankJob.setOutputFormat(SequenceFileOutputFormat.class);
pageRankJob.setInputFormat(SequenceFileInputFormat.class);
pageRankJob.setMapperClass(MapReduce.PageRankCalcMapper.class);
pageRankJob.setReducerClass(MapReduce.PageRankCalcReducer.class);
// Delete the output directory if it exists already
Path tempPageRankDir = new Path(outputPath);
FileSystem.get(tempDir.toUri(), conf1).delete(tempPageRankDir, true);
startTime = System.currentTimeMillis();
// Run Pure Page Rank Calculation Job
RunningJob runningJob = JobClient.runJob(pageRankJob);
sLogger.info(
"PP Job"
+ counter
+ "Finished in "
+ (System.currentTimeMillis() - startTime) / 1000.0
+ " seconds");
// Delete the input directory if it exists already
Path tempInputPageRankDir = new Path(inputPath);
FileSystem.get(tempDir.toUri(), conf1).delete(tempInputPageRankDir, true);
// Set the output path of this iteration to be the inputpath for the next iteration
inputPath = outputPath;
// Check for convergence after every five iterations
if (counter % 5 == 0) {
Configuration conf = getConf();
if (outputPath != null) {
sLogger.info("Attempting to open file: " + outputPath + File.separator + "part-00000");
System.out.println(
"Attempting to open file: " + outputPath + File.separator + "part-00000");
} else {
sLogger.info("OUTPUT PATH IS NULL");
System.out.println("OUTPUT PATH IS NULL");
}
Path cFile = new Path(outputPath + File.separator + "part-00000");
FileSystem fs = FileSystem.get(cFile.toUri(), conf);
sequenceFilereader = new Reader(fs, cFile, conf);
for (int i = 0; i < 5; i++) {
MapReduceNode readValue = new MapReduceNode();
Text readKey = new Text();
sequenceFilereader.next(readKey, readValue);
if (!(readValue.hasConverged())) {
break;
}
if (i == 4) {
hasConverged = true;
sequenceFilereader.close();
}
}
sequenceFilereader.close();
}
if (counter == 75) {
sLogger.info("****************** Exiting (purposefully) after 75th iteration");
hasConverged = true;
}
}
// Normalised Page Rank Calculation Job 3
outputPathForNormalisedPagerank = args[1] + "normalizedPageRank";
// Normalised Page Rank Calculation Job Setup
JobConf normalizationJob = new JobConf(getConf(), MapReduce.class);
FileInputFormat.setInputPaths(normalizationJob, new Path(inputPath));
FileOutputFormat.setOutputPath(normalizationJob, new Path(outputPathForNormalisedPagerank));
FileOutputFormat.setCompressOutput(normalizationJob, false);
normalizationJob.setJobName("Normalised Pagerank Output");
normalizationJob.setNumMapTasks(mapTasks);
normalizationJob.setNumReduceTasks(1);
normalizationJob.setOutputKeyClass(Text.class);
normalizationJob.setOutputValueClass(DoubleWritable.class);
normalizationJob.setInputFormat(SequenceFileInputFormat.class);
normalizationJob.setOutputFormat(SequenceFileOutputFormat.class);
normalizationJob.setMapperClass(NormalisationMapper.class);
normalizationJob.setReducerClass(NormalisationReducer.class);
// Delete the output directory if it exists already
Path tempUpdatedPageRankDir = new Path(outputPathForNormalisedPagerank);
FileSystem.get(tempDir.toUri(), conf1).delete(tempUpdatedPageRankDir, true);
startTime = System.currentTimeMillis();
// Run Normalised Page Rank Calculation Job
RunningJob runningUpdateJob = JobClient.runJob(normalizationJob);
sLogger.info(
"Normalisation Job Finished in "
+ (System.currentTimeMillis() - startTime) / 1000.0
+ " seconds");
// Sorting and Output Job 4
// Delete the intermediary files created
Path tempNormalizationInputPath = new Path(inputPath);
FileSystem.get(tempDir.toUri(), conf1).delete(tempNormalizationInputPath, true);
inputPath = outputPathForNormalisedPagerank;
outputPathforFinalSortedPagerank = args[1] + "FinalSortedPageRank";
// Sorting and Output Job Setup
JobConf outputJob = new JobConf(getConf(), MapReduce.class);
FileInputFormat.setInputPaths(outputJob, new Path(inputPath));
FileOutputFormat.setOutputPath(outputJob, new Path(outputPathforFinalSortedPagerank));
FileOutputFormat.setCompressOutput(outputJob, false);
outputJob.setJobName("Final Pagerank Output");
sLogger.info("Starting final sotirng job -> this will output a single file");
outputJob.setNumMapTasks(1);
outputJob.setNumReduceTasks(1);
outputJob.setOutputKeyClass(DoubleWritable.class);
outputJob.setOutputValueClass(Text.class);
outputJob.setInputFormat(SequenceFileInputFormat.class);
outputJob.setOutputFormat(TextOutputFormat.class);
outputJob.setMapperClass(OutputMapper.class);
outputJob.setReducerClass(OutputReducer.class);
outputJob.setOutputKeyComparatorClass(ReverseComparator.class);
startTime = System.currentTimeMillis();
// Run Sorting and Output Job
RunningJob runningSortingJob = JobClient.runJob(outputJob);
// Delete the intermediary files created
Path tempFinalSortedInputPath = new Path(inputPath);
FileSystem.get(tempDir.toUri(), conf1).delete(tempFinalSortedInputPath, true);
sLogger.info(
"Final Sorting Job Finished in "
+ (System.currentTimeMillis() - startTime) / 1000.0
+ " seconds");
sLogger.info(
"The program lasted "
+ (System.currentTimeMillis() - startWholeProgram) / 1000.0
+ "s ("
+ (System.currentTimeMillis() - startWholeProgram) / 60000.0
+ " mins)");
return 0;
}