List of usage examples for org.apache.hadoop.mapred Task setConf
public void setConf(Configuration conf)
From source file:com.scaleoutsoftware.soss.hserver.hadoop.MapperWrapperMapred.java
License:Apache License
/** * Runs mapper for the single split./*from w ww . j av a 2 s . com*/ * * @param mapOutputAccumulator mapOutputAccumulator to use * @param split split ot run on */ @Override @SuppressWarnings("unchecked") public void runSplit(final MapOutputAccumulator<OUTKEY, OUTVALUE> mapOutputAccumulator, Object split, int splitIndex) throws IOException, ClassNotFoundException, InterruptedException { JobConf jobConf = new JobConf(this.jobConf); //Clone JobConf to prevent unexpected task interaction TaskAttemptID taskAttemptID = TaskAttemptID .downgrade(hadoopVersionSpecificCode.createTaskAttemptId(jobId, true, splitIndex)); ReducerWrapperMapred.updateJobConf(jobConf, taskAttemptID, splitIndex); updateJobWithSplit(jobConf, split); InputFormat inputFormat = jobConf.getInputFormat(); Reporter reporter = Reporter.NULL; //Create RecordReader org.apache.hadoop.mapred.RecordReader<INKEY, INVALUE> recordReader = inputFormat .getRecordReader((InputSplit) split, jobConf, reporter); //Make a mapper org.apache.hadoop.mapred.Mapper<INKEY, INVALUE, OUTKEY, OUTVALUE> mapper; try { mapper = (org.apache.hadoop.mapred.Mapper<INKEY, INVALUE, OUTKEY, OUTVALUE>) mapperConstructor .newInstance(); mapper.configure(jobConf); } catch (Exception e) { throw new RuntimeException("Cannot instantiate mapper " + mapperConstructor.getDeclaringClass(), e); } //These are to support map only jobs which write output directly to HDFS. final RecordWriter outputRecordWriter; OutputCommitter outputCommitter = null; TaskAttemptContext taskAttemptContext = null; if (mapOnlyJob) { taskAttemptContext = hadoopVersionSpecificCode.createTaskAttemptContextMapred(jobConf, taskAttemptID); OutputFormat outputFormat = jobConf.getOutputFormat(); FileSystem fs = FileSystem.get(jobConf); outputRecordWriter = (org.apache.hadoop.mapred.RecordWriter<OUTKEY, OUTVALUE>) outputFormat .getRecordWriter(fs, jobConf, ReducerWrapperMapred.getOutputName(splitIndex), Reporter.NULL); outputCommitter = jobConf.getOutputCommitter(); //Create task object so it can handle file format initialization //The MapTask is private in the Hadoop 1.x so we have to go through reflection. try { Class reduceTask = Class.forName("org.apache.hadoop.mapred.MapTask"); Constructor reduceTaskConstructor = reduceTask.getDeclaredConstructor(String.class, TaskAttemptID.class, int.class, JobSplit.TaskSplitIndex.class, int.class); reduceTaskConstructor.setAccessible(true); Task task = (Task) reduceTaskConstructor.newInstance(null, taskAttemptID, splitIndex, new JobSplit.TaskSplitIndex(), 0); task.setConf(jobConf); task.initialize(jobConf, jobId, Reporter.NULL, false); } catch (Exception e) { throw new IOException("Cannot initialize MapTask", e); } outputCommitter.setupTask(taskAttemptContext); } else { outputRecordWriter = null; } OutputCollector<OUTKEY, OUTVALUE> outputCollector; if (!mapOnlyJob) { outputCollector = new OutputCollector<OUTKEY, OUTVALUE>() { @Override public void collect(OUTKEY outkey, OUTVALUE outvalue) throws IOException { try { mapOutputAccumulator.combine(outkey, outvalue); } catch (InterruptedException e) { Thread.currentThread().interrupt(); } } }; } else { outputCollector = new OutputCollector<OUTKEY, OUTVALUE>() { @Override public void collect(OUTKEY outkey, OUTVALUE outvalue) throws IOException { outputRecordWriter.write(outkey, outvalue); } }; } INKEY key = recordReader.createKey(); INVALUE value = recordReader.createValue(); while (recordReader.next(key, value)) { mapper.map(key, value, outputCollector, reporter); } mapper.close(); recordReader.close(); if (mapOnlyJob) { outputRecordWriter.close(Reporter.NULL); outputCommitter.commitTask(taskAttemptContext); } }
From source file:com.scaleoutsoftware.soss.hserver.hadoop.ReducerWrapperMapred.java
License:Apache License
public ReducerWrapperMapred(HServerInvocationParameters invocationParameters, int hadoopPartition, int appId, int region, boolean sort) throws IOException, ClassNotFoundException, InterruptedException { this.invocationParameters = invocationParameters; JobConf jobConf = new JobConf((Configuration) invocationParameters.getConfiguration()); //Clone JobConf, so the temporary settings do not pollute other tasks LOG.info("Starting reducer:" + HadoopInvocationParameters.dumpConfiguration(jobConf)); JobID jobID = (JobID) invocationParameters.getJobId(); this.hadoopPartition = hadoopPartition; hadoopVersionSpecificCode = HadoopVersionSpecificCode.getInstance(invocationParameters.getHadoopVersion(), jobConf);/*from w w w .j a va 2 s .com*/ TaskAttemptID taskAttemptID = TaskAttemptID .downgrade(hadoopVersionSpecificCode.createTaskAttemptId(jobID, false, hadoopPartition)); updateJobConf(jobConf, taskAttemptID, region); context = hadoopVersionSpecificCode.createTaskAttemptContextMapred(jobConf, taskAttemptID); reducer = (org.apache.hadoop.mapred.Reducer<INKEY, INVALUE, OUTKEY, OUTVALUE>) ReflectionUtils .newInstance(jobConf.getReducerClass(), jobConf); reducer.configure(jobConf); OutputFormat outputFormat = jobConf.getOutputFormat(); FileSystem fs = FileSystem.get(jobConf); recordWriter = (org.apache.hadoop.mapred.RecordWriter<OUTKEY, OUTVALUE>) outputFormat.getRecordWriter(fs, jobConf, getOutputName(hadoopPartition), Reporter.NULL); committer = jobConf.getOutputCommitter(); //Create task object so it can handle file format initialization //The ReduceTask is private in the Hadoop 1.x so we have to go through reflection. try { Class reduceTask = Class.forName("org.apache.hadoop.mapred.ReduceTask"); Constructor reduceTaskConstructor = reduceTask.getDeclaredConstructor(String.class, TaskAttemptID.class, int.class, int.class, int.class); reduceTaskConstructor.setAccessible(true); Task task = (Task) reduceTaskConstructor.newInstance(null, taskAttemptID, hadoopPartition, 0, 0); task.setConf(jobConf); task.initialize(jobConf, jobID, Reporter.NULL, false); } catch (Exception e) { throw new IOException("Cannot initialize ReduceTask", e); } committer.setupTask(context); Class<INKEY> keyClass = (Class<INKEY>) jobConf.getMapOutputKeyClass(); WritableSerializerDeserializer<INKEY> firstKeySerializer = new WritableSerializerDeserializer<INKEY>( keyClass, null); WritableSerializerDeserializer<INKEY> secondKeySerializer = new WritableSerializerDeserializer<INKEY>( keyClass, null); Class<INVALUE> valueClass = (Class<INVALUE>) jobConf.getMapOutputValueClass(); WritableSerializerDeserializer<INVALUE> valueSerializer = new WritableSerializerDeserializer<INVALUE>( valueClass, null); DataGridReaderParameters<INKEY, INVALUE> params = new DataGridReaderParameters<INKEY, INVALUE>(region, appId, HServerParameters.getSetting(REDUCE_USEMEMORYMAPPEDFILES, jobConf) > 0, firstKeySerializer, valueSerializer, invocationParameters.getSerializationMode(), secondKeySerializer, keyClass, valueClass, sort, HServerParameters.getSetting(REDUCE_CHUNKSTOREADAHEAD, jobConf), 1024 * HServerParameters.getSetting(REDUCE_INPUTCHUNKSIZE_KB, jobConf), HServerParameters.getSetting(REDUCE_CHUNKREADTIMEOUT, jobConf)); transport = DataGridChunkedCollectionReader.getGridReader(params); outputCollector = new OutputCollector<OUTKEY, OUTVALUE>() { @Override public void collect(OUTKEY outkey, OUTVALUE outvalue) throws IOException { recordWriter.write(outkey, outvalue); } }; }