use of org.apache.tez.runtime.library.common.task.local.output.TezTaskOutput in project tez by apache.
the class TestUnorderedPartitionedKVWriter method baseTestWithPipelinedTransfer.
@SuppressWarnings("unchecked")
private void baseTestWithPipelinedTransfer(int numRecords, int numPartitions, Set<Integer> skippedPartitions, boolean shouldCompress) throws IOException, InterruptedException {
PartitionerForTest partitioner = new PartitionerForTest();
ApplicationId appId = ApplicationId.newInstance(10000000, 1);
TezCounters counters = new TezCounters();
String uniqueId = UUID.randomUUID().toString();
int dagId = 1;
String auxiliaryService = defaultConf.get(TezConfiguration.TEZ_AM_SHUFFLE_AUXILIARY_SERVICE_ID, TezConfiguration.TEZ_AM_SHUFFLE_AUXILIARY_SERVICE_ID_DEFAULT);
OutputContext outputContext = createMockOutputContext(counters, appId, uniqueId, auxiliaryService);
Configuration conf = createConfiguration(outputContext, IntWritable.class, LongWritable.class, shouldCompress, -1);
conf.setBoolean(TezRuntimeConfiguration.TEZ_RUNTIME_ENABLE_FINAL_MERGE_IN_OUTPUT, false);
conf.setBoolean(TezRuntimeConfiguration.TEZ_RUNTIME_PIPELINED_SHUFFLE_ENABLED, true);
CompressionCodec codec = null;
if (shouldCompress) {
codec = new DefaultCodec();
((Configurable) codec).setConf(conf);
}
int numOutputs = numPartitions;
long availableMemory = 2048;
int numRecordsWritten = 0;
UnorderedPartitionedKVWriter kvWriter = new UnorderedPartitionedKVWriterForTest(outputContext, conf, numOutputs, availableMemory);
int sizePerBuffer = kvWriter.sizePerBuffer;
// IntW + LongW
int sizePerRecord = 4 + 8;
// Record + META_OVERHEAD
int sizePerRecordWithOverhead = sizePerRecord + 12;
BitSet partitionsWithData = new BitSet(numPartitions);
IntWritable intWritable = new IntWritable();
LongWritable longWritable = new LongWritable();
for (int i = 0; i < numRecords; i++) {
intWritable.set(i);
longWritable.set(i);
int partition = partitioner.getPartition(intWritable, longWritable, numOutputs);
if (skippedPartitions != null && skippedPartitions.contains(partition)) {
continue;
}
partitionsWithData.set(partition);
kvWriter.write(intWritable, longWritable);
numRecordsWritten++;
}
int recordsPerBuffer = sizePerBuffer / sizePerRecordWithOverhead;
int numExpectedSpills = numRecordsWritten / recordsPerBuffer;
ArgumentCaptor<List> eventCaptor = ArgumentCaptor.forClass(List.class);
List<Event> lastEvents = kvWriter.close();
if (numPartitions == 1) {
assertEquals(false, kvWriter.skipBuffers);
}
// no events are sent to kvWriter upon close with pipelining
assertTrue(lastEvents.size() == 0);
verify(outputContext, atLeast(numExpectedSpills)).sendEvents(eventCaptor.capture());
int numOfCapturedEvents = eventCaptor.getAllValues().size();
lastEvents = eventCaptor.getAllValues().get(numOfCapturedEvents - 1);
VertexManagerEvent VMEvent = (VertexManagerEvent) lastEvents.get(0);
for (int i = 0; i < numOfCapturedEvents; i++) {
List<Event> events = eventCaptor.getAllValues().get(i);
if (i < numOfCapturedEvents - 1) {
assertTrue(events.size() == 1);
assertTrue(events.get(0) instanceof CompositeDataMovementEvent);
} else {
assertTrue(events.size() == 2);
assertTrue(events.get(0) instanceof VertexManagerEvent);
assertTrue(events.get(1) instanceof CompositeDataMovementEvent);
}
}
verifyPartitionStats(VMEvent, partitionsWithData);
verify(outputContext, never()).reportFailure(any(TaskFailureType.class), any(Throwable.class), any(String.class));
assertNull(kvWriter.currentBuffer);
assertEquals(0, kvWriter.availableBuffers.size());
// Verify the counters
TezCounter outputRecordBytesCounter = counters.findCounter(TaskCounter.OUTPUT_BYTES);
TezCounter outputRecordsCounter = counters.findCounter(TaskCounter.OUTPUT_RECORDS);
TezCounter outputBytesWithOverheadCounter = counters.findCounter(TaskCounter.OUTPUT_BYTES_WITH_OVERHEAD);
TezCounter fileOutputBytesCounter = counters.findCounter(TaskCounter.OUTPUT_BYTES_PHYSICAL);
TezCounter spilledRecordsCounter = counters.findCounter(TaskCounter.SPILLED_RECORDS);
TezCounter additionalSpillBytesWritternCounter = counters.findCounter(TaskCounter.ADDITIONAL_SPILLS_BYTES_WRITTEN);
TezCounter additionalSpillBytesReadCounter = counters.findCounter(TaskCounter.ADDITIONAL_SPILLS_BYTES_READ);
TezCounter numAdditionalSpillsCounter = counters.findCounter(TaskCounter.ADDITIONAL_SPILL_COUNT);
assertEquals(numRecordsWritten * sizePerRecord, outputRecordBytesCounter.getValue());
assertEquals(numRecordsWritten, outputRecordsCounter.getValue());
assertEquals(numRecordsWritten * sizePerRecordWithOverhead, outputBytesWithOverheadCounter.getValue());
long fileOutputBytes = fileOutputBytesCounter.getValue();
if (numRecordsWritten > 0) {
assertTrue(fileOutputBytes > 0);
if (!shouldCompress) {
assertTrue(fileOutputBytes > outputRecordBytesCounter.getValue());
}
} else {
assertEquals(0, fileOutputBytes);
}
// due to multiple threads, buffers could be merged in chunks in scheduleSpill.
assertTrue(recordsPerBuffer * numExpectedSpills >= spilledRecordsCounter.getValue());
long additionalSpillBytesWritten = additionalSpillBytesWritternCounter.getValue();
long additionalSpillBytesRead = additionalSpillBytesReadCounter.getValue();
// No additional spill bytes written when final merge is disabled.
assertEquals(additionalSpillBytesWritten, 0);
// No additional spills when final merge is disabled.
assertTrue(additionalSpillBytesWritten == additionalSpillBytesRead);
// No additional spills when final merge is disabled.
assertEquals(numAdditionalSpillsCounter.getValue(), 0);
assertTrue(lastEvents.size() > 0);
// Get the last event
int index = lastEvents.size() - 1;
assertTrue(lastEvents.get(index) instanceof CompositeDataMovementEvent);
CompositeDataMovementEvent cdme = (CompositeDataMovementEvent) lastEvents.get(index);
assertEquals(0, cdme.getSourceIndexStart());
assertEquals(numOutputs, cdme.getCount());
DataMovementEventPayloadProto eventProto = DataMovementEventPayloadProto.parseFrom(ByteString.copyFrom(cdme.getUserPayload()));
// Ensure that this is the last event
assertTrue(eventProto.getLastEvent());
verifyEmptyPartitions(eventProto, numRecordsWritten, numPartitions, skippedPartitions);
verify(outputContext, atLeast(1)).notifyProgress();
// Verify if all spill files are available.
TezTaskOutput taskOutput = new TezTaskOutputFiles(conf, uniqueId, dagId);
if (numRecordsWritten > 0) {
int numSpills = kvWriter.numSpills.get();
for (int i = 0; i < numSpills; i++) {
Path outputFile = taskOutput.getSpillFileForWrite(i, 10);
Path indexFile = taskOutput.getSpillIndexFileForWrite(i, 10);
assertTrue(localFs.exists(outputFile));
assertTrue(localFs.exists(indexFile));
assertEquals("Incorrect output permissions", (short) 0640, localFs.getFileStatus(outputFile).getPermission().toShort());
assertEquals("Incorrect index permissions", (short) 0640, localFs.getFileStatus(indexFile).getPermission().toShort());
}
} else {
return;
}
}
Aggregations