/*
* 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.flink.runtime.operators;
import java.io.EOFException;
import java.io.IOException;
import java.util.List;
import org.apache.flink.api.common.ExecutionConfig;
import org.apache.flink.metrics.Counter;
import org.apache.flink.runtime.operators.util.metrics.CountingCollector;
import org.apache.flink.runtime.operators.hash.InPlaceMutableHashTable;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.flink.api.common.functions.ReduceFunction;
import org.apache.flink.api.common.typeutils.TypeComparator;
import org.apache.flink.api.common.typeutils.TypeSerializer;
import org.apache.flink.api.common.typeutils.TypeSerializerFactory;
import org.apache.flink.core.memory.MemorySegment;
import org.apache.flink.runtime.memory.MemoryManager;
import org.apache.flink.runtime.operators.sort.FixedLengthRecordSorter;
import org.apache.flink.runtime.operators.sort.InMemorySorter;
import org.apache.flink.runtime.operators.sort.NormalizedKeySorter;
import org.apache.flink.runtime.operators.sort.QuickSort;
import org.apache.flink.util.Collector;
import org.apache.flink.util.MutableObjectIterator;
/**
* Combine operator for Reduce functions, standalone (not chained).
* Sorts and groups and reduces data, but never spills the sort. May produce multiple
* partially aggregated groups.
*
* @param <T> The data type consumed and produced by the combiner.
*/
public class ReduceCombineDriver<T> implements Driver<ReduceFunction<T>, T> {
private static final Logger LOG = LoggerFactory.getLogger(ReduceCombineDriver.class);
/** Fix length records with a length below this threshold will be in-place sorted, if possible. */
private static final int THRESHOLD_FOR_IN_PLACE_SORTING = 32;
private TaskContext<ReduceFunction<T>, T> taskContext;
private TypeSerializer<T> serializer;
private TypeComparator<T> comparator;
private ReduceFunction<T> reducer;
private Collector<T> output;
private DriverStrategy strategy;
private InMemorySorter<T> sorter;
private QuickSort sortAlgo = new QuickSort();
private InPlaceMutableHashTable<T> table;
private InPlaceMutableHashTable<T>.ReduceFacade reduceFacade;
private List<MemorySegment> memory;
private volatile boolean running;
private boolean objectReuseEnabled = false;
// ------------------------------------------------------------------------
@Override
public void setup(TaskContext<ReduceFunction<T>, T> context) {
taskContext = context;
running = true;
}
@Override
public int getNumberOfInputs() {
return 1;
}
@Override
public Class<ReduceFunction<T>> getStubType() {
@SuppressWarnings("unchecked")
final Class<ReduceFunction<T>> clazz = (Class<ReduceFunction<T>>) (Class<?>) ReduceFunction.class;
return clazz;
}
@Override
public int getNumberOfDriverComparators() {
return 1;
}
@Override
public void prepare() throws Exception {
final Counter numRecordsOut = taskContext.getMetricGroup().getIOMetricGroup().getNumRecordsOutCounter();
strategy = taskContext.getTaskConfig().getDriverStrategy();
// instantiate the serializer / comparator
final TypeSerializerFactory<T> serializerFactory = taskContext.getInputSerializer(0);
comparator = taskContext.getDriverComparator(0);
serializer = serializerFactory.getSerializer();
reducer = taskContext.getStub();
output = new CountingCollector<>(this.taskContext.getOutputCollector(), numRecordsOut);
MemoryManager memManager = taskContext.getMemoryManager();
final int numMemoryPages = memManager.computeNumberOfPages(
taskContext.getTaskConfig().getRelativeMemoryDriver());
memory = memManager.allocatePages(taskContext.getContainingTask(), numMemoryPages);
ExecutionConfig executionConfig = taskContext.getExecutionConfig();
objectReuseEnabled = executionConfig.isObjectReuseEnabled();
if (LOG.isDebugEnabled()) {
LOG.debug("ReduceCombineDriver object reuse: " + (objectReuseEnabled ? "ENABLED" : "DISABLED") + ".");
}
switch (strategy) {
case SORTED_PARTIAL_REDUCE:
// instantiate a fix-length in-place sorter, if possible, otherwise the out-of-place sorter
if (comparator.supportsSerializationWithKeyNormalization() &&
serializer.getLength() > 0 && serializer.getLength() <= THRESHOLD_FOR_IN_PLACE_SORTING) {
sorter = new FixedLengthRecordSorter<T>(serializer, comparator.duplicate(), memory);
} else {
sorter = new NormalizedKeySorter<T>(serializer, comparator.duplicate(), memory);
}
break;
case HASHED_PARTIAL_REDUCE:
table = new InPlaceMutableHashTable<T>(serializer, comparator, memory);
reduceFacade = table.new ReduceFacade(reducer, output, objectReuseEnabled);
break;
default:
throw new Exception("Invalid strategy " + taskContext.getTaskConfig().getDriverStrategy() + " for reduce combiner.");
}
}
@Override
public void run() throws Exception {
if (LOG.isDebugEnabled()) {
LOG.debug("Combiner starting.");
}
final Counter numRecordsIn = taskContext.getMetricGroup().getIOMetricGroup().getNumRecordsInCounter();
final MutableObjectIterator<T> in = taskContext.getInput(0);
final TypeSerializer<T> serializer = this.serializer;
switch (strategy) {
case SORTED_PARTIAL_REDUCE:
if (objectReuseEnabled) {
T value = serializer.createInstance();
while (running && (value = in.next(value)) != null) {
numRecordsIn.inc();
// try writing to the sorter first
if (sorter.write(value)) {
continue;
}
// do the actual sorting, combining, and data writing
sortAndCombine();
sorter.reset();
// write the value again
if (!sorter.write(value)) {
throw new IOException("Cannot write record to fresh sort buffer. Record too large.");
}
}
} else {
T value;
while (running && (value = in.next()) != null) {
numRecordsIn.inc();
// try writing to the sorter first
if (sorter.write(value)) {
continue;
}
// do the actual sorting, combining, and data writing
sortAndCombine();
sorter.reset();
// write the value again
if (!sorter.write(value)) {
throw new IOException("Cannot write record to fresh sort buffer. Record too large.");
}
}
}
// sort, combine, and send the final batch
sortAndCombine();
break;
case HASHED_PARTIAL_REDUCE:
table.open();
if (objectReuseEnabled) {
T value = serializer.createInstance();
while (running && (value = in.next(value)) != null) {
numRecordsIn.inc();
try {
reduceFacade.updateTableEntryWithReduce(value);
} catch (EOFException ex) {
// the table has run out of memory
reduceFacade.emitAndReset();
// try again
reduceFacade.updateTableEntryWithReduce(value);
}
}
} else {
T value;
while (running && (value = in.next()) != null) {
numRecordsIn.inc();
try {
reduceFacade.updateTableEntryWithReduce(value);
} catch (EOFException ex) {
// the table has run out of memory
reduceFacade.emitAndReset();
// try again
reduceFacade.updateTableEntryWithReduce(value);
}
}
}
// send the final batch
reduceFacade.emit();
table.close();
break;
default:
throw new Exception("Invalid strategy " + taskContext.getTaskConfig().getDriverStrategy() + " for reduce combiner.");
}
}
private void sortAndCombine() throws Exception {
final InMemorySorter<T> sorter = this.sorter;
if (!sorter.isEmpty()) {
sortAlgo.sort(sorter);
final TypeSerializer<T> serializer = this.serializer;
final TypeComparator<T> comparator = this.comparator;
final ReduceFunction<T> function = this.reducer;
final Collector<T> output = this.output;
final MutableObjectIterator<T> input = sorter.getIterator();
if (objectReuseEnabled) {
// We only need two objects. The first reference stores results and is
// eventually collected. New values are read into the second.
//
// The output value must have the same key fields as the input values.
T reuse1 = input.next();
T reuse2 = serializer.createInstance();
T value = reuse1;
// iterate over key groups
while (running && value != null) {
comparator.setReference(value);
// iterate within a key group
while ((reuse2 = input.next(reuse2)) != null) {
if (comparator.equalToReference(reuse2)) {
// same group, reduce
value = function.reduce(value, reuse2);
// we must never read into the object returned
// by the user, so swap the reuse objects
if (value == reuse2) {
T tmp = reuse1;
reuse1 = reuse2;
reuse2 = tmp;
}
} else {
// new key group
break;
}
}
output.collect(value);
// swap the value from the new key group into the first object
T tmp = reuse1;
reuse1 = reuse2;
reuse2 = tmp;
value = reuse1;
}
} else {
T value = input.next();
// iterate over key groups
while (running && value != null) {
comparator.setReference(value);
T res = value;
// iterate within a key group
while ((value = input.next()) != null) {
if (comparator.equalToReference(value)) {
// same group, reduce
res = function.reduce(res, value);
} else {
// new key group
break;
}
}
output.collect(res);
}
}
}
}
@Override
public void cleanup() {
try {
if (sorter != null) {
sorter.dispose();
}
if (table != null) {
table.close();
}
} catch (Exception e) {
// may happen during concurrent modification
}
taskContext.getMemoryManager().release(memory);
}
@Override
public void cancel() {
running = false;
cleanup();
}
}