jihoonson commented on a change in pull request #6629: Add support parallel combine in brokers URL: https://github.com/apache/incubator-druid/pull/6629#discussion_r240882437
########## File path: core/src/main/java/org/apache/druid/java/util/common/guava/ParallelMergeCombineSequence.java ########## @@ -0,0 +1,315 @@ +/* + * 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.druid.java.util.common.guava; + +import com.google.common.collect.Ordering; +import org.apache.druid.collections.ReferenceCountingResourceHolder; +import org.apache.druid.java.util.common.ISE; +import org.apache.druid.java.util.common.Pair; +import org.apache.druid.java.util.common.guava.BaseSequence.IteratorMaker; +import org.apache.druid.java.util.common.guava.nary.BinaryFn; +import org.apache.druid.query.ParallelCombines; +import org.apache.druid.query.PrioritizedRunnable; +import org.apache.druid.query.ThreadResource; + +import javax.annotation.Nullable; +import java.util.Iterator; +import java.util.List; +import java.util.NoSuchElementException; +import java.util.concurrent.ArrayBlockingQueue; +import java.util.concurrent.BlockingQueue; +import java.util.concurrent.ExecutionException; +import java.util.concurrent.ExecutorService; +import java.util.concurrent.Future; +import java.util.concurrent.TimeUnit; +import java.util.concurrent.TimeoutException; +import java.util.function.Supplier; +import java.util.stream.IntStream; + +/** + * This sequence merges underlying {@link #baseSequences} and combines (aggregates) them in parallel. + * It creates a tree to merge and combine input streams in parallel, consisting of processing threads connected by + * blocking queues. In leaf nodes, the processing threads combines values from the {@link #baseSequences} and stores + * aggregates in its blocking queue. Intermediate threads reads values from the blocking queue of their children and + * stores computed aggregates its blocking queue. Finally, the caller thread of this class (usually it's an http thread + * in query processing) reads values from the queue of the root thread. Filling blocking queue and reading from it are + * done asynchronously. + */ +public class ParallelMergeCombineSequence<T> extends YieldingSequenceBase<T> +{ + private static final int MINIMUM_LEAF_COMBINE_DEGREE = 2; + + private final ExecutorService exec; + private final List<Sequence<T>> baseSequences; + private final Ordering<T> ordering; + private final BinaryFn<T, T, T> combineFn; + private final List<ReferenceCountingResourceHolder<ThreadResource>> processingThreadHolders; + private final int queueSize; + private final boolean hasTimeout; + private final long timeoutAt; + private final int queryPriority; + + public ParallelMergeCombineSequence( + ExecutorService exec, + List<? extends Sequence<? extends T>> baseSequences, + Ordering<T> ordering, + BinaryFn<T, T, T> combineFn, + List<ReferenceCountingResourceHolder<ThreadResource>> processingThreadHolders, + int queueSize, + boolean hasTimeout, + long timeout, + int queryPriority + ) + { + this.exec = exec; + this.baseSequences = (List<Sequence<T>>) baseSequences; + this.ordering = ordering; + this.combineFn = combineFn; + this.processingThreadHolders = processingThreadHolders; + this.queueSize = queueSize; + this.hasTimeout = hasTimeout; + this.timeoutAt = System.currentTimeMillis() + timeout; + this.queryPriority = queryPriority; + } + + @Override + public <OutType> Yielder<OutType> toYielder(OutType initValue, YieldingAccumulator<OutType, T> accumulator) + { + final Pair<Integer, Integer> combineDegreeAndNumThreads = findCombineDegreeAndNumThreads( + processingThreadHolders.size(), + baseSequences.size() + ); + final int combineDegree = combineDegreeAndNumThreads.lhs; + final int numThreads = combineDegreeAndNumThreads.rhs; + + // Early release unnecessary processing threads + IntStream.range(0, processingThreadHolders.size() - numThreads) + .forEach(i -> processingThreadHolders.remove(0).close()); + + final Supplier<ReferenceCountingResourceHolder<ThreadResource>> processingThreadSupplier = + new Supplier<ReferenceCountingResourceHolder<ThreadResource>>() + { + private int next = 0; + + @Override + public ReferenceCountingResourceHolder<ThreadResource> get() + { + if (next < processingThreadHolders.size()) { + return processingThreadHolders.get(next++); + } else { + throw new ISE( + "WTH? current pointer[%d] is larger than available threads[%d]", + next, + processingThreadHolders.size() + ); + } + } + }; + + // In the result combine tree, nodes and edges are processing threads and blocking queues, respectively. + // In the leaf nodes, processing threads read data from historicals and fill its blocking queue. + // In the intermediate nodes, processing threads read data from the blocking queues of their children and fill + // its blocking queue. + // The caller thread reads the blocking queue of the root node. + final Pair<Sequence<T>, List<Future>> rootAndFutures = ParallelCombines.buildCombineTree( + baseSequences, + combineDegree, + combineDegree, + sequences -> runCombine(processingThreadSupplier.get(), sequences) + ); + + // Ignore futures since they are handled in the generated BaseSequence.IteratorMaker.cleanup(). + return rootAndFutures.lhs.toYielder(initValue, accumulator); + } + + private static Pair<Integer, Integer> findCombineDegreeAndNumThreads(int numAvailableThreads, int numLeafNodes) + { + for (int combineDegree = MINIMUM_LEAF_COMBINE_DEGREE; combineDegree <= numLeafNodes; combineDegree++) { + final int numRequiredThreads = computeNumRequiredThreads(numLeafNodes, combineDegree); + if (numRequiredThreads <= numAvailableThreads) { + return Pair.of(combineDegree, numRequiredThreads); + } + } + + throw new ISE( + "Cannot find a proper combine degree for the combining tree. " + + "Each node of the combining tree requires a single thread. " + + "Try increasing druid.processing.numThreads or " + + "reducing numBrokerParallelCombineThreads[%d] in the query context for a smaller tree", + numAvailableThreads + ); + } + + private static int computeNumRequiredThreads(int numChildNodes, int combineDegree) + { + // numChildNodes is used to determine that the last node is needed for the current level. + // Please see ParallelCombines.buildCombineTree() for more details. + final int numChildrenForLastNode = numChildNodes % combineDegree; + final int numCurLevelNodes = numChildNodes / combineDegree + (numChildrenForLastNode > 1 ? 1 : 0); + final int numChildOfParentNodes = numCurLevelNodes + (numChildrenForLastNode == 1 ? 1 : 0); + + if (numChildOfParentNodes == 1) { + return numCurLevelNodes; + } else { + return numCurLevelNodes + + computeNumRequiredThreads(numChildOfParentNodes, combineDegree); + } + } + + private Pair<Sequence<T>, Future> runCombine( + ReferenceCountingResourceHolder processingThreadHolder, + List<Sequence<T>> sequenceList + ) + { + final Sequence<? extends Sequence<T>> sequences = Sequences.simple(sequenceList); + final CombiningSequence<T> combiningSequence = CombiningSequence.create( + new MergeSequence<>(ordering, sequences), + ordering, + combineFn + ); + + final BlockingQueue<ValueHolder> queue = new ArrayBlockingQueue<>(queueSize); + + final Future future = exec.submit( + new PrioritizedRunnable() + { + @Override + public int getPriority() + { + return queryPriority; + } + + @Override + public void run() + { + try { + combiningSequence.accumulate( + queue, + (theQueue, v) -> { + try { + addToQueue(theQueue, new ValueHolder(v)); + } + catch (InterruptedException e) { + throw new RuntimeException(e); + } + return theQueue; + } + ); + // add a null to indicate this is the last one + addToQueue(queue, new ValueHolder(null)); + } + catch (InterruptedException e) { + throw new RuntimeException(e); + } + finally { + processingThreadHolder.close(); + } + } + } + ); + + final Sequence<T> backgroundCombineSequence = new BaseSequence<>( + new IteratorMaker<T, Iterator<T>>() + { + @Override + public Iterator<T> make() + { + return new Iterator<T>() + { + private T nextVal; + + @Override + public boolean hasNext() + { + try { + final ValueHolder holder; + if (!hasTimeout) { + holder = queue.take(); + } else { + final long timeout = timeoutAt - System.currentTimeMillis(); Review comment: `ArrayBlockingQueue` immediately returns if `timeout` is negative. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
