AMashenkov commented on a change in pull request #9081: URL: https://github.com/apache/ignite/pull/9081#discussion_r645168223
########## File path: modules/core/src/main/java/org/apache/ignite/internal/processors/cache/query/MergeSortDistributedCacheQueryReducer.java ########## @@ -0,0 +1,298 @@ +/* + * 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.ignite.internal.processors.cache.query; + +import java.lang.reflect.Array; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collection; +import java.util.Comparator; +import java.util.List; +import java.util.Map; +import java.util.NoSuchElementException; +import java.util.UUID; +import java.util.concurrent.ConcurrentHashMap; +import org.apache.ignite.IgniteCheckedException; +import org.apache.ignite.cluster.ClusterNode; +import org.apache.ignite.internal.util.typedef.internal.U; +import org.jetbrains.annotations.Nullable; + +/** + * Reducer of distirbuted query that sort result through all nodes. Note that it's assumed that every node + * returns pre-sorted collection of data. + */ +public class MergeSortDistributedCacheQueryReducer<R> extends UnsortedDistributedCacheQueryReducer<R> { + /** Map of Node ID to stream. Used for inserting new pages. */ + private final Map<UUID, NodePageStream> streamsMap; + + /** Array of streams. Used for iteration over result pages. */ + private final NodePageStream[] streams; + + /** If {@code true} then there wasn't call {@link #hasNext()} on this reducer. */ + private boolean first = true; + + /** + * Offset of current stream to get next value. This offset only increments if a stream is done. The array {@link + * #streams} is sorted to put stream with lowest value on this offset. + */ + private int streamOff; + + /** Compares head of streams to get lowest value at the moment. */ + private final Comparator<NodePageStream> streamCmp; + + /** + * @param rowCmp Comparator to sort query results from different nodes. + */ + public MergeSortDistributedCacheQueryReducer(GridCacheQueryFutureAdapter fut, long reqId, + CacheQueryPageRequester fetcher, Collection<ClusterNode> nodes, Comparator<R> rowCmp + ) { + super(fut, reqId, fetcher, nodes); + + synchronized (queueLock()) { + streamsMap = new ConcurrentHashMap<>(nodes.size()); + streams = (NodePageStream[])Array.newInstance(NodePageStream.class, nodes.size()); + + int i = 0; + + for (ClusterNode node : nodes) { + streams[i] = new NodePageStream(node.id()); + streamsMap.put(node.id(), streams[i++]); + } + } + + streamCmp = (o1, o2) -> { + if (o1 == o2) + return 0; + + if (o1 == null) + return -1; + + if (o2 == null) + return 1; + + return rowCmp.compare(o1.head, o2.head); + }; + } + + /** {@inheritDoc} */ + @Override public R next() throws IgniteCheckedException { + for (int i = streamOff; i < streams.length; i++) { + NodePageStream s = streams[i]; + + // Skip if stream has head: on previous next() head extracts from stream[streamOff], others streams do + // have head or it is the first run. + if (s.head() != null) + break; Review comment: After you got the last item from the first page, you will be blocked on `next()` call until the second page will be received from the node, otherwise you will not be able to resort streams correctly. You need to have at least one page from every node to be ready at any time for non-blocking reducing. I wouldn't bet on equal data distribution. Assume, one node will have the largest part of the query result, then you will end up in synchronous ping-pong with a single node. Effective prefetching is to request next page right after you start processing the previous one. If you bother about fetching few additional pages, then you can configure twice smaller page, aren't you? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
