clintropolis commented on a change in pull request #10082: URL: https://github.com/apache/druid/pull/10082#discussion_r447520607
########## File path: server/src/main/java/org/apache/druid/query/RetryQueryRunner.java ########## @@ -0,0 +1,250 @@ +/* + * 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.query; + +import com.fasterxml.jackson.core.type.TypeReference; +import com.fasterxml.jackson.databind.ObjectMapper; +import com.google.common.annotations.VisibleForTesting; +import com.google.common.base.Preconditions; +import org.apache.druid.java.util.common.ISE; +import org.apache.druid.java.util.common.guava.BaseSequence; +import org.apache.druid.java.util.common.guava.BaseSequence.IteratorMaker; +import org.apache.druid.java.util.common.guava.MergeSequence; +import org.apache.druid.java.util.common.guava.Sequence; +import org.apache.druid.java.util.common.guava.Yielder; +import org.apache.druid.java.util.common.guava.YieldingAccumulator; +import org.apache.druid.java.util.common.guava.YieldingSequenceBase; +import org.apache.druid.java.util.common.logger.Logger; +import org.apache.druid.query.context.ResponseContext; +import org.apache.druid.query.context.ResponseContext.Key; +import org.apache.druid.segment.SegmentMissingException; + +import java.util.ArrayList; +import java.util.Collections; +import java.util.Iterator; +import java.util.List; +import java.util.NoSuchElementException; +import java.util.concurrent.ConcurrentHashMap; +import java.util.function.BiFunction; + +public class RetryQueryRunner<T> implements QueryRunner<T> +{ + private static final Logger LOG = new Logger(RetryQueryRunner.class); + + private final QueryRunner<T> baseRunner; + private final BiFunction<Query<T>, List<SegmentDescriptor>, QueryRunner<T>> retryRunnerCreateFn; + private final RetryQueryRunnerConfig config; + private final ObjectMapper jsonMapper; + + /** + * Runnable executed after the broker creates query distribution tree for the first attempt. This is only + * for testing and must not be used in production code. + */ + private final Runnable runnableAfterFirstAttempt; + + private int totalNumRetries; + + public RetryQueryRunner( + QueryRunner<T> baseRunner, + BiFunction<Query<T>, List<SegmentDescriptor>, QueryRunner<T>> retryRunnerCreateFn, + RetryQueryRunnerConfig config, + ObjectMapper jsonMapper + ) + { + this(baseRunner, retryRunnerCreateFn, config, jsonMapper, () -> {}); + } + + /** + * Constructor only for testing. + */ + @VisibleForTesting + RetryQueryRunner( + QueryRunner<T> baseRunner, + BiFunction<Query<T>, List<SegmentDescriptor>, QueryRunner<T>> retryRunnerCreateFn, + RetryQueryRunnerConfig config, + ObjectMapper jsonMapper, + Runnable runnableAfterFirstAttempt + ) + { + this.baseRunner = baseRunner; + this.retryRunnerCreateFn = retryRunnerCreateFn; + this.config = config; + this.jsonMapper = jsonMapper; + this.runnableAfterFirstAttempt = runnableAfterFirstAttempt; + } + + @VisibleForTesting + int getTotalNumRetries() + { + return totalNumRetries; + } + + @Override + public Sequence<T> run(final QueryPlus<T> queryPlus, final ResponseContext context) + { + return new YieldingSequenceBase<T>() + { + @Override + public <OutType> Yielder<OutType> toYielder(OutType initValue, YieldingAccumulator<OutType, T> accumulator) + { + final Sequence<Sequence<T>> retryingSequence = new BaseSequence<>( + new IteratorMaker<Sequence<T>, RetryingSequenceIterator>() + { + @Override + public RetryingSequenceIterator make() + { + return new RetryingSequenceIterator(queryPlus, context, baseRunner, runnableAfterFirstAttempt); + } + + @Override + public void cleanup(RetryingSequenceIterator iterFromMake) + { + totalNumRetries = iterFromMake.retryCount; + } + } + ); + return new MergeSequence<>(queryPlus.getQuery().getResultOrdering(), retryingSequence) + .toYielder(initValue, accumulator); + } + }; + } + + private List<SegmentDescriptor> getMissingSegments(QueryPlus<T> queryPlus, final ResponseContext context) + { + // Sanity check before retrieving missingSegments from responseContext. + // The missingSegments in the responseContext is only valid when all servers have responded to the broker. + // The remainingResponses must be not null but 0 in the responseContext at this point. + final ConcurrentHashMap<String, Integer> idToRemainingResponses = + (ConcurrentHashMap<String, Integer>) Preconditions.checkNotNull( + context.get(Key.REMAINING_RESPONSES_FROM_QUERY_SERVERS), + "%s in responseContext", + Key.REMAINING_RESPONSES_FROM_QUERY_SERVERS.getName() + ); + + final int remainingResponses = Preconditions.checkNotNull( + idToRemainingResponses.get(queryPlus.getQuery().getMostRelevantId()), + "Number of remaining responses for query[%s]", + queryPlus.getQuery().getMostRelevantId() + ); + if (remainingResponses > 0) { + throw new ISE("Failed to check missing segments due to missing responds from [%d] servers", remainingResponses); + } + + final Object maybeMissingSegments = context.get(ResponseContext.Key.MISSING_SEGMENTS); + if (maybeMissingSegments == null) { + return Collections.emptyList(); + } + + return jsonMapper.convertValue( + maybeMissingSegments, + new TypeReference<List<SegmentDescriptor>>() + { + } + ); + } + + /** + * A lazy iterator populating {@link Sequence} by retrying the query. The first returned sequence is always the base + * sequence from the baseQueryRunner. Subsequent sequences are created dynamically whenever it retries the query. All + * the sequences populated by this iterator will be merged (not combined) with the base sequence. + * + * The design of this iterator depends on how {@link MergeSequence} works; the MergeSequence pops an item from + * each underlying sequence and pushes them to a {@link java.util.PriorityQueue}. Whenever it pops from the queue, + * it pushes a new item from the sequence where the returned item was originally from. Since the first returned + * sequence from this iterator is always the base sequence, the MergeSequence will call {@link Sequence#toYielder} + * on the base sequence first which in turn initializing query distribution tree. Once this tree is built, the query + * servers (historicals and realtime tasks) will lock all segments to read and report missing segments to the broker. + * If there are missing segments reported, this iterator will rewrite the query with those reported segments and + * reissue the rewritten query. + * + * @see org.apache.druid.client.CachingClusteredClient + * @see org.apache.druid.client.DirectDruidClient + */ + private class RetryingSequenceIterator implements Iterator<Sequence<T>> + { + private final QueryPlus<T> queryPlus; + private final ResponseContext context; + private final QueryRunner<T> baseQueryRunner; + private final Runnable runnableAfterFirstAttempt; + + private boolean first = true; + private Sequence<T> sequence = null; + private int retryCount = 0; + + private RetryingSequenceIterator( + QueryPlus<T> queryPlus, + ResponseContext context, + QueryRunner<T> baseQueryRunner, + Runnable runnableAfterFirstAttempt + ) + { + this.queryPlus = queryPlus; + this.context = context; + this.baseQueryRunner = baseQueryRunner; + this.runnableAfterFirstAttempt = runnableAfterFirstAttempt; + } + + @Override + public boolean hasNext() + { + if (first) { + sequence = baseQueryRunner.run(queryPlus, context); + // runnableAfterFirstAttempt is only for testing, it must be no-op for production code. + runnableAfterFirstAttempt.run(); Review comment: I suppose we can agree to disagree; I don't feel super strongly about this particular case since it's not a hot iterator, but I still think the test code in here complicates things overall and makes it harder to casually know what's going on, and wouldn't have done this in this way. ########## File path: server/src/main/java/org/apache/druid/client/CachingClusteredClient.java ########## @@ -283,13 +301,23 @@ public CachingClusteredClient( return contextBuilder.build(); } - Sequence<T> run(final UnaryOperator<TimelineLookup<String, ServerSelector>> timelineConverter) + /** + * Builds a query distribution and merge plan. + * + * This method returns an empty sequence if the query datasource is unknown or there is matching result-level cache. + * Otherwise, it creates a sequence merging sequences from the regular broker cache and remote servers. If parallel + * merge is enabled, it can merge and *combine* the underlying sequences in parallel. + * + * @return a pair of a sequence merging results from remote query servers and the number of remote servers + * participating in query processing. + */ + NonnullPair<Sequence<T>, Integer> run(final UnaryOperator<TimelineLookup<String, ServerSelector>> timelineConverter) Review comment: nit: I think a dedicated type would be more appropriate than using a generic Pair, since it makes it a lot easier to follow what is going on than having to keep a mental model of what `lhs` and `rhs` are. Something like: ``` class ClusterQueryResults<T> { Sequence<T> results; int numServers; } ``` This would be much easier to understand, and has the added benefit of when we realize we need a 3rd thing we can just add it. ---------------------------------------------------------------- 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: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@druid.apache.org For additional commands, e-mail: commits-h...@druid.apache.org