abhishekrb19 commented on code in PR #19689: URL: https://github.com/apache/druid/pull/19689#discussion_r3594818165
########## indexing-service/src/main/java/org/apache/druid/indexing/seekablestream/DimensionValueSetCollector.java: ########## @@ -0,0 +1,191 @@ +/* + * 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.indexing.seekablestream; + +import com.google.common.collect.Sets; +import org.apache.druid.data.input.InputRow; +import org.apache.druid.java.util.common.logger.Logger; +import org.apache.druid.timeline.DataSegment; +import org.apache.druid.timeline.SegmentId; +import org.apache.druid.timeline.partition.DimensionValueSetShardSpec; + +import javax.annotation.Nullable; +import java.util.ArrayList; +import java.util.Collection; +import java.util.Collections; +import java.util.Comparator; +import java.util.HashMap; +import java.util.HashSet; +import java.util.List; +import java.util.Map; +import java.util.Set; +import java.util.concurrent.ConcurrentHashMap; + +/** + * {@link StreamingShardSpecCollector} for {@link DimensionValueSetPartitionsSpec}: it records, per segment, the + * distinct values observed for each configured dimension and stamps each segment with a + * {@link DimensionValueSetShardSpec} at publish time so the broker can prune it. + * + * <p>A {@code null} element denotes an observed null/missing value (kept distinct from {@code ""}) so that + * {@code IS NULL} queries are not pruned. + * + * <p>Thread-safety follows the {@link StreamingShardSpecCollector} contract: + * {@link #observedPartitionDimValuesBySegment} is a + * {@link ConcurrentHashMap} keyed by {@link SegmentId} whose value sets are {@link Collections#synchronizedSet} written + * by the run loop ({@link #collect}) and snapshotted under their own monitor by the publish path ({@link #annotate}). + */ +public class DimensionValueSetCollector implements StreamingShardSpecCollector +{ + private static final Logger log = new Logger(DimensionValueSetCollector.class); + + private final List<String> partitionDimensions; + + /** + * If non-null, the maximum number of distinct values to record for a single dimension on a single segment. A segment + * whose observed values for a dimension exceed this cap omits that dimension from its stamped filter map (pruning + * disabled for it on that segment), guarding against shard-spec bloat. + */ + @Nullable + private final Integer maxValuesPerDimension; + + /** + * Observed values per tracked dimension, keyed by segment identifier. Inner sets permit null and are written by the + * run loop / read by the publish path under their own monitor. Entries are removed on successful publish via + * {@link #onSegmentPublished}; a publish failure is terminal for the task, so any remaining entries are reclaimed at + * task teardown. + */ + private final ConcurrentHashMap<SegmentId, Map<String, Set<String>>> observedPartitionDimValuesBySegment = + new ConcurrentHashMap<>(); + + /** + * Segment identifiers restored from disk at startup (i.e. spanning a task restart). Their pre-restart rows are not + * re-read, so {@link #observedPartitionDimValuesBySegment} would under-include values; to avoid wrongly pruning them, + * such segments are published with an empty-filter (non-pruning) {@link DimensionValueSetShardSpec} instead of one + * declaring observed values. + */ + private final Set<SegmentId> restartSpannedSegments = Sets.newConcurrentHashSet(); + + public DimensionValueSetCollector(List<String> partitionDimensions, @Nullable Integer maxValuesPerDimension) + { + this.partitionDimensions = partitionDimensions; + this.maxValuesPerDimension = maxValuesPerDimension; + } + + @Override + public void collect(SegmentId segmentId, InputRow row) + { + if (partitionDimensions.isEmpty()) { + return; + } + final Map<String, Set<String>> segValues = + observedPartitionDimValuesBySegment.computeIfAbsent(segmentId, k -> new ConcurrentHashMap<>()); + for (String dim : partitionDimensions) { + final Set<String> dimSet = segValues.computeIfAbsent( + dim, + k -> Collections.synchronizedSet(new HashSet<>()) + ); + // Empty getDimension result means a null/missing value; record null so IS NULL is not pruned + // (distinct from "", which getDimension returns as [""]). + final List<String> dimValues = row.getDimension(dim); + if (dimValues == null || dimValues.isEmpty()) { + dimSet.add(null); + } else { + dimSet.addAll(dimValues); + } + } + } + + @Override + public void onSegmentsRestored(Collection<SegmentId> segmentIds) + { + if (segmentIds.isEmpty()) { + return; + } + restartSpannedSegments.addAll(segmentIds); + log.warn( + "Disabling partition-filter pruning for %d segment(s) restored across a task restart: %s", + segmentIds.size(), + segmentIds + ); + } + + /** + * Stamps a segment with a {@link DimensionValueSetShardSpec} declaring its observed dimension values so the broker + * can prune it. We always return a {@link DimensionValueSetShardSpec}, falling back to an empty (non-pruning) filter + * map when values can't be safely declared, so segments in an interval stay class-uniform for + * {@link org.apache.druid.segment.realtime.appenderator.SegmentPublisherHelper}. A null observed value is carried + * through (distinct from {@code ""}) so {@code IS NULL} queries are not pruned. + */ + @Override + public DataSegment annotate(DataSegment s) + { + final Map<String, List<String>> snapshotFilters = new HashMap<>(); + final SegmentId lookupKey = s.getId(); + final Map<String, Set<String>> segObserved = observedPartitionDimValuesBySegment.get(lookupKey); + // Leave filters empty for restart-spanned segments: their pre-restart values can't be re-observed. + if (!restartSpannedSegments.contains(lookupKey) && segObserved != null) { + for (String dim : partitionDimensions) { + final Set<String> vals = segObserved.get(dim); + if (vals == null) { + continue; + } + // vals is a synchronized set written by the run loop; copy it under its monitor to iterate safely. + final List<String> snapshot; + synchronized (vals) { + if (vals.isEmpty()) { + continue; + } + // Over-cap: omit this dim from the stamped filter map (still a DimensionValueSetShardSpec for + // class-uniformity; possibleInDomain treats an absent dim as unconstrained, so pruning is disabled + // for it on this segment). + if (maxValuesPerDimension != null && vals.size() > maxValuesPerDimension) { Review Comment: > I wonder if it's worth somehow making this exclusion condition dependent on the frequency of the column values? Yes, frequency-based should be possible. I think it'd either require introducing a mode into this existing shard type or adding a new shard spec altogether (that resembles a bloom value set use case). -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
