YuweiXiao commented on code in PR #4958: URL: https://github.com/apache/hudi/pull/4958#discussion_r917392986
########## hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/client/clustering/plan/strategy/SparkConsistentBucketClusteringPlanStrategy.java: ########## @@ -0,0 +1,332 @@ +/* + * 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.hudi.client.clustering.plan.strategy; + +import org.apache.hudi.avro.model.HoodieClusteringGroup; +import org.apache.hudi.client.WriteStatus; +import org.apache.hudi.common.engine.HoodieEngineContext; +import org.apache.hudi.common.fs.FSUtils; +import org.apache.hudi.common.model.ConsistentHashingNode; +import org.apache.hudi.common.model.FileSlice; +import org.apache.hudi.common.model.HoodieConsistentHashingMetadata; +import org.apache.hudi.common.model.HoodieFileFormat; +import org.apache.hudi.common.model.HoodieKey; +import org.apache.hudi.common.model.HoodieRecord; +import org.apache.hudi.common.model.HoodieRecordPayload; +import org.apache.hudi.common.table.timeline.HoodieTimeline; +import org.apache.hudi.common.table.view.SyncableFileSystemView; +import org.apache.hudi.common.util.Option; +import org.apache.hudi.common.util.StringUtils; +import org.apache.hudi.common.util.ValidationUtils; +import org.apache.hudi.common.util.collection.Triple; +import org.apache.hudi.config.HoodieClusteringConfig; +import org.apache.hudi.config.HoodieWriteConfig; +import org.apache.hudi.exception.HoodieClusteringException; +import org.apache.hudi.index.bucket.ConsistentBucketIdentifier; +import org.apache.hudi.index.bucket.HoodieSparkConsistentBucketIndex; +import org.apache.hudi.table.HoodieTable; +import org.apache.hudi.table.action.cluster.strategy.PartitionAwareClusteringPlanStrategy; + +import org.apache.log4j.LogManager; +import org.apache.log4j.Logger; +import org.apache.spark.api.java.JavaRDD; + +import java.io.IOException; +import java.util.ArrayList; +import java.util.Collections; +import java.util.Comparator; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.stream.Collectors; +import java.util.stream.IntStream; +import java.util.stream.Stream; + +/** + * Clustering plan strategy specifically for consistent bucket index + */ +public class SparkConsistentBucketClusteringPlanStrategy<T extends HoodieRecordPayload<T>> + extends PartitionAwareClusteringPlanStrategy<T, JavaRDD<HoodieRecord<T>>, JavaRDD<HoodieKey>, JavaRDD<WriteStatus>> { + + private static final Logger LOG = LogManager.getLogger(SparkConsistentBucketClusteringPlanStrategy.class); + + public static final String METADATA_PARTITION_KEY = "clustering.group.partition"; + public static final String METADATA_CHILD_NODE_KEY = "clustering.group.child.node"; + public static final String METADATA_SEQUENCE_NUMBER_KEY = "clustering.group.sequence.no"; + + public SparkConsistentBucketClusteringPlanStrategy(HoodieTable table, HoodieEngineContext engineContext, HoodieWriteConfig writeConfig) { + super(table, engineContext, writeConfig); + validate(); + } + + /** + * TODO maybe add force config to schedule the clustering. It could allow clustering on partitions that are not doing write operation. + * Block clustering if there is any ongoing concurrent writers + * + * @return true if the schedule can proceed + */ + @Override + public boolean checkPrecondition() { + HoodieTimeline timeline = getHoodieTable().getActiveTimeline().getDeltaCommitTimeline().filterInflightsAndRequested(); + if (!timeline.empty()) { + LOG.warn("When using consistent bucket, clustering cannot be scheduled async if there are concurrent writers. " + + "Writer instant: " + timeline.getInstants().collect(Collectors.toList())); + return false; + } + return true; + } + + /** + * Generate candidate clustering file slices of the given partition. + * If there is inflight / requested clustering working on the partition, then return empty list + * to ensure serialized update to the hashing metadata. + * + * @return candidate file slices to be clustered (i.e., sort, bucket split or merge) + */ + @Override + protected Stream<FileSlice> getFileSlicesEligibleForClustering(String partition) { + SyncableFileSystemView fileSystemView = (SyncableFileSystemView) getHoodieTable().getSliceView(); + boolean isPartitionInClustering = fileSystemView.getFileGroupsInPendingClustering().anyMatch(p -> p.getLeft().getPartitionPath().equals(partition)); + if (isPartitionInClustering) { + LOG.info("Partition: " + partition + " is already in clustering, skip"); + return Stream.empty(); + } + + return super.getFileSlicesEligibleForClustering(partition); + } + + @Override + protected Map<String, String> getStrategyParams() { + Map<String, String> params = new HashMap<>(); + if (!StringUtils.isNullOrEmpty(getWriteConfig().getClusteringSortColumns())) { + params.put(HoodieClusteringConfig.PLAN_STRATEGY_SORT_COLUMNS.key(), getWriteConfig().getClusteringSortColumns()); + } + return params; + } + + /** + * Generate cluster group based on split, merge and sort rules + */ + @Override + protected Stream<HoodieClusteringGroup> buildClusteringGroupsForPartition(String partitionPath, List<FileSlice> fileSlices) { + ValidationUtils.checkArgument(getHoodieTable().getIndex() instanceof HoodieSparkConsistentBucketIndex, + "Mismatch of index type and the clustering strategy, index: " + getHoodieTable().getIndex().getClass().getSimpleName()); + HoodieConsistentHashingMetadata metadata = HoodieSparkConsistentBucketIndex.loadMetadata(getHoodieTable(), partitionPath); + ValidationUtils.checkArgument(metadata != null, "Metadata is null for partition: " + partitionPath); + ConsistentBucketIdentifier identifier = new ConsistentBucketIdentifier(metadata); + + // Apply split rule + int splitSlot = getWriteConfig().getBucketIndexMaxNumBuckets() - identifier.getNumBuckets(); + Triple<List<HoodieClusteringGroup>, Integer, List<FileSlice>> splitResult = + buildSplitClusteringGroups(identifier, fileSlices, splitSlot); + List<HoodieClusteringGroup> ret = new ArrayList<>(splitResult.getLeft()); + + // Apply merge rule + int mergeSlot = identifier.getNumBuckets() - getWriteConfig().getBucketIndexMinNumBuckets() + splitResult.getMiddle(); + Triple<List<HoodieClusteringGroup>, Integer, List<FileSlice>> mergeResult = + buildMergeClusteringGroup(identifier, splitResult.getRight(), mergeSlot); + ret.addAll(mergeResult.getLeft()); + + // Apply sort only to the remaining file groups + ret.addAll(mergeResult.getRight().stream().map(fs -> { + ConsistentHashingNode oldNode = identifier.getBucketByFileId(fs.getFileId()); + ConsistentHashingNode newNode = new ConsistentHashingNode(oldNode.getValue(), FSUtils.createNewFileIdPfx(), ConsistentHashingNode.NodeTag.REPLACE); + return HoodieClusteringGroup.newBuilder() + .setSlices(getFileSliceInfo(Collections.singletonList(fs))) + .setNumOutputFileGroups(1) + .setMetrics(buildMetrics(Collections.singletonList(fs))) + .setExtraMetadata(constructExtraMetadata(fs.getPartitionPath(), Collections.singletonList(newNode), identifier.getMetadata().getSeqNo())) + .build(); + }).collect(Collectors.toList())); + + return ret.stream(); + } + + /** + * Generate clustering groups according to split rules。 + * Currently, we always split bucket into two sub-buckets. + * + * @param identifier bucket identifier + * @param fileSlices file slice candidate to be built as split clustering groups + * @param splitSlot number of new bucket allowed to produce, in order to constrain the upper bound of the total number of bucket + * @return list of clustering group, number of new buckets generated, remaining file slice (that does not split) + */ + protected Triple<List<HoodieClusteringGroup>, Integer, List<FileSlice>> buildSplitClusteringGroups( + ConsistentBucketIdentifier identifier, List<FileSlice> fileSlices, int splitSlot) { + List<HoodieClusteringGroup> retGroup = new ArrayList<>(); + List<FileSlice> fsUntouched = new ArrayList<>(); + long splitSize = getSplitSize(); + int remainingSplitSlot = splitSlot; + for (FileSlice fs : fileSlices) { + boolean needSplit = fs.getTotalFileSize() > splitSize; + if (!needSplit || remainingSplitSlot == 0) { + fsUntouched.add(fs); + continue; + } + + Option<List<ConsistentHashingNode>> nodes = identifier.splitBucket(fs.getFileId()); + + // Bucket cannot be split + if (!nodes.isPresent()) { + fsUntouched.add(fs); + continue; + } + + remainingSplitSlot--; + List<FileSlice> fsList = Collections.singletonList(fs); + retGroup.add(HoodieClusteringGroup.newBuilder() + .setSlices(getFileSliceInfo(fsList)) + .setNumOutputFileGroups(2) + .setMetrics(buildMetrics(fsList)) + .setExtraMetadata(constructExtraMetadata(fs.getPartitionPath(), nodes.get(), identifier.getMetadata().getSeqNo())) + .build()); + } + return Triple.of(retGroup, splitSlot - remainingSplitSlot, fsUntouched); + } + + /** + * Generate clustering group according to merge rules + * + * @param identifier bucket identifier + * @param fileSlices file slice candidates to be built as merge clustering groups + * @param mergeSlot number of bucket allowed to be merged, in order to guarantee the lower bound of the total number of bucket + * @return list of clustering group, number of buckets merged (removed), remaining file slice (that does not be merged) + */ + protected Triple<List<HoodieClusteringGroup>, Integer, List<FileSlice>> buildMergeClusteringGroup( + ConsistentBucketIdentifier identifier, List<FileSlice> fileSlices, int mergeSlot) { + if (fileSlices.size() <= 1) { + return Triple.of(Collections.emptyList(), 0, fileSlices); + } + + long mergeSize = getMergeSize(); + int remainingMergeSlot = mergeSlot; + List<HoodieClusteringGroup> groups = new ArrayList<>(); + boolean[] added = new boolean[fileSlices.size()]; + + fileSlices.sort(Comparator.comparingInt(a -> identifier.getBucketByFileId(a.getFileId()).getValue())); + // In each round, we check if the ith file slice can be merged with its predecessors and successors + for (int i = 0; i < fileSlices.size(); ++i) { + if (added[i] || fileSlices.get(i).getTotalFileSize() > mergeSize) { + continue; + } + + int startIdx = i; + int endIdx = i; + // Backward check + long totalSize = fileSlices.get(i).getTotalFileSize(); + do { + int preIdx = startIdx >= 1 ? startIdx - 1 : fileSlices.size() - 1; + boolean isNeighbour = identifier.getBucketByFileId(fileSlices.get(preIdx).getFileId()) == identifier.getFormerBucket(fileSlices.get(startIdx).getFileId()); + /** + * Merge condition: + * 1. there is still slot to merge bucket + * 2. the previous file slices is not merged + * 3. the previous file slice and current file slice are neighbour in the hash ring + * 4. Both the total file size up to now and the previous file slice size are smaller than merge size threshold + */ + if (remainingMergeSlot == 0 || added[preIdx] || !isNeighbour || totalSize > mergeSize || fileSlices.get(preIdx).getTotalFileSize() > mergeSize) { + break; + } + + // Mark preIdx as merge candidate + totalSize += fileSlices.get(preIdx).getTotalFileSize(); + startIdx = preIdx; + remainingMergeSlot--; + } while (startIdx != i); + + // Forward check + do { + int nextIdx = endIdx + 1 < fileSlices.size() ? endIdx + 1 : 0; + boolean isNeighbour = identifier.getBucketByFileId(fileSlices.get(endIdx).getFileId()) == identifier.getFormerBucket(fileSlices.get(nextIdx).getFileId()); + + if (remainingMergeSlot == 0 || added[nextIdx] || !isNeighbour || totalSize > mergeSize || fileSlices.get(nextIdx).getTotalFileSize() > mergeSize) { + break; + } + + // Mark nextIdx as merge candidate + totalSize += fileSlices.get(nextIdx).getTotalFileSize(); + endIdx = nextIdx; + remainingMergeSlot--; Review Comment: Yeah, sure! -- 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]
