klsince commented on code in PR #13347: URL: https://github.com/apache/pinot/pull/13347#discussion_r1709966912
########## pinot-segment-local/src/main/java/org/apache/pinot/segment/local/upsert/ConcurrentMapTableUpsertMetadataManagerForConsistentDeletes.java: ########## @@ -0,0 +1,75 @@ +/** + * 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.pinot.segment.local.upsert; + +import java.io.IOException; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.concurrent.ConcurrentHashMap; +import org.apache.pinot.segment.spi.SegmentContext; + + +/** + * Implementation of {@link TableUpsertMetadataManager} that is backed by a {@link ConcurrentHashMap} and ensures + * consistent deletions. This should be used when the table is configured with 'consistentDeletes' set to true. + */ +public class ConcurrentMapTableUpsertMetadataManagerForConsistentDeletes extends BaseTableUpsertMetadataManager { Review Comment: Consider to extract a base class? ########## pinot-segment-local/src/main/java/org/apache/pinot/segment/local/upsert/ConcurrentMapPartitionUpsertMetadataManagerForConsistentDeletes.java: ########## @@ -0,0 +1,385 @@ +/** + * 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.pinot.segment.local.upsert; + +import com.google.common.annotations.VisibleForTesting; +import java.util.HashMap; +import java.util.HashSet; +import java.util.Iterator; +import java.util.Map; +import java.util.Objects; +import java.util.Set; +import java.util.concurrent.ConcurrentHashMap; +import java.util.concurrent.atomic.AtomicBoolean; +import java.util.concurrent.atomic.AtomicInteger; +import javax.annotation.Nullable; +import javax.annotation.concurrent.ThreadSafe; +import org.apache.pinot.common.metrics.ServerMeter; +import org.apache.pinot.segment.local.indexsegment.immutable.ImmutableSegmentImpl; +import org.apache.pinot.segment.local.segment.readers.LazyRow; +import org.apache.pinot.segment.local.utils.HashUtils; +import org.apache.pinot.segment.spi.ImmutableSegment; +import org.apache.pinot.segment.spi.IndexSegment; +import org.apache.pinot.segment.spi.MutableSegment; +import org.apache.pinot.segment.spi.index.mutable.ThreadSafeMutableRoaringBitmap; +import org.apache.pinot.spi.data.readers.GenericRow; +import org.apache.pinot.spi.data.readers.PrimaryKey; +import org.roaringbitmap.buffer.MutableRoaringBitmap; + + +/** + * Implementation of {@link PartitionUpsertMetadataManager} that is backed by a {@link ConcurrentHashMap} and ensures + * consistent deletions. This should be used when the table is configured with 'enableConsistentDeletes' set to true. + * + * Consistent deletion ensures that when deletedKeysTTL is enabled with UpsertCompaction, the key metadata is + * removed from the HashMap only after all other records in the old segments are compacted. This guarantees + * data consistency. Without this, there can be a scenario where a deleted record is compacted first, while an + * old record remains non-compacted in a previous segment. During a server restart, this could lead to the old + * record reappearing. For the end-user, this would result in a data loss or inconsistency scenario, as the + * record was marked for deletion. + */ +@SuppressWarnings({"rawtypes", "unchecked"}) +@ThreadSafe +public class ConcurrentMapPartitionUpsertMetadataManagerForConsistentDeletes + extends BasePartitionUpsertMetadataManager { + + @VisibleForTesting + final ConcurrentHashMap<Object, ConcurrentMapPartitionUpsertMetadataManagerForConsistentDeletes.RecordLocation> + _primaryKeyToRecordLocationMap = new ConcurrentHashMap<>(); + // Used to initialize a reference to previous row for merging in partial upsert + private final LazyRow _reusePreviousRow = new LazyRow(); + private final Map<String, Object> _reuseMergeResultHolder = new HashMap<>(); + + public ConcurrentMapPartitionUpsertMetadataManagerForConsistentDeletes(String tableNameWithType, int partitionId, + UpsertContext context) { + super(tableNameWithType, partitionId, context); + } + + @Override + protected long getNumPrimaryKeys() { + return _primaryKeyToRecordLocationMap.size(); + } + + @Override + protected void doAddOrReplaceSegment(ImmutableSegmentImpl segment, ThreadSafeMutableRoaringBitmap validDocIds, + @Nullable ThreadSafeMutableRoaringBitmap queryableDocIds, Iterator<RecordInfo> recordInfoIterator, + @Nullable IndexSegment oldSegment, @Nullable MutableRoaringBitmap validDocIdsForOldSegment) { + String segmentName = segment.getSegmentName(); + segment.enableUpsert(this, validDocIds, queryableDocIds); + + AtomicInteger numKeysInWrongSegment = new AtomicInteger(); + while (recordInfoIterator.hasNext()) { + RecordInfo recordInfo = recordInfoIterator.next(); + int newDocId = recordInfo.getDocId(); + Comparable newComparisonValue = recordInfo.getComparisonValue(); + _primaryKeyToRecordLocationMap.compute(HashUtils.hashPrimaryKey(recordInfo.getPrimaryKey(), _hashFunction), + (primaryKey, currentRecordLocation) -> { + if (currentRecordLocation != null) { + // Existing primary key + IndexSegment currentSegment = currentRecordLocation.getSegment(); + int currentDocId = currentRecordLocation.getDocId(); + Comparable currentComparisonValue = currentRecordLocation.getComparisonValue(); + int comparisonResult = newComparisonValue.compareTo(currentComparisonValue); + int currentDistinctSegmentCount = currentRecordLocation.getDistinctSegmentCount(); + + // The current record is in the same segment + // Update the record location when there is a tie to keep the newer record. Note that the record info + // iterator will return records with incremental doc ids. + if (currentSegment == segment) { + if (comparisonResult >= 0) { + replaceDocId(segment, validDocIds, queryableDocIds, currentDocId, newDocId, recordInfo); + return new ConcurrentMapPartitionUpsertMetadataManagerForConsistentDeletes.RecordLocation(segment, + newDocId, newComparisonValue, currentDistinctSegmentCount); + } else { + return currentRecordLocation; + } + } + + // The current record is in an old segment being replaced + // This could happen when committing a consuming segment, or reloading a completed segment. In this + // case, we want to update the record location when there is a tie because the record locations should + // point to the new added segment instead of the old segment being replaced. Also, do not update the valid + // doc ids for the old segment because it has not been replaced yet. We pass in an optional valid doc ids + // snapshot for the old segment, which can be updated and used to track the docs not replaced yet. + if (currentSegment == oldSegment) { + if (comparisonResult >= 0) { + addDocId(segment, validDocIds, queryableDocIds, newDocId, recordInfo); + if (validDocIdsForOldSegment != null) { + validDocIdsForOldSegment.remove(currentDocId); + } + return new RecordLocation(segment, newDocId, newComparisonValue, + RecordLocation.incrementSegmentCount(currentDistinctSegmentCount)); + } else { + return new RecordLocation(currentSegment, currentDocId, currentComparisonValue, + RecordLocation.incrementSegmentCount(currentDistinctSegmentCount)); + } + } + + // This should not happen because the previously replaced segment should have all keys removed. We still + // handle it here, and also track the number of keys not properly replaced previously. + String currentSegmentName = currentSegment.getSegmentName(); + if (currentSegmentName.equals(segmentName)) { + numKeysInWrongSegment.getAndIncrement(); + if (comparisonResult >= 0) { + addDocId(segment, validDocIds, queryableDocIds, newDocId, recordInfo); + return new RecordLocation(segment, newDocId, newComparisonValue, + RecordLocation.incrementSegmentCount(currentDistinctSegmentCount)); + } else { + return currentRecordLocation; + } + } + + // The current record is in a different segment + // Update the record location when getting a newer comparison value, or the value is the same as the + // current value, but the segment has a larger sequence number (the segment is newer than the current + // segment). + if (comparisonResult > 0 || (comparisonResult == 0 && shouldReplaceOnComparisonTie(segmentName, + currentSegmentName, segment.getSegmentMetadata().getIndexCreationTime(), + currentSegment.getSegmentMetadata().getIndexCreationTime()))) { + replaceDocId(segment, validDocIds, queryableDocIds, currentSegment, currentDocId, newDocId, recordInfo); + return new RecordLocation(segment, newDocId, newComparisonValue, + RecordLocation.incrementSegmentCount(currentDistinctSegmentCount)); + } else { + return new RecordLocation(currentSegment, currentDocId, currentComparisonValue, + RecordLocation.incrementSegmentCount(currentDistinctSegmentCount)); + } + } else { + // New primary key + addDocId(segment, validDocIds, queryableDocIds, newDocId, recordInfo); + return new RecordLocation(segment, newDocId, newComparisonValue, 1); + } + }); + } + int numKeys = numKeysInWrongSegment.get(); + if (numKeys > 0) { + _logger.warn("Found {} primary keys in the wrong segment when adding segment: {}", numKeys, segmentName); + _serverMetrics.addMeteredTableValue(_tableNameWithType, ServerMeter.UPSERT_KEYS_IN_WRONG_SEGMENT, numKeys); + } + } + + @Override + protected void addSegmentWithoutUpsert(ImmutableSegmentImpl segment, ThreadSafeMutableRoaringBitmap validDocIds, + @Nullable ThreadSafeMutableRoaringBitmap queryableDocIds, Iterator<RecordInfo> recordInfoIterator) { + throw new UnsupportedOperationException("Consistent-deletion does not support preloading of segments."); + } + + @Override + protected void doRemoveSegment(IndexSegment segment) { + String segmentName = segment.getSegmentName(); + _logger.info("Removing {} segment: {}, current primary key count: {}", + segment instanceof ImmutableSegment ? "immutable" : "mutable", segmentName, getNumPrimaryKeys()); + long startTimeMs = System.currentTimeMillis(); + + try ( + UpsertUtils.PrimaryKeyReader primaryKeyReader = new UpsertUtils.PrimaryKeyReader(segment, _primaryKeyColumns)) { + removeSegment(segment, + UpsertUtils.getPrimaryKeyIterator(primaryKeyReader, segment.getSegmentMetadata().getTotalDocs())); + } catch (Exception e) { + throw new RuntimeException( + String.format("Caught exception while removing segment: %s, table: %s", segment.getSegmentName(), + _tableNameWithType), e); + } + + // Update metrics + long numPrimaryKeys = getNumPrimaryKeys(); + updatePrimaryKeyGauge(numPrimaryKeys); + _logger.info("Finished removing segment: {} in {}ms, current primary key count: {}", segmentName, + System.currentTimeMillis() - startTimeMs, numPrimaryKeys); + } + + @Override + protected void removeSegment(IndexSegment segment, Iterator<PrimaryKey> primaryKeyIterator) { + Set<Object> deDuplicatedPrimaryKeys = new HashSet<>(); + while (primaryKeyIterator.hasNext()) { + PrimaryKey primaryKey = primaryKeyIterator.next(); + _primaryKeyToRecordLocationMap.computeIfPresent(HashUtils.hashPrimaryKey(primaryKey, _hashFunction), + (pk, recordLocation) -> { + if (deDuplicatedPrimaryKeys.contains(pk)) { + return recordLocation; + } + deDuplicatedPrimaryKeys.add(pk); + if (recordLocation.getSegment() == segment) { + return null; + } + return new RecordLocation(recordLocation.getSegment(), recordLocation.getDocId(), + recordLocation.getComparisonValue(), + RecordLocation.decrementSegmentCount(recordLocation.getDistinctSegmentCount())); + }); + } + } + + @Override + public void doRemoveExpiredPrimaryKeys() { + AtomicInteger numDeletedTTLKeysRemoved = new AtomicInteger(); + double largestSeenComparisonValue = _largestSeenComparisonValue.get(); + double deletedKeysThreshold; + if (_deletedKeysTTL > 0) { + deletedKeysThreshold = largestSeenComparisonValue - _deletedKeysTTL; + } else { + deletedKeysThreshold = Double.MIN_VALUE; + } + + _primaryKeyToRecordLocationMap.forEach((primaryKey, recordLocation) -> { + double comparisonValue = ((Number) recordLocation.getComparisonValue()).doubleValue(); + if (_deletedKeysTTL > 0 && comparisonValue < deletedKeysThreshold + && recordLocation.getDistinctSegmentCount() <= 1) { Review Comment: perhaps add some comments on this check, or add comment on the disctinctSegmentCount field in recordLocation that it's used here, as it's kinda critical for the correctness of this feature. also looks like a lot of methods are same with ConcurrentMapPartitionUpsertMetadataManager. How about extracting a base class like BaseConcurrentMapPartitionUpsertMetadataManager, for both to reuse. -- 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: commits-unsubscr...@pinot.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@pinot.apache.org For additional commands, e-mail: commits-h...@pinot.apache.org
