jtao15 commented on a change in pull request #7178: URL: https://github.com/apache/pinot/pull/7178#discussion_r685401281
########## File path: pinot-plugins/pinot-minion-tasks/pinot-minion-builtin-tasks/src/main/java/org/apache/pinot/plugin/minion/tasks/merge_rollup/MergeRollupTaskGenerator.java ########## @@ -0,0 +1,491 @@ +/** + * 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.plugin.minion.tasks.merge_rollup; + +import com.google.common.base.Preconditions; +import java.util.ArrayList; +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.TreeMap; +import java.util.TreeSet; +import org.I0Itec.zkclient.exception.ZkException; +import org.apache.commons.lang3.StringUtils; +import org.apache.helix.ZNRecord; +import org.apache.helix.task.TaskState; +import org.apache.pinot.common.lineage.SegmentLineage; +import org.apache.pinot.common.lineage.SegmentLineageUtils; +import org.apache.pinot.common.metadata.segment.OfflineSegmentZKMetadata; +import org.apache.pinot.common.minion.MergeRollupTaskMetadata; +import org.apache.pinot.controller.helix.core.minion.ClusterInfoAccessor; +import org.apache.pinot.controller.helix.core.minion.generator.PinotTaskGenerator; +import org.apache.pinot.controller.helix.core.minion.generator.TaskGeneratorUtils; +import org.apache.pinot.core.common.MinionConstants; +import org.apache.pinot.core.common.MinionConstants.MergeRollupTask; +import org.apache.pinot.core.minion.PinotTaskConfig; +import org.apache.pinot.spi.annotations.minion.TaskGenerator; +import org.apache.pinot.spi.config.table.ColumnPartitionConfig; +import org.apache.pinot.spi.config.table.SegmentPartitionConfig; +import org.apache.pinot.spi.config.table.TableConfig; +import org.apache.pinot.spi.config.table.TableTaskConfig; +import org.apache.pinot.spi.config.table.TableType; +import org.apache.pinot.spi.utils.IngestionConfigUtils; +import org.apache.pinot.spi.utils.TimeUtils; +import org.apache.pinot.spi.utils.builder.TableNameBuilder; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + + +/** + * A {@link PinotTaskGenerator} implementation for generating tasks of type {@link MergeRollupTask} + * + * TODO: Add the support for realtime table + * + * Steps: + * + * - Pre-select segments: + * - Fetch all segments, select segments based on segment lineage (removing segmentsFrom for COMPLETED lineage entry and + * segmentsTo for IN_PROGRESS lineage entry) + * - Sort segments based on startTime and endTime in ascending order + * + * For each mergeLevel (from lowest to highest, e.g. Hourly -> Daily -> Monthly -> Yearly): + * - Skip scheduling if there's incomplete task for the mergeLevel + * - Calculate merge/rollup window: + * - Read watermarkMs from the {@link MergeRollupTaskMetadata} ZNode + * found at MINION_TASK_METADATA/MergeRollupTask/<tableNameWithType> + * In case of cold-start, no ZNode will exist. + * A new ZNode will be created, with watermarkMs as the smallest time found in all segments truncated to the + * closest bucket start time. + * - The execution window for the task is calculated as, + * windowStartMs = watermarkMs, windowEndMs = windowStartMs + bucketTimeMs + * - Skip scheduling if the window is invalid: + * - If the execution window is not older than bufferTimeMs, no task will be generated + * - The windowEndMs of higher merge level should be less or equal to the waterMarkMs of lower level + * - Bump up target window and watermark if needed. + * - If there's no unmerged segments (by checking segment zk metadata {mergeRollupTask.mergeLevel: level}) for current window, + * keep bumping up the watermark and target window until unmerged segments are found. Else skip the scheduling. + * - Select all segments for the target window + * - Create tasks (per partition for partitioned table) based on maxNumRecordsPerTask + */ +@TaskGenerator +public class MergeRollupTaskGenerator implements PinotTaskGenerator { + private static final int DEFAULT_MAX_NUM_RECORDS_PER_TASK = 50_000_000; + private static final String REFRESH = "REFRESH"; + private static final Logger LOGGER = LoggerFactory.getLogger(MergeRollupTaskGenerator.class); + + private ClusterInfoAccessor _clusterInfoAccessor; + + @Override + public void init(ClusterInfoAccessor clusterInfoAccessor) { + _clusterInfoAccessor = clusterInfoAccessor; + } + + @Override + public String getTaskType() { + return MergeRollupTask.TASK_TYPE; + } + + @Override + public List<PinotTaskConfig> generateTasks(List<TableConfig> tableConfigs) { + String taskType = MergeRollupTask.TASK_TYPE; + List<PinotTaskConfig> pinotTaskConfigs = new ArrayList<>(); + + for (TableConfig tableConfig : tableConfigs) { + if (!validate(tableConfig, taskType)) { + continue; + } + String offlineTableName = tableConfig.getTableName(); + LOGGER.info("Start generating task configs for table: {} for task: {}", offlineTableName, taskType); + + // Get all segment metadata + Set<OfflineSegmentZKMetadata> allSegments = + new HashSet<>(_clusterInfoAccessor.getOfflineSegmentsMetadata(offlineTableName)); + + // Select current segment snapshot based on lineage + SegmentLineage segmentLineage = _clusterInfoAccessor.getSegmentLineage(offlineTableName); + Set<String> preSelectedSegmentsBasedOnLineage = new HashSet<>(); + for (OfflineSegmentZKMetadata offlineSegmentZKMetadata : allSegments) { + preSelectedSegmentsBasedOnLineage.add(offlineSegmentZKMetadata.getSegmentName()); + } + SegmentLineageUtils.filterSegmentsBasedOnLineageInplace(preSelectedSegmentsBasedOnLineage, segmentLineage); + + // Sort segments based on startTimeMs, endTimeMs and segmentName in ascending order + TreeSet<OfflineSegmentZKMetadata> preSelectedSegments = new TreeSet<>((a, b) -> { + long aStartTime = a.getStartTimeMs(); + long bStartTime = b.getStartTimeMs(); + if (aStartTime != bStartTime) { + return Long.compare(aStartTime, bStartTime); + } + long aEndTime = a.getEndTimeMs(); + long bEndTime = b.getEndTimeMs(); + return aEndTime != bEndTime ? Long.compare(aEndTime, bEndTime) + : a.getSegmentName().compareTo(b.getSegmentName()); + }); + for (OfflineSegmentZKMetadata segment : allSegments) { + if (preSelectedSegmentsBasedOnLineage.contains(segment.getSegmentName())) { + preSelectedSegments.add(segment); + } + } + + if (preSelectedSegments.isEmpty()) { + LOGGER.info("Skip generating task: {} for table: {}, no segment is found to merge.", taskType, + offlineTableName); + continue; + } + + // Sort merge levels based on bucket time period + Map<String, String> taskConfigs = tableConfig.getTaskConfig().getConfigsForTaskType(taskType); + Map<String, Map<String, String>> mergeLevelToConfigs = MergeRollupTaskUtils.getLevelToConfigMap(taskConfigs); + List<Map.Entry<String, Map<String, String>>> sortedMergeLevelConfigs = + new ArrayList<>(mergeLevelToConfigs.entrySet()); + sortedMergeLevelConfigs.sort(Comparator.comparingLong( + e -> TimeUtils.convertPeriodToMillis(e.getValue().get(MinionConstants.MergeTask.BUCKET_TIME_PERIOD_KEY)))); + + // Get incomplete merge levels + Set<String> inCompleteMergeLevels = new HashSet<>(); + for (Map.Entry<String, TaskState> entry : TaskGeneratorUtils.getIncompleteTasks(taskType, offlineTableName, + _clusterInfoAccessor).entrySet()) { + for (PinotTaskConfig taskConfig : _clusterInfoAccessor.getTaskConfigs(entry.getKey())) { + inCompleteMergeLevels.add(taskConfig.getConfigs().get(MergeRollupTask.MERGE_LEVEL_KEY)); + } + } + + ZNRecord mergeRollupTaskZNRecord = _clusterInfoAccessor.getMinionMergeRollupTaskZNRecord(offlineTableName); + int expectedVersion = mergeRollupTaskZNRecord != null ? mergeRollupTaskZNRecord.getVersion() : -1; + MergeRollupTaskMetadata mergeRollupTaskMetadata = + mergeRollupTaskZNRecord != null ? MergeRollupTaskMetadata.fromZNRecord(mergeRollupTaskZNRecord) + : new MergeRollupTaskMetadata(offlineTableName, new TreeMap<>()); + List<PinotTaskConfig> pinotTaskConfigsForTable = new ArrayList<>(); + + // Schedule tasks from lowest to highest merge level (e.g. Hourly -> Daily -> Monthly -> Yearly) + for (int i = 0; i < sortedMergeLevelConfigs.size(); i++) { + String lowerMergeLevel = i > 0 ? sortedMergeLevelConfigs.get(i - 1).getKey() : null; + String mergeLevel = sortedMergeLevelConfigs.get(i).getKey(); + Map<String, String> mergeConfigs = sortedMergeLevelConfigs.get(i).getValue(); + + // Skip scheduling if there's incomplete task for current mergeLevel + if (inCompleteMergeLevels.contains(mergeLevel)) { + LOGGER.info("Found incomplete task of merge level: {} for the same table: {}, Skipping task generation: {}", + mergeLevel, offlineTableName, taskType); + continue; + } + + // Get the bucket size and buffer + long bucketMs = + TimeUtils.convertPeriodToMillis(mergeConfigs.get(MinionConstants.MergeTask.BUCKET_TIME_PERIOD_KEY)); + long bufferMs = + TimeUtils.convertPeriodToMillis(mergeConfigs.get(MinionConstants.MergeTask.BUFFER_TIME_PERIOD_KEY)); + + // Get watermark from MergeRollupTaskMetadata ZNode + // windowStartMs = watermarkMs, windowEndMs = windowStartMs + bucketTimeMs + long waterMarkMs = + getWatermarkMs(preSelectedSegments.first().getStartTimeMs(), bucketMs, mergeLevel, mergeRollupTaskMetadata); + long windowStartMs = waterMarkMs; + long windowEndMs = windowStartMs + bucketMs; + + if (!isValidMergeWindowEndTime(windowEndMs, bufferMs, lowerMergeLevel, mergeRollupTaskMetadata)) { + LOGGER.info( + "Window with start: {} and end: {} of mergeLevel: {} is not a valid merge window, Skipping task generation: {}", + windowStartMs, windowEndMs, mergeLevel, taskType); + continue; + } + + // Find all segments overlapping with the merge window, if all overlapping segments are merged, bump up the target window + List<OfflineSegmentZKMetadata> selectedSegments = new ArrayList<>(); + boolean hasUnMergedSegments = false; + boolean isValidMergeWindow = true; + + // The for loop terminates in following cases: + // 1. Found unmerged segments in target merge window, need to bump up watermark if windowStartMs > watermarkMs, + // will schedule tasks + // 2. All segments are merged in the merge window and we have loop through all segments, need to bump up + // watermark to windowStartMs + bucketTimeMs, skip scheduling + // 3. Current watermark > endTimeMs of all segments, won't bump up watermark, skip scheduling + // 4. Merge window is invalid (windowEndMs > System.currentTimeMillis() - bufferMs || windowEndMs > waterMark of + // the lower mergeLevel), may need to bump up watermark if windowStartMs > watermarkMs, skip scheduling + for (OfflineSegmentZKMetadata preSelectedSegment : preSelectedSegments) { + if (windowStartMs <= preSelectedSegment.getEndTimeMs() && preSelectedSegment.getStartTimeMs() < windowEndMs) { + // For segments overlapping with current window, add to the result list + selectedSegments.add(preSelectedSegment); + if (!isMergedSegment(preSelectedSegment, mergeLevel)) { + hasUnMergedSegments = true; + } + } else if (windowEndMs <= preSelectedSegment.getStartTimeMs()) { + // Has gone through all overlapping segments for current window + if (hasUnMergedSegments) { + // Found unmerged segments, schedule merge task for current window + break; + } else { + // No unmerged segments found, clean up selected segments and bump up the merge window + // TODO: If there are many small merged segments, we should merge them again + selectedSegments = new ArrayList<>(); + selectedSegments.add(preSelectedSegment); + if (!isMergedSegment(preSelectedSegment, mergeLevel)) { + hasUnMergedSegments = true; + } + while (windowEndMs <= preSelectedSegment.getStartTimeMs()) { + LOGGER.debug("No unmerged segments found for window [{}, {}), Update window to: [{}. {})", + windowStartMs, windowEndMs, windowStartMs + bucketMs, windowEndMs + bucketMs); + windowStartMs += bucketMs; + windowEndMs += bucketMs; + if (!isValidMergeWindowEndTime(windowEndMs, bufferMs, lowerMergeLevel, mergeRollupTaskMetadata)) { + isValidMergeWindow = false; + break; + } + } + + if (!isValidMergeWindow) { + break; + } + } + } + // windowStartMs > preSelectedSegment.getEndTimeMs() + // Haven't find the first overlapping segment, continue to the next segment + } + + // Update watermark in 3 cases: + // 1. Cold start, no watermark value in MergeRollupTaskMetadata + // 2. Merge window start time > current watermark + // 3. All segments for the table are merged (for loop termination case 2), bump up watermark to the next window + if (mergeRollupTaskMetadata.getWatermarkMap().get(mergeLevel) == null || windowStartMs > waterMarkMs) { + updateWatermarkMs(offlineTableName, mergeLevel, windowStartMs, mergeRollupTaskMetadata); + } + if (isValidMergeWindow && !hasUnMergedSegments && selectedSegments.size() > 0) { + LOGGER.debug("No unmerged segments found for window [{}, {}), Update window to: [{}. {})", windowStartMs, + windowEndMs, windowStartMs + bucketMs, windowEndMs + bucketMs); + windowStartMs += bucketMs; + updateWatermarkMs(offlineTableName, mergeLevel, windowStartMs, mergeRollupTaskMetadata); + } + + if (!isValidMergeWindow) { + LOGGER.info( + "Window with start: {} and end: {} of mergeLevel: {} is not a valid merge window, Skipping task generation: {}", + windowStartMs, windowEndMs, mergeLevel, taskType); + continue; + } + + if (!hasUnMergedSegments) { + LOGGER.info("No unmerged segments found for mergeLevel:{} for table: {}, Skipping task generation: {}", + mergeLevel, offlineTableName, taskType); + continue; + } + + // Create task configs + int maxNumRecordsPerTask = + mergeConfigs.get(MergeRollupTask.MAX_NUM_RECORDS_PER_TASK_KEY) != null ? Integer.parseInt( + mergeConfigs.get(MergeRollupTask.MAX_NUM_RECORDS_PER_TASK_KEY)) : DEFAULT_MAX_NUM_RECORDS_PER_TASK; + SegmentPartitionConfig segmentPartitionConfig = tableConfig.getIndexingConfig().getSegmentPartitionConfig(); + if (segmentPartitionConfig == null) { + pinotTaskConfigsForTable.addAll( + createPinotTaskConfigs(selectedSegments, offlineTableName, maxNumRecordsPerTask, mergeLevel, mergeConfigs, + taskConfigs)); + } else { + // For partitioned table, schedule separate tasks for each partition + Map<String, ColumnPartitionConfig> columnPartitionMap = segmentPartitionConfig.getColumnPartitionMap(); + Preconditions.checkState(columnPartitionMap.size() == 1, "Cannot partition on multiple columns for table: %s", + tableConfig.getTableName()); + Map.Entry<String, ColumnPartitionConfig> partitionEntry = columnPartitionMap.entrySet().iterator().next(); + String partitionColumn = partitionEntry.getKey(); + + Map<Integer, List<OfflineSegmentZKMetadata>> partitionToSegments = new HashMap<>(); + List<OfflineSegmentZKMetadata> outlierSegments = + new ArrayList<>(); // Handle segments that have multiple partitions or no partition info + for (OfflineSegmentZKMetadata selectedSegment : selectedSegments) { + if (selectedSegment.getPartitionMetadata() == null + || selectedSegment.getPartitionMetadata().getPartitions(partitionColumn).size() > 1) { + outlierSegments.add(selectedSegment); + } else { + int partition = selectedSegment.getPartitionMetadata().getPartitions(partitionColumn).iterator().next(); + partitionToSegments.computeIfAbsent(partition, k -> new ArrayList<>()).add(selectedSegment); + } + } + + for (Map.Entry<Integer, List<OfflineSegmentZKMetadata>> partitionToSegmentsEntry : partitionToSegments.entrySet()) { + pinotTaskConfigsForTable.addAll( + createPinotTaskConfigs(partitionToSegmentsEntry.getValue(), offlineTableName, maxNumRecordsPerTask, + mergeLevel, mergeConfigs, taskConfigs)); + } + + if (!outlierSegments.isEmpty()) { + pinotTaskConfigsForTable.addAll( + createPinotTaskConfigs(outlierSegments, offlineTableName, maxNumRecordsPerTask, mergeLevel, + mergeConfigs, taskConfigs)); + } + } + } + + // Write updated watermark map to zookeeper + try { + _clusterInfoAccessor.setMergeRollupTaskMetadata(mergeRollupTaskMetadata, expectedVersion); + } catch (ZkException e) { + LOGGER.error( + "Version changed while updating merge/rollup task metadata for table: {}, skip scheduling. There are multiple task schedulers for the same table, need to investigate!", + offlineTableName); + continue; + } + pinotTaskConfigs.addAll(pinotTaskConfigsForTable); + LOGGER.info("Finished generating task configs for table: {} for task: {}", offlineTableName, taskType); + } + return pinotTaskConfigs; + } + + /** + * Validate table config for merge/rollup task + */ + private boolean validate(TableConfig tableConfig, String taskType) { + String offlineTableName = tableConfig.getTableName(); + if (tableConfig.getTableType() != TableType.OFFLINE) { + LOGGER.warn("Skip generating task: {} for non-OFFLINE table: {}, REALTIME table is not supported yet", taskType, + offlineTableName); + return false; + } + + if (REFRESH.equalsIgnoreCase(IngestionConfigUtils.getBatchSegmentIngestionType(tableConfig))) { + LOGGER.warn("Skip generating task: {} for non-APPEND table: {}, REFRESH table is not supported", taskType, + offlineTableName); + return false; + } + + TableTaskConfig tableTaskConfig = tableConfig.getTaskConfig(); + Preconditions.checkState(tableTaskConfig != null); + Map<String, String> taskConfigs = tableTaskConfig.getConfigsForTaskType(taskType); + Preconditions.checkState(taskConfigs != null, "Task config shouldn't be null for table: {}", offlineTableName); + return true; + } + + /** + * Check if the segment is merged for give merge level + */ + private boolean isMergedSegment(OfflineSegmentZKMetadata segmentZKMetadata, String mergeLevel) { + return segmentZKMetadata.getCustomMap() != null && mergeLevel.equalsIgnoreCase( + segmentZKMetadata.getCustomMap().get(MergeRollupTask.SEGMENT_ZK_METADATA_MERGE_LEVEL_KEY)); + } + + /** + * Check if the merge window end time is valid + */ + private boolean isValidMergeWindowEndTime(long windowEndMs, long bufferMs, String lowerMergeLevel, + MergeRollupTaskMetadata mergeRollupTaskMetadata) { + // Check that execution window endTimeMs <= now - bufferTime + if (windowEndMs > System.currentTimeMillis() - bufferMs) { + return false; + } + // Check that execution window endTimeMs <= waterMark of the lower mergeLevel + return lowerMergeLevel == null || windowEndMs <= mergeRollupTaskMetadata.getWatermarkMap().get(lowerMergeLevel); Review comment: `watermark` for the previous level? I think by current logic, cold start for the first merge level is the only case that the previous `watermark` is null. For this case, the `lowerMergeLevel` is `null`, so this will not cause NPE. -- 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
