jugomezv commented on code in PR #9994: URL: https://github.com/apache/pinot/pull/9994#discussion_r1060963331
########## pinot-core/src/main/java/org/apache/pinot/core/data/manager/realtime/ConsumptionDelayTracker.java: ########## @@ -0,0 +1,361 @@ +/** + * 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.core.data.manager.realtime; + +import com.google.common.annotations.VisibleForTesting; +import java.util.HashSet; +import java.util.List; +import java.util.Timer; +import java.util.TimerTask; +import java.util.concurrent.ConcurrentHashMap; +import org.apache.pinot.common.metrics.ServerGauge; +import org.apache.pinot.common.metrics.ServerMetrics; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + + +/** + * A Class to track maximum realtime delay for a given table on a given server. + * Highlights: + * 1-An object of this class is hosted by each RealtimeTableDataManager. + * 2-The object tracks ingestion delays for all partitions hosted by the current server for the given Realtime table. + * 3-Partition delays are updated by all LLRealtimeSegmentDataManager objects hosted in the corresponding + * RealtimeTableDataManager. + * 4-The class tracks the maximum of all ingestion delays observed for all partitions of the given table. + * 5-A Metric is derived from reading the maximum tracked by this class. + * 6-Delay reported for partitions that do not have events to consume is reported as zero. + * 7-We track the time at which each delay sample was collected so that delay can be increased when partition stops + * consuming for any reason other than no events being available for consumption. + * 8-Segments that go from CONSUMING to DROPPED states stop being tracked so their delays do not cloud + * delays of active partitions. + * 9-When a segment goes from CONSUMING to ONLINE, we start ticking time for the corresponding partition. + * If no consumption is noticed after a timeout, then we read ideal state to confirm the server still hosts the + * partition. If not, we stop tracking the respective partition. + * 10-A timer thread is started by this object to track timeouts of partitions and drive the reading of their ideal + * state. + * + * The following diagram illustrates the object interactions with main external APIs + * + * (CONSUMING -> ONLINE state change) + * | + * markPartitionForConfirmation(partitionId) + * | |<-storeConsumptionDelay()-{LLRealtimeSegmentDataManager(Partition 0}} + * | | + * ___________V_________________________V_ + * | (Table X) |<-storeConsumptionDelay()-{LLRealtimeSegmentDataManager(Partition 1}} + * | ConsumptionDelayTracker | ... + * |____________________________________|<-storeConsumptionDelay()-{LLRealtimeSegmentDataManager (Partition n}} + * ^ ^ + * | \ + * timeoutInactivePartitions() stopTrackingPartitionConsumptionDelay(partitionId) + * _________|__________ \ + * | TimerTrackingTask | (CONSUMING -> DROPPED state change) + * |___________________| + * + */ + +public class ConsumptionDelayTracker { + + // Sleep interval for timer thread that triggers read of ideal state + private static final int TIMER_THREAD_TICK_INTERVAL_MS = 300000; // 5 minutes +/- precision in timeouts + // Once a partition is marked for verification, we wait 10 minutes to pull its ideal state. + private static final int PARTITION_TIMEOUT_MS = 600000; // 10 minutes timeouts + // Delay Timer thread for this time after starting timer + private static final int INITIAL_TIMER_THREAD_DELAY_MS = 100; + + /* + * Class to keep a Pinot Consumption Delay measure and the time when the sample was taken (i.e. sample time) + * We will use the sample time to increase consumption delay when a partition stops consuming: the time + * difference between the sample time and current time will be added to the metric when read. + */ + static private class DelayMeasure { + public DelayMeasure(long t, long d) { + _delayMilliseconds = d; + _sampleTime = t; + } + public long _delayMilliseconds; + public long _sampleTime; + } + + /* + * Class to handle timer thread that will track inactive partitions + */ + private class TrackingTimerTask extends TimerTask { + final private ConsumptionDelayTracker _tracker; + + public TrackingTimerTask(ConsumptionDelayTracker tracker) { + _tracker = tracker; + } + + @Override + public void run() { + // tick inactive partitions every interval of time to keep tracked partitions up to date + _tracker.timeoutInactivePartitions(); + } + } + + // HashMap used to store delay measures for all partitions active for the current table. + // _partitionsToDelaySampleMap<Key=PartitionGroupId,Value=DelaySample> + private ConcurrentHashMap<Integer, DelayMeasure> _partitionToDelaySampleMap = new ConcurrentHashMap<>(); + // We mark partitions that go from CONSUMING to ONLINE in _partitionsMarkedForVerification: if they do not + // go back to CONSUMING in some period of time, we confirm whether they are still hosted in this server by reading + // ideal state. This is done with the goal of minimizing reading ideal state for efficiency reasons. + // _partitionsMarkedForVerification<Key=PartitionGroupId,Value=TimePartitionWasMarkedForVerificationMilliseconds> + private ConcurrentHashMap<Integer, Long> _partitionsMarkedForVerification = new ConcurrentHashMap<>(); + // Mutable versions of timer constants so we can test with smaller delays + final int _timerThreadTickIntervalMs; + // Timer task to check partitions that are inactive against ideal state. + final private Timer _timer; + + final private ServerMetrics _serverMetrics; + final private String _tableNameWithType; + + private boolean _enableAging; + private boolean _enablePerPartitionMetric = true; + private boolean _enableAggregateMetric = true; + final private Logger _logger; + + final private RealtimeTableDataManager _realTimeTableDataManager; + + /* + * Helper function to update the maximum when the current maximum is removed or updated. + * If no samples left we set maximum to minimum so new samples can be recorded. + */ + private DelayMeasure getMaximumDelay() { + DelayMeasure newMax = null; + for (int partitionGroupId : _partitionToDelaySampleMap.keySet()) { + DelayMeasure currentMeasure = _partitionToDelaySampleMap.get(partitionGroupId); + if ((newMax == null) + || + (currentMeasure != null) && (currentMeasure._delayMilliseconds > newMax._delayMilliseconds)) { + newMax = currentMeasure; + } + } + return newMax; + } + + /* + * Helper function to age a delay measure. Aging means adding the time elapsed since the measure was + * taken till the measure is being reported. + */ + private long getAgedDelay(DelayMeasure currentDelay) { + if (currentDelay == null) { + return 0; // return 0 when not initialized + } + // Add age of measure to the reported value + long measureAgeInMs = _enableAging ? (System.currentTimeMillis() - currentDelay._sampleTime) : 0; + // Correct to zero for any time shifts due to NTP or time reset. + measureAgeInMs = Math.max(measureAgeInMs, 0); + return currentDelay._delayMilliseconds + measureAgeInMs; + } + + private List<Integer> getPartitionsHostedByThisServerPerIdealState() { + return _realTimeTableDataManager.getHostedPartitionsGroupIds(); + } + /* + * Helper function to be called when we should stop tracking a given partition. Removes the partition from + * all our maps, it also updates the maximum if the tracked partition was the previous maximum. + * + * @param partitionGroupId partition ID which we should stop tracking. + */ + private void removePartitionId(int partitionGroupId) { + _partitionToDelaySampleMap.remove(partitionGroupId); + // If we are removing a partition we should stop reading its ideal state. + _partitionsMarkedForVerification.remove(partitionGroupId); + if (_enablePerPartitionMetric) { + _serverMetrics.removeTableGauge(getPerPartitionMetricName(partitionGroupId), + ServerGauge.PER_PARTITION_CONSUMPTION_DELAY_MS); + } + } + + /* + * Helper function to generate a per partition metric name. + * + * @param partitionGroupId the partition group id to be appended to the table name so we + * can differentiate between metrics for various partitions. + * + * @return a metric name with the following structure: tableNameWithType + partitionGroupId + */ + private String getPerPartitionMetricName(int partitionGroupId) { + return _tableNameWithType + partitionGroupId; + } + + // Custom Constructor + public ConsumptionDelayTracker(ServerMetrics serverMetrics, String tableNameWithType, + RealtimeTableDataManager realtimeTableDataManager, int timerThreadTickIntervalMs, String metricNamePrefix, + boolean enableAggregateMetric, boolean enablePerPartitionMetric) + throws RuntimeException { + _logger = LoggerFactory.getLogger(tableNameWithType + "-" + getClass().getSimpleName()); + _serverMetrics = serverMetrics; + _tableNameWithType = metricNamePrefix + tableNameWithType; + _realTimeTableDataManager = realtimeTableDataManager; + // Handle negative timer values + if (timerThreadTickIntervalMs <= 0) { + throw new RuntimeException("Illegal timer timeout argument, expected > 0, got=" + timerThreadTickIntervalMs); + } + _enableAging = true; + _enablePerPartitionMetric = enablePerPartitionMetric; + _enableAggregateMetric = enableAggregateMetric; + _timerThreadTickIntervalMs = timerThreadTickIntervalMs; + _timer = new Timer("ConsumptionDelayTimerThread" + tableNameWithType); + _timer.schedule(new TrackingTimerTask(this), INITIAL_TIMER_THREAD_DELAY_MS, _timerThreadTickIntervalMs); Review Comment: Done -- 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
