Guosmilesmile commented on code in PR #15151: URL: https://github.com/apache/iceberg/pull/15151#discussion_r2734903211
########## flink/v2.1/flink/src/main/java/org/apache/iceberg/flink/maintenance/operator/TriggerManagerOperator.java: ########## @@ -0,0 +1,327 @@ +/* + * 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.iceberg.flink.maintenance.operator; + +import java.util.List; +import org.apache.flink.annotation.Experimental; +import org.apache.flink.annotation.Internal; +import org.apache.flink.annotation.VisibleForTesting; +import org.apache.flink.api.common.operators.ProcessingTimeService.ProcessingTimeCallback; +import org.apache.flink.api.common.state.ListState; +import org.apache.flink.api.common.state.ListStateDescriptor; +import org.apache.flink.api.common.typeinfo.TypeInformation; +import org.apache.flink.api.common.typeinfo.Types; +import org.apache.flink.metrics.Counter; +import org.apache.flink.metrics.MetricGroup; +import org.apache.flink.runtime.operators.coordination.OperatorEvent; +import org.apache.flink.runtime.operators.coordination.OperatorEventGateway; +import org.apache.flink.runtime.operators.coordination.OperatorEventHandler; +import org.apache.flink.runtime.state.StateInitializationContext; +import org.apache.flink.runtime.state.StateSnapshotContext; +import org.apache.flink.streaming.api.operators.AbstractStreamOperator; +import org.apache.flink.streaming.api.operators.OneInputStreamOperator; +import org.apache.flink.streaming.api.operators.StreamOperatorParameters; +import org.apache.flink.streaming.api.watermark.Watermark; +import org.apache.flink.streaming.runtime.streamrecord.StreamRecord; +import org.apache.flink.streaming.runtime.tasks.ProcessingTimeService; +import org.apache.iceberg.flink.maintenance.api.Trigger; +import org.apache.iceberg.relocated.com.google.common.base.Preconditions; +import org.apache.iceberg.relocated.com.google.common.collect.Iterables; +import org.apache.iceberg.relocated.com.google.common.collect.Lists; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +/** + * TriggerManagerOperator sends events to the coordinator to acquire a lock, then waits for the + * response. If the response indicates that the lock has been acquired, it fires a trigger; + * otherwise, it schedules the next attempt. When the job recovers from a failure, tasks from + * different execution paths of the previous run may still be running. Therefore, it first needs to + * send a lock with the maximum timestamp, and then send a recovery trigger. Only after the + * downstream removes this lock can we be sure that all tasks have fully stopped. + */ +@Experimental +@Internal +public class TriggerManagerOperator extends AbstractStreamOperator<Trigger> + implements OneInputStreamOperator<TableChange, Trigger>, + OperatorEventHandler, + ProcessingTimeCallback { + private static final Logger LOG = LoggerFactory.getLogger(TriggerManagerOperator.class); + + private final List<String> maintenanceTaskNames; + private final List<TriggerEvaluator> evaluators; + private transient Long nextEvaluationTime; + private final long minFireDelayMs; + private final OperatorEventGateway operatorEventGateway; + private transient List<TableChange> accumulatedChanges; + private transient ListState<Long> nextEvaluationTimeState; + private transient ListState<TableChange> accumulatedChangesState; + private transient ListState<Long> lastTriggerTimesState; + private transient Counter rateLimiterTriggeredCounter; + private transient Counter concurrentRunThrottledCounter; + private transient Counter nothingToTriggerCounter; + private transient List<Counter> triggerCounters; + private final long lockCheckDelayMs; + private transient List<Long> lastTriggerTimes; + // To keep the task scheduling fair we keep the last triggered task position in memory. + // If we find a task to trigger, then we run it, but after it is finished, we start from the given + // position to prevent "starvation" of the tasks. + // When there is nothing to trigger, we start from the beginning, as the order of the tasks might + // be important (RewriteDataFiles first, and then RewriteManifestFiles later) + private transient int startsFrom = 0; + private transient boolean triggered = false; + private final String tableName; + private transient boolean restoreTasks = false; + private transient boolean lockHeld = false; + + public TriggerManagerOperator( + StreamOperatorParameters<Trigger> parameters, + OperatorEventGateway operatorEventGateway, + List<String> maintenanceTaskNames, + List<TriggerEvaluator> evaluators, + long minFireDelayMs, + long lockCheckDelayMs, + String tableName) { + super(parameters); + Preconditions.checkArgument( + maintenanceTaskNames != null && !maintenanceTaskNames.isEmpty(), + "Invalid maintenance task names: null or empty"); + Preconditions.checkArgument( + evaluators != null && !evaluators.isEmpty(), "Invalid evaluators: null or empty"); + Preconditions.checkArgument( + maintenanceTaskNames.size() == evaluators.size(), + "Provide a name and evaluator for all of the maintenance tasks"); + Preconditions.checkArgument(minFireDelayMs > 0, "Minimum fire delay should be at least 1."); + Preconditions.checkArgument( + lockCheckDelayMs > 0, "Minimum lock delay rate should be at least 1 ms."); + + this.maintenanceTaskNames = maintenanceTaskNames; + this.evaluators = evaluators; + this.minFireDelayMs = minFireDelayMs; + this.lockCheckDelayMs = lockCheckDelayMs; + this.tableName = tableName; + this.operatorEventGateway = operatorEventGateway; + } + + @Override + public void open() throws Exception { + super.open(); + MetricGroup mainGroup = TableMaintenanceMetrics.groupFor(getRuntimeContext(), tableName); + this.rateLimiterTriggeredCounter = + mainGroup.counter(TableMaintenanceMetrics.RATE_LIMITER_TRIGGERED); + this.concurrentRunThrottledCounter = + mainGroup.counter(TableMaintenanceMetrics.CONCURRENT_RUN_THROTTLED); + this.nothingToTriggerCounter = mainGroup.counter(TableMaintenanceMetrics.NOTHING_TO_TRIGGER); + this.triggerCounters = Lists.newArrayListWithExpectedSize(maintenanceTaskNames.size()); + for (int taskIndex = 0; taskIndex < maintenanceTaskNames.size(); ++taskIndex) { + triggerCounters.add( + TableMaintenanceMetrics.groupFor( + mainGroup, maintenanceTaskNames.get(taskIndex), taskIndex) + .counter(TableMaintenanceMetrics.TRIGGERED)); + } + } + + @Override + public void initializeState(StateInitializationContext context) throws Exception { + super.initializeState(context); + this.nextEvaluationTimeState = + context + .getOperatorStateStore() + .getListState(new ListStateDescriptor<>("triggerManagerNextTriggerTime", Types.LONG)); + + this.accumulatedChangesState = + context + .getOperatorStateStore() + .getListState( + new ListStateDescriptor<>( + "triggerManagerAccumulatedChange", TypeInformation.of(TableChange.class))); + + this.lastTriggerTimesState = + context + .getOperatorStateStore() + .getListState(new ListStateDescriptor<>("triggerManagerLastTriggerTime", Types.LONG)); + + long current = getProcessingTimeService().getCurrentProcessingTime(); + + // Initialize from state + if (!Iterables.isEmpty(nextEvaluationTimeState.get())) { + nextEvaluationTime = Iterables.getOnlyElement(nextEvaluationTimeState.get()); + } + + this.accumulatedChanges = Lists.newArrayList(accumulatedChangesState.get()); + this.lastTriggerTimes = Lists.newArrayList(lastTriggerTimesState.get()); + + // Initialize if the state was empty + if (accumulatedChanges.isEmpty()) { + for (int i = 0; i < evaluators.size(); ++i) { + accumulatedChanges.add(TableChange.empty()); + lastTriggerTimes.add(current); + } + } + + // register the lock register event + operatorEventGateway.sendEventToCoordinator(new LockRegisterEvent(tableName, current)); + + if (context.isRestored()) { + // When the job state is restored, there could be ongoing tasks. + // To prevent collision with the new triggers the following is done: + // - add a recovery lock + // - fire a recovery trigger + // This ensures that the tasks of the previous trigger are executed, and the lock is removed + // in the end. The result of the 'tryLock' is ignored as an already existing lock prevents + // collisions as well. + // register the recover lock + restoreTasks = true; + lockHeld = false; + output.collect(new StreamRecord<>(Trigger.recovery(current), current)); + if (nextEvaluationTime == null) { + schedule(getProcessingTimeService(), current + minFireDelayMs); + } else { + schedule(getProcessingTimeService(), nextEvaluationTime); + } + } else { + restoreTasks = false; + lockHeld = false; + } + } + + @Override + public void snapshotState(StateSnapshotContext context) throws Exception { + nextEvaluationTimeState.clear(); + if (nextEvaluationTime != null) { + nextEvaluationTimeState.add(nextEvaluationTime); + } + + accumulatedChangesState.update(accumulatedChanges); + lastTriggerTimesState.update(lastTriggerTimes); + LOG.info( + "Storing state: nextEvaluationTime {}, accumulatedChanges {}, lastTriggerTimes {}", + nextEvaluationTime, + accumulatedChanges, + lastTriggerTimes); + } + + @Override + public void handleOperatorEvent(OperatorEvent event) { + if (event instanceof LockReleasedEvent) { + LOG.info("Received lock released event: {}", event); + handleLockReleaseResult((LockReleasedEvent) event); + } else { + throw new IllegalArgumentException( + "Invalid operator event type: " + event.getClass().getCanonicalName()); + } + } + + @SuppressWarnings("FutureReturnValueIgnored") + @Override + public void processElement(StreamRecord<TableChange> streamRecord) throws Exception { + TableChange change = streamRecord.getValue(); + accumulatedChanges.forEach(tableChange -> tableChange.merge(change)); + if (nextEvaluationTime == null) { + checkAndFire(getProcessingTimeService()); + } else { + LOG.info( + "Trigger manager rate limiter triggered current: {}, next: {}, accumulated changes: {},{}", + getProcessingTimeService().getCurrentProcessingTime(), + nextEvaluationTime, + accumulatedChanges, + maintenanceTaskNames); + rateLimiterTriggeredCounter.inc(); + } + } + + @Override + public void processWatermark(Watermark mark) throws Exception { + super.processWatermark(mark); + } + + @Override + public void onProcessingTime(long l) { + this.nextEvaluationTime = null; + checkAndFire(getProcessingTimeService()); + } + + @Override + public void close() throws Exception { + super.close(); + this.lockHeld = false; + this.restoreTasks = false; + } + + @VisibleForTesting + void handleLockReleaseResult(LockReleasedEvent event) { + if (event.lockId().equals(tableName)) { + this.lockHeld = false; + this.restoreTasks = false; Review Comment: I think we can differentiate between the restoreLock release and the lockHeld release from the source from processWatermark or processElement, from processWatermark is recove. So I add a flag in LockReleasedEvent. And TriggerManagerOperator only release restoreTasksLock when it is from watermark -- 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]
