[scheduling] implement backtracking of intermediate results

For batch programs, we currently schedule all tasks which are sources
and let them kick off the execution of the connected tasks. This
approach bears some problems when executing large dataflows with many
branches. With backtracking, we traverse the execution graph
output-centrically (from the sinks) in a depth-first manner. This
enables us to use resources differently.

In the course of backtracking, only tasks will be executed that are
required to supply inputs to the current task. When a job is newly
submitted, this means that the backtracking will reach the
sources. When the job has been previously executed and intermediate
results are available, old ResultPartitions to resume from can be
requested while backtracking.

Backtracking is disabled by default. It can be enabled by setting the
ScheduleMode in JobGraph to BACKTRACKING.

CHANGELOG
- new scheduling mode: backtracking
- backtracks from the sinks of an ExecutionGraph
- checks the availability of IntermediatePartitionResults
- marks ExecutionVertex to be scheduled
- caches ResultPartitions and reloads them
- resumes from intermediate results
- test for general behavior of backtracking (BacktrackingTest)
- test for resuming from an intermediate result (ResumeITCase)
- test for releasing of cached ResultPartitions (ResultPartitionManagerTest)
- allow multiple consumers per blocking intermediate result (batch)

flink-runtime/src/main/java/org/apache/flink/runtime/executiongraph/Backtracking.java

@@ -0,0 +1,254 @@
+/*
+ * 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:https://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.flink.runtime.executiongraph;
+
+import akka.actor.ActorRef;
+import akka.dispatch.OnComplete;
+import akka.pattern.Patterns;
+import com.google.common.base.Preconditions;
+import org.apache.flink.runtime.akka.AkkaUtils;
+import org.apache.flink.runtime.execution.ExecutionState;
+import org.apache.flink.runtime.jobgraph.IntermediateResultPartitionID;
+import org.apache.flink.runtime.messages.TaskMessages;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import scala.concurrent.Future;
+
+import java.util.ArrayDeque;
+import java.util.Collection;
+import java.util.Deque;
+import java.util.HashMap;
+import java.util.Iterator;
+import java.util.Map;
+
+
+/**
+ * Backtracking is a mechanism to schedule only those Execution Vertices of an Execution Graph which
+ * do not have an intermediate result available. This is in contrast to the simple way of scheduling
+ * a job, where all Execution Vertices are executed starting from the source. The Backtracking starts
+ * from the sinks and traverses the Execution Graph to the sources. It only reaches the sources if
+ * no intermediate result could be found on the way.
+ *
+ * @see ExecutionGraph
+ * @see ExecutionVertex
+ * @see Execution
+ */
+public class Backtracking {
+
+ private static final Logger LOG = LoggerFactory.getLogger(Backtracking.class);
+
+ private final Deque<TaskRequirement> taskRequirements = new ArrayDeque<TaskRequirement>();
+
+ private final Map<IntermediateResultPartitionID, Boolean> visitedPartitions = new HashMap<IntermediateResultPartitionID, Boolean>();
+
+ private ScheduleAction scheduleAction;
+ private Runnable postBacktrackingHook;
+
+ public Backtracking(Collection<ExecutionJobVertex> vertices) {
+ Preconditions.checkNotNull(vertices);
+
+ // add all sinks found to the stack
+ for (ExecutionJobVertex ejv : vertices) {
+ if (ejv.getJobVertex().isOutputVertex()) {
+ for (ExecutionVertex ev : ejv.getTaskVertices()) {
+ if (ev.getExecutionState() == ExecutionState.CREATED) {
+ taskRequirements.add(new TaskRequirement(ev));
+ }
+ }
+ }
+ }
+
+ this.scheduleAction = new ScheduleAction() {
+ @Override
+ public void schedule(ExecutionVertex ev) {}
+ };
+
+ this.postBacktrackingHook = new Runnable() {
+ @Override
+ public void run() {}
+ };
+ }
+
+ /**
+ * Scheduling to be performed when an ExecutionVertex is encountered that cannot be resumed
+ */
+ public interface ScheduleAction {
+ void schedule(ExecutionVertex ev);
+ }
+
+ /**
+ * A TaskRequirement encapsulates an ExecutionVertex and its IntermediateResultPartitions which
+ * are required for execution.
+ */
+ private class TaskRequirement {
+
+ private final ExecutionVertex executionVertex;
+ private final Deque<IntermediateResultPartition> pendingInputs = new ArrayDeque<IntermediateResultPartition>();
+ private final int numInputs;
+
+ private int availableInputs = 0;
+
+ public TaskRequirement(ExecutionVertex executionVertex) {
+ this.executionVertex = executionVertex;
+ this.pendingInputs.addAll(executionVertex.getInputs());
+ this.numInputs = pendingInputs.size();
+ }
+
+ public ExecutionVertex getExecutionVertex() {
+ return executionVertex;
+ }
+
+ public boolean pendingRequirements() {
+ Iterator<IntermediateResultPartition> iter = pendingInputs.iterator();
+ while (iter.hasNext()) {
+ Boolean visitedPartition = visitedPartitions.get(iter.next().getPartitionId());
+ if (visitedPartition == null) {
+ return true;
+ } else {
+ if (visitedPartition) {
+ availableInputs++;
+ }
+ iter.remove();
+ }
+ }
+ return false;
+ }
+
+ public IntermediateResultPartition getNextRequirement() {
+ return pendingInputs.pop();
+ }
+
+ public boolean needsToBeScheduled() {
+ return numInputs == availableInputs;
+ }
+
+ public void inputFound() {
+ availableInputs++;
+ }
+
+ @Override
+ public String toString() {
+ return "TaskRequirement{" +
+ "executionVertex=" + executionVertex +
+ ", pendingInputs=" + pendingInputs.size() +
+ '}';
+ }
+ }
+
+ /**
+ * Action to be performed on an ExecutionVertex when it is determined to be scheduled.
+ * @param scheduleAction A ScheduleAction which receives an ExecutionVertex.
+ */
+ public void setScheduleAction(ScheduleAction scheduleAction) {
+ Preconditions.checkNotNull(scheduleAction);
+ this.scheduleAction = scheduleAction;
+ }
+
+ /**
+ * Hook executed after backtracking finishes. Note that because of the use of futures, this may
+ * not be when the scheduleUsingBacktracking() method returns.
+ * @param postBacktrackingHook A Runnable that is executed after backtracking finishes.
+ */
+ public void setPostBacktrackingHook(Runnable postBacktrackingHook) {
+ Preconditions.checkNotNull(postBacktrackingHook);
+ this.postBacktrackingHook = postBacktrackingHook;
+ }
+
+ /* Visit the ExecutionGraph from the previously determined sinks using a pre-order depth-first
+ * iterative traversal.
+ */
+ public void scheduleUsingBacktracking() {
+
+ while (!taskRequirements.isEmpty()) {
+
+ final TaskRequirement taskRequirement = taskRequirements.peek();
+ final ExecutionVertex task = taskRequirement.getExecutionVertex();
+ task.getCurrentExecutionAttempt().setScheduled();
+
+ if (task.getExecutionState() != ExecutionState.CREATED && task.getExecutionState() != ExecutionState.DEPLOYING) {
+ LOG.debug("Resetting ExecutionVertex {} from {} to CREATED.", task, task.getExecutionState());
+ task.resetForNewExecution();
+ task.getCurrentExecutionAttempt().setScheduled();
+ }
+
+ if (taskRequirement.pendingRequirements()) {
+
+ final IntermediateResultPartition resultRequired = taskRequirement.getNextRequirement();
+
+ if (resultRequired.isLocationAvailable()) {
+ ActorRef taskManager = resultRequired.getLocation().getTaskManager();
+
+ LOG.debug("Requesting availability of IntermediateResultPartition " + resultRequired.getPartitionId());
+ // pin ResulPartition for this intermediate result
+ Future<Object> future = Patterns.ask(
+ taskManager,
+ new TaskMessages.LockResultPartition(
+ resultRequired.getPartitionId(),
+ // We lock this result for all consumers. We have to make sure
+ // to release it once a job finishes.
+ resultRequired.getNumConsumers()
+ ),
+ 5000 // 5 seconds timeout
+ );
+
+ /** BEGIN Asynchronous callback **/
+ future.onComplete(new OnComplete<Object>() {
+ @Override
+ public void onComplete(Throwable failure, Object success) {
+ if (success instanceof TaskMessages.LockResultPartitionReply &&
+ ((TaskMessages.LockResultPartitionReply) success).locked()) {
+ LOG.debug("Resuming from IntermediateResultPartition " + resultRequired.getPartitionId());
+ visitedPartitions.put(resultRequired.getPartitionId(), true);
+ taskRequirement.inputFound();
+ } else {
+ // intermediate result not available
+ visitedPartitions.put(resultRequired.getPartitionId(), false);
+ taskRequirements.push(new TaskRequirement(resultRequired.getProducer()));
+ }
+ // TODO try again in case of errors?
+ // continue with backtracking
+ scheduleUsingBacktracking();
+ }
+ }, AkkaUtils.globalExecutionContext());
+ /** END Asynchronous callback **/
+
+ // interrupt backtracking here and continue once future is complete
+ return;
+
+ } else {
+ visitedPartitions.put(resultRequired.getPartitionId(), false);
+ taskRequirements.push(new TaskRequirement(resultRequired.getProducer()));
+ }
+
+ } else {
+ taskRequirements.pop();
+
+ if (taskRequirement.needsToBeScheduled()) {
+ scheduleAction.schedule(task);
+ }
+
+ }
+
+ }
+
+ LOG.debug("Finished backtracking.");
+ postBacktrackingHook.run();
+ }
+
+}