聊聊flink的RichParallelSourceFunction

go4it發表於2018-11-28

序

本文主要研究一下flink的RichParallelSourceFunction

RichParallelSourceFunction

/**
 * Base class for implementing a parallel data source. Upon execution, the runtime will
 * execute as many parallel instances of this function function as configured parallelism
 * of the source.
 *
 * <p>The data source has access to context information (such as the number of parallel
 * instances of the source, and which parallel instance the current instance is)
 * via {@link #getRuntimeContext()}. It also provides additional life-cycle methods
 * ({@link #open(org.apache.flink.configuration.Configuration)} and {@link #close()}.</p>
 *
 * @param <OUT> The type of the records produced by this source.
 */
@Public
public abstract class RichParallelSourceFunction<OUT> extends AbstractRichFunction
		implements ParallelSourceFunction<OUT> {

	private static final long serialVersionUID = 1L;
}
複製程式碼

RichParallelSourceFunction實現了ParallelSourceFunction介面，同時繼承了AbstractRichFunction

ParallelSourceFunction

flink-streaming-java_2.11-1.6.2-sources.jar!/org/apache/flink/streaming/api/functions/source/ParallelSourceFunction.java

/**
 * A stream data source that is executed in parallel. Upon execution, the runtime will
 * execute as many parallel instances of this function function as configured parallelism
 * of the source.
 *
 * <p>This interface acts only as a marker to tell the system that this source may
 * be executed in parallel. When different parallel instances are required to perform
 * different tasks, use the {@link RichParallelSourceFunction} to get access to the runtime
 * context, which reveals information like the number of parallel tasks, and which parallel
 * task the current instance is.
 *
 * @param <OUT> The type of the records produced by this source.
 */
@Public
public interface ParallelSourceFunction<OUT> extends SourceFunction<OUT> {
}
複製程式碼

ParallelSourceFunction繼承了SourceFunction介面，它並沒有定義其他額外的方法，僅僅是用介面名來表達意圖，即可以被並行執行的stream data source

AbstractRichFunction

flink-core-1.6.2-sources.jar!/org/apache/flink/api/common/functions/AbstractRichFunction.java

/**
 * An abstract stub implementation for rich user-defined functions.
 * Rich functions have additional methods for initialization ({@link #open(Configuration)}) and
 * teardown ({@link #close()}), as well as access to their runtime execution context via
 * {@link #getRuntimeContext()}.
 */
@Public
public abstract class AbstractRichFunction implements RichFunction, Serializable {

	private static final long serialVersionUID = 1L;

	// --------------------------------------------------------------------------------------------
	//  Runtime context access
	// --------------------------------------------------------------------------------------------

	private transient RuntimeContext runtimeContext;

	@Override
	public void setRuntimeContext(RuntimeContext t) {
		this.runtimeContext = t;
	}

	@Override
	public RuntimeContext getRuntimeContext() {
		if (this.runtimeContext != null) {
			return this.runtimeContext;
		} else {
			throw new IllegalStateException("The runtime context has not been initialized.");
		}
	}

	@Override
	public IterationRuntimeContext getIterationRuntimeContext() {
		if (this.runtimeContext == null) {
			throw new IllegalStateException("The runtime context has not been initialized.");
		} else if (this.runtimeContext instanceof IterationRuntimeContext) {
			return (IterationRuntimeContext) this.runtimeContext;
		} else {
			throw new IllegalStateException("This stub is not part of an iteration step function.");
		}
	}

	// --------------------------------------------------------------------------------------------
	//  Default life cycle methods
	// --------------------------------------------------------------------------------------------

	@Override
	public void open(Configuration parameters) throws Exception {}

	@Override
	public void close() throws Exception {}
}
複製程式碼

AbstractRichFunction主要實現了RichFunction介面的setRuntimeContext、getRuntimeContext、getIterationRuntimeContext方法；open及close方法都是空操作

RuntimeContext

flink-core-1.6.2-sources.jar!/org/apache/flink/api/common/functions/RuntimeContext.java

/**
 * A RuntimeContext contains information about the context in which functions are executed. Each parallel instance
 * of the function will have a context through which it can access static contextual information (such as
 * the current parallelism) and other constructs like accumulators and broadcast variables.
 *
 * <p>A function can, during runtime, obtain the RuntimeContext via a call to
 * {@link AbstractRichFunction#getRuntimeContext()}.
 */
@Public
public interface RuntimeContext {

	/**
	 * Returns the name of the task in which the UDF runs, as assigned during plan construction.
	 *
	 * @return The name of the task in which the UDF runs.
	 */
	String getTaskName();

	/**
	 * Returns the metric group for this parallel subtask.
	 *
	 * @return The metric group for this parallel subtask.
	 */
	@PublicEvolving
	MetricGroup getMetricGroup();

	/**
	 * Gets the parallelism with which the parallel task runs.
	 *
	 * @return The parallelism with which the parallel task runs.
	 */
	int getNumberOfParallelSubtasks();

	/**
	 * Gets the number of max-parallelism with which the parallel task runs.
	 *
	 * @return The max-parallelism with which the parallel task runs.
	 */
	@PublicEvolving
	int getMaxNumberOfParallelSubtasks();

	/**
	 * Gets the number of this parallel subtask. The numbering starts from 0 and goes up to
	 * parallelism-1 (parallelism as returned by {@link #getNumberOfParallelSubtasks()}).
	 *
	 * @return The index of the parallel subtask.
	 */
	int getIndexOfThisSubtask();

	/**
	 * Gets the attempt number of this parallel subtask. First attempt is numbered 0.
	 *
	 * @return Attempt number of the subtask.
	 */
	int getAttemptNumber();

	/**
	 * Returns the name of the task, appended with the subtask indicator, such as "MyTask (3/6)",
	 * where 3 would be ({@link #getIndexOfThisSubtask()} + 1), and 6 would be
	 * {@link #getNumberOfParallelSubtasks()}.
	 *
	 * @return The name of the task, with subtask indicator.
	 */
	String getTaskNameWithSubtasks();

	/**
	 * Returns the {@link org.apache.flink.api.common.ExecutionConfig} for the currently executing
	 * job.
	 */
	ExecutionConfig getExecutionConfig();

	//.......
}
複製程式碼

RuntimeContext定義了很多方法，這裡我們看下getNumberOfParallelSubtasks方法，它可以返回當前的task的parallelism；而getIndexOfThisSubtask則可以獲取當前parallel subtask的下標；可以根據這些資訊，開發既能並行執行但各自發射的資料又不重複的ParallelSourceFunction

JobMaster.startJobExecution

flink-runtime_2.11-1.6.2-sources.jar!/org/apache/flink/runtime/jobmaster/JobMaster.java

	private Acknowledge startJobExecution(JobMasterId newJobMasterId) throws Exception {
		validateRunsInMainThread();

		checkNotNull(newJobMasterId, "The new JobMasterId must not be null.");

		if (Objects.equals(getFencingToken(), newJobMasterId)) {
			log.info("Already started the job execution with JobMasterId {}.", newJobMasterId);

			return Acknowledge.get();
		}

		setNewFencingToken(newJobMasterId);

		startJobMasterServices();

		log.info("Starting execution of job {} ({})", jobGraph.getName(), jobGraph.getJobID());

		resetAndScheduleExecutionGraph();

		return Acknowledge.get();
	}

	private void resetAndScheduleExecutionGraph() throws Exception {
		validateRunsInMainThread();

		final CompletableFuture<Void> executionGraphAssignedFuture;

		if (executionGraph.getState() == JobStatus.CREATED) {
			executionGraphAssignedFuture = CompletableFuture.completedFuture(null);
		} else {
			suspendAndClearExecutionGraphFields(new FlinkException("ExecutionGraph is being reset in order to be rescheduled."));
			final JobManagerJobMetricGroup newJobManagerJobMetricGroup = jobMetricGroupFactory.create(jobGraph);
			final ExecutionGraph newExecutionGraph = createAndRestoreExecutionGraph(newJobManagerJobMetricGroup);

			executionGraphAssignedFuture = executionGraph.getTerminationFuture().handleAsync(
				(JobStatus ignored, Throwable throwable) -> {
					assignExecutionGraph(newExecutionGraph, newJobManagerJobMetricGroup);
					return null;
				},
				getMainThreadExecutor());
		}

		executionGraphAssignedFuture.thenRun(this::scheduleExecutionGraph);
	}

	private void scheduleExecutionGraph() {
		checkState(jobStatusListener == null);
		// register self as job status change listener
		jobStatusListener = new JobManagerJobStatusListener();
		executionGraph.registerJobStatusListener(jobStatusListener);

		try {
			executionGraph.scheduleForExecution();
		}
		catch (Throwable t) {
			executionGraph.failGlobal(t);
		}
	}
複製程式碼

這裡呼叫了resetAndScheduleExecutionGraph方法，而resetAndScheduleExecutionGraph則組合了scheduleExecutionGraph方法；scheduleExecutionGraph這裡呼叫executionGraph.scheduleForExecution()來排程執行

ExecutionGraph.scheduleForExecution

flink-runtime_2.11-1.6.2-sources.jar!/org/apache/flink/runtime/executiongraph/ExecutionGraph.java

	public void scheduleForExecution() throws JobException {

		final long currentGlobalModVersion = globalModVersion;

		if (transitionState(JobStatus.CREATED, JobStatus.RUNNING)) {

			final CompletableFuture<Void> newSchedulingFuture;

			switch (scheduleMode) {

				case LAZY_FROM_SOURCES:
					newSchedulingFuture = scheduleLazy(slotProvider);
					break;

				case EAGER:
					newSchedulingFuture = scheduleEager(slotProvider, allocationTimeout);
					break;

				default:
					throw new JobException("Schedule mode is invalid.");
			}

			if (state == JobStatus.RUNNING && currentGlobalModVersion == globalModVersion) {
				schedulingFuture = newSchedulingFuture;

				newSchedulingFuture.whenCompleteAsync(
					(Void ignored, Throwable throwable) -> {
						if (throwable != null && !(throwable instanceof CancellationException)) {
							// only fail if the scheduling future was not canceled
							failGlobal(ExceptionUtils.stripCompletionException(throwable));
						}
					},
					futureExecutor);
			} else {
				newSchedulingFuture.cancel(false);
			}
		}
		else {
			throw new IllegalStateException("Job may only be scheduled from state " + JobStatus.CREATED);
		}
	}
複製程式碼

這裡走的是EAGER模式，因而呼叫scheduleEager方法

ExecutionGraph.scheduleEager

flink-runtime_2.11-1.6.2-sources.jar!/org/apache/flink/runtime/executiongraph/ExecutionGraph.java

	/**
	 *
	 *
	 * @param slotProvider  The resource provider from which the slots are allocated
	 * @param timeout       The maximum time that the deployment may take, before a
	 *                      TimeoutException is thrown.
	 * @returns Future which is completed once the {@link ExecutionGraph} has been scheduled.
	 * The future can also be completed exceptionally if an error happened.
	 */
	private CompletableFuture<Void> scheduleEager(SlotProvider slotProvider, final Time timeout) {
		checkState(state == JobStatus.RUNNING, "job is not running currently");

		// Important: reserve all the space we need up front.
		// that way we do not have any operation that can fail between allocating the slots
		// and adding them to the list. If we had a failure in between there, that would
		// cause the slots to get lost
		final boolean queued = allowQueuedScheduling;

		// collecting all the slots may resize and fail in that operation without slots getting lost
		final ArrayList<CompletableFuture<Execution>> allAllocationFutures = new ArrayList<>(getNumberOfExecutionJobVertices());

		// allocate the slots (obtain all their futures
		for (ExecutionJobVertex ejv : getVerticesTopologically()) {
			// these calls are not blocking, they only return futures
			Collection<CompletableFuture<Execution>> allocationFutures = ejv.allocateResourcesForAll(
				slotProvider,
				queued,
				LocationPreferenceConstraint.ALL,
				allocationTimeout);

			allAllocationFutures.addAll(allocationFutures);
		}

		// this future is complete once all slot futures are complete.
		// the future fails once one slot future fails.
		final ConjunctFuture<Collection<Execution>> allAllocationsFuture = FutureUtils.combineAll(allAllocationFutures);

		final CompletableFuture<Void> currentSchedulingFuture = allAllocationsFuture
			.thenAccept(
				(Collection<Execution> executionsToDeploy) -> {
					for (Execution execution : executionsToDeploy) {
						try {
							execution.deploy();
						} catch (Throwable t) {
							throw new CompletionException(
								new FlinkException(
									String.format("Could not deploy execution %s.", execution),
									t));
						}
					}
				})
			// Generate a more specific failure message for the eager scheduling
			.exceptionally(
				(Throwable throwable) -> {
					final Throwable strippedThrowable = ExceptionUtils.stripCompletionException(throwable);
					final Throwable resultThrowable;

					if (strippedThrowable instanceof TimeoutException) {
						int numTotal = allAllocationsFuture.getNumFuturesTotal();
						int numComplete = allAllocationsFuture.getNumFuturesCompleted();
						String message = "Could not allocate all requires slots within timeout of " +
							timeout + ". Slots required: " + numTotal + ", slots allocated: " + numComplete;

						resultThrowable = new NoResourceAvailableException(message);
					} else {
						resultThrowable = strippedThrowable;
					}

					throw new CompletionException(resultThrowable);
				});

		return currentSchedulingFuture;
	}
複製程式碼

scheduleEager方法這裡先呼叫getVerticesTopologically來獲取ExecutionJobVertex
之後呼叫ExecutionJobVertex.allocateResourcesForAll來分配資源得到Collection<CompletableFuture<Execution>>
最後通過FutureUtils.combineAll(allAllocationFutures)等待這批Future，之後挨個呼叫execution.deploy()進行部署

ExecutionJobVertex.allocateResourcesForAll

flink-runtime_2.11-1.6.2-sources.jar!/org/apache/flink/runtime/executiongraph/ExecutionJobVertex.java

	/**
	 * Acquires a slot for all the execution vertices of this ExecutionJobVertex. The method returns
	 * pairs of the slots and execution attempts, to ease correlation between vertices and execution
	 * attempts.
	 *
	 * <p>If this method throws an exception, it makes sure to release all so far requested slots.
	 *
	 * @param resourceProvider The resource provider from whom the slots are requested.
	 * @param queued if the allocation can be queued
	 * @param locationPreferenceConstraint constraint for the location preferences
	 * @param allocationTimeout timeout for allocating the individual slots
	 */
	public Collection<CompletableFuture<Execution>> allocateResourcesForAll(
			SlotProvider resourceProvider,
			boolean queued,
			LocationPreferenceConstraint locationPreferenceConstraint,
			Time allocationTimeout) {
		final ExecutionVertex[] vertices = this.taskVertices;
		final CompletableFuture<Execution>[] slots = new CompletableFuture[vertices.length];

		// try to acquire a slot future for each execution.
		// we store the execution with the future just to be on the safe side
		for (int i = 0; i < vertices.length; i++) {
			// allocate the next slot (future)
			final Execution exec = vertices[i].getCurrentExecutionAttempt();
			final CompletableFuture<Execution> allocationFuture = exec.allocateAndAssignSlotForExecution(
				resourceProvider,
				queued,
				locationPreferenceConstraint,
				allocationTimeout);
			slots[i] = allocationFuture;
		}

		// all good, we acquired all slots
		return Arrays.asList(slots);
	}
複製程式碼

這裡根據ExecutionJobVertex的taskVertices來挨個呼叫exec.allocateAndAssignSlotForExecution進行分配；可以發現整個並行度由taskVertices來決定

Execution.deploy

flink-runtime_2.11-1.6.2-sources.jar!/org/apache/flink/runtime/executiongraph/Execution.java

	/**
	 * Deploys the execution to the previously assigned resource.
	 *
	 * @throws JobException if the execution cannot be deployed to the assigned resource
	 */
	public void deploy() throws JobException {
		final LogicalSlot slot  = assignedResource;

		checkNotNull(slot, "In order to deploy the execution we first have to assign a resource via tryAssignResource.");

		//......

		try {

			//......

			final TaskDeploymentDescriptor deployment = vertex.createDeploymentDescriptor(
				attemptId,
				slot,
				taskRestore,
				attemptNumber);

			// null taskRestore to let it be GC'ed
			taskRestore = null;

			final TaskManagerGateway taskManagerGateway = slot.getTaskManagerGateway();

			final CompletableFuture<Acknowledge> submitResultFuture = taskManagerGateway.submitTask(deployment, rpcTimeout);

			submitResultFuture.whenCompleteAsync(
				(ack, failure) -> {
					// only respond to the failure case
					if (failure != null) {
						if (failure instanceof TimeoutException) {
							String taskname = vertex.getTaskNameWithSubtaskIndex() + " (" + attemptId + ')';

							markFailed(new Exception(
								"Cannot deploy task " + taskname + " - TaskManager (" + getAssignedResourceLocation()
									+ ") not responding after a rpcTimeout of " + rpcTimeout, failure));
						} else {
							markFailed(failure);
						}
					}
				},
				executor);
		}
		catch (Throwable t) {
			markFailed(t);
			ExceptionUtils.rethrow(t);
		}
	}
複製程式碼

Execution.deploy會建立TaskDeploymentDescriptor，之後通過taskManagerGateway.submitTask提交這個deployment；之後就是觸發TaskExecutor去觸發Task的run方法

ExecutionJobVertex