Search in sources :

Example 6 with IntermediateResultPartition

use of org.apache.flink.runtime.executiongraph.IntermediateResultPartition in project flink by apache.

the class JobManagerTest method testRequestPartitionStateMoreRecentExecutionAttempt.

/**
	 * Tests the JobManager response when the execution is not registered with
	 * the ExecutionGraph anymore and a new execution attempt is available.
	 */
@Test
public void testRequestPartitionStateMoreRecentExecutionAttempt() throws Exception {
    new JavaTestKit(system) {

        {
            new Within(duration("15 seconds")) {

                @Override
                protected void run() {
                    // Setup
                    TestingCluster cluster = null;
                    try {
                        cluster = startTestingCluster(4, 1, DEFAULT_AKKA_ASK_TIMEOUT());
                        final IntermediateDataSetID rid = new IntermediateDataSetID();
                        // Create a task
                        final JobVertex sender = new JobVertex("Sender");
                        sender.setParallelism(1);
                        // just finish
                        sender.setInvokableClass(NoOpInvokable.class);
                        sender.createAndAddResultDataSet(rid, PIPELINED);
                        final JobVertex sender2 = new JobVertex("Blocking Sender");
                        sender2.setParallelism(1);
                        // just block
                        sender2.setInvokableClass(BlockingNoOpInvokable.class);
                        sender2.createAndAddResultDataSet(new IntermediateDataSetID(), PIPELINED);
                        final JobGraph jobGraph = new JobGraph("Fast finishing producer test job", sender, sender2);
                        final JobID jid = jobGraph.getJobID();
                        final ActorGateway jobManagerGateway = cluster.getLeaderGateway(TestingUtils.TESTING_DURATION());
                        // we can set the leader session ID to None because we don't use this gateway to send messages
                        final ActorGateway testActorGateway = new AkkaActorGateway(getTestActor(), null);
                        // Submit the job and wait for all vertices to be running
                        jobManagerGateway.tell(new SubmitJob(jobGraph, ListeningBehaviour.EXECUTION_RESULT), testActorGateway);
                        expectMsgClass(JobManagerMessages.JobSubmitSuccess.class);
                        jobManagerGateway.tell(new WaitForAllVerticesToBeRunningOrFinished(jid), testActorGateway);
                        expectMsgClass(TestingJobManagerMessages.AllVerticesRunning.class);
                        Future<Object> egFuture = jobManagerGateway.ask(new RequestExecutionGraph(jobGraph.getJobID()), remaining());
                        ExecutionGraphFound egFound = (ExecutionGraphFound) Await.result(egFuture, remaining());
                        ExecutionGraph eg = (ExecutionGraph) egFound.executionGraph();
                        ExecutionVertex vertex = eg.getJobVertex(sender.getID()).getTaskVertices()[0];
                        while (vertex.getExecutionState() != ExecutionState.FINISHED) {
                            Thread.sleep(1);
                        }
                        IntermediateResultPartition partition = vertex.getProducedPartitions().values().iterator().next();
                        ResultPartitionID partitionId = new ResultPartitionID(partition.getPartitionId(), vertex.getCurrentExecutionAttempt().getAttemptId());
                        // Reset execution => new execution attempt
                        vertex.resetForNewExecution();
                        // Producer finished, request state
                        Object request = new JobManagerMessages.RequestPartitionProducerState(jid, rid, partitionId);
                        Future<?> producerStateFuture = jobManagerGateway.ask(request, getRemainingTime());
                        try {
                            Await.result(producerStateFuture, getRemainingTime());
                            fail("Did not fail with expected Exception");
                        } catch (PartitionProducerDisposedException ignored) {
                        }
                    } catch (Exception e) {
                        e.printStackTrace();
                        fail(e.getMessage());
                    } finally {
                        if (cluster != null) {
                            cluster.shutdown();
                        }
                    }
                }
            };
        }
    };
}
Also used : AkkaActorGateway(org.apache.flink.runtime.instance.AkkaActorGateway) RequestPartitionProducerState(org.apache.flink.runtime.messages.JobManagerMessages.RequestPartitionProducerState) ExecutionVertex(org.apache.flink.runtime.executiongraph.ExecutionVertex) TestingUtils.startTestingCluster(org.apache.flink.runtime.testingUtils.TestingUtils.startTestingCluster) TestingCluster(org.apache.flink.runtime.testingUtils.TestingCluster) TestingJobManagerMessages(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages) ActorGateway(org.apache.flink.runtime.instance.ActorGateway) AkkaActorGateway(org.apache.flink.runtime.instance.AkkaActorGateway) RequestExecutionGraph(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.RequestExecutionGraph) ExecutionGraphFound(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.ExecutionGraphFound) ResultPartitionID(org.apache.flink.runtime.io.network.partition.ResultPartitionID) SubmitJob(org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob) JobManagerMessages(org.apache.flink.runtime.messages.JobManagerMessages) TestingJobManagerMessages(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages) JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) WaitForAllVerticesToBeRunningOrFinished(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.WaitForAllVerticesToBeRunningOrFinished) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) IntermediateResultPartition(org.apache.flink.runtime.executiongraph.IntermediateResultPartition) ExecutionGraph(org.apache.flink.runtime.executiongraph.ExecutionGraph) RequestExecutionGraph(org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.RequestExecutionGraph) IntermediateDataSetID(org.apache.flink.runtime.jobgraph.IntermediateDataSetID) JavaTestKit(akka.testkit.JavaTestKit) JobID(org.apache.flink.api.common.JobID) Test(org.junit.Test)

Example 7 with IntermediateResultPartition

use of org.apache.flink.runtime.executiongraph.IntermediateResultPartition in project flink by apache.

the class InputChannelDeploymentDescriptor method fromEdges.

// ------------------------------------------------------------------------
/**
	 * Creates an input channel deployment descriptor for each partition.
	 */
public static InputChannelDeploymentDescriptor[] fromEdges(ExecutionEdge[] edges, SimpleSlot consumerSlot, boolean allowLazyDeployment) throws ExecutionGraphException {
    final ResourceID consumerTaskManager = consumerSlot.getTaskManagerID();
    final InputChannelDeploymentDescriptor[] icdd = new InputChannelDeploymentDescriptor[edges.length];
    // Each edge is connected to a different result partition
    for (int i = 0; i < edges.length; i++) {
        final IntermediateResultPartition consumedPartition = edges[i].getSource();
        final Execution producer = consumedPartition.getProducer().getCurrentExecutionAttempt();
        final ExecutionState producerState = producer.getState();
        final SimpleSlot producerSlot = producer.getAssignedResource();
        final ResultPartitionLocation partitionLocation;
        // The producing task needs to be RUNNING or already FINISHED
        if (consumedPartition.isConsumable() && producerSlot != null && (producerState == ExecutionState.RUNNING || producerState == ExecutionState.FINISHED || producerState == ExecutionState.SCHEDULED || producerState == ExecutionState.DEPLOYING)) {
            final TaskManagerLocation partitionTaskManagerLocation = producerSlot.getTaskManagerLocation();
            final ResourceID partitionTaskManager = partitionTaskManagerLocation.getResourceID();
            if (partitionTaskManager.equals(consumerTaskManager)) {
                // Consuming task is deployed to the same TaskManager as the partition => local
                partitionLocation = ResultPartitionLocation.createLocal();
            } else {
                // Different instances => remote
                final ConnectionID connectionId = new ConnectionID(partitionTaskManagerLocation, consumedPartition.getIntermediateResult().getConnectionIndex());
                partitionLocation = ResultPartitionLocation.createRemote(connectionId);
            }
        } else if (allowLazyDeployment) {
            // The producing task might not have registered the partition yet
            partitionLocation = ResultPartitionLocation.createUnknown();
        } else if (producerState == ExecutionState.CANCELING || producerState == ExecutionState.CANCELED || producerState == ExecutionState.FAILED) {
            String msg = "Trying to schedule a task whose inputs were canceled or failed. " + "The producer is in state " + producerState + ".";
            throw new ExecutionGraphException(msg);
        } else {
            String msg = String.format("Trying to eagerly schedule a task whose inputs " + "are not ready (partition consumable? %s, producer state: %s, producer slot: %s).", consumedPartition.isConsumable(), producerState, producerSlot);
            throw new ExecutionGraphException(msg);
        }
        final ResultPartitionID consumedPartitionId = new ResultPartitionID(consumedPartition.getPartitionId(), producer.getAttemptId());
        icdd[i] = new InputChannelDeploymentDescriptor(consumedPartitionId, partitionLocation);
    }
    return icdd;
}
Also used : ExecutionState(org.apache.flink.runtime.execution.ExecutionState) ExecutionGraphException(org.apache.flink.runtime.executiongraph.ExecutionGraphException) TaskManagerLocation(org.apache.flink.runtime.taskmanager.TaskManagerLocation) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) ConnectionID(org.apache.flink.runtime.io.network.ConnectionID) IntermediateResultPartition(org.apache.flink.runtime.executiongraph.IntermediateResultPartition) Execution(org.apache.flink.runtime.executiongraph.Execution) ResourceID(org.apache.flink.runtime.clusterframework.types.ResourceID) ResultPartitionID(org.apache.flink.runtime.io.network.partition.ResultPartitionID)

Aggregations

IntermediateResultPartition (org.apache.flink.runtime.executiongraph.IntermediateResultPartition)7 ResultPartitionID (org.apache.flink.runtime.io.network.partition.ResultPartitionID)6 ExecutionVertex (org.apache.flink.runtime.executiongraph.ExecutionVertex)5 Test (org.junit.Test)5 ExecutionState (org.apache.flink.runtime.execution.ExecutionState)4 JavaTestKit (akka.testkit.JavaTestKit)3 JobID (org.apache.flink.api.common.JobID)3 ResourceID (org.apache.flink.runtime.clusterframework.types.ResourceID)3 ExecutionGraph (org.apache.flink.runtime.executiongraph.ExecutionGraph)3 ActorGateway (org.apache.flink.runtime.instance.ActorGateway)3 AkkaActorGateway (org.apache.flink.runtime.instance.AkkaActorGateway)3 SimpleSlot (org.apache.flink.runtime.instance.SimpleSlot)3 IntermediateDataSetID (org.apache.flink.runtime.jobgraph.IntermediateDataSetID)3 IntermediateResultPartitionID (org.apache.flink.runtime.jobgraph.IntermediateResultPartitionID)3 JobGraph (org.apache.flink.runtime.jobgraph.JobGraph)3 JobVertex (org.apache.flink.runtime.jobgraph.JobVertex)3 RequestPartitionProducerState (org.apache.flink.runtime.messages.JobManagerMessages.RequestPartitionProducerState)3 SubmitJob (org.apache.flink.runtime.messages.JobManagerMessages.SubmitJob)3 TestingCluster (org.apache.flink.runtime.testingUtils.TestingCluster)3 RequestExecutionGraph (org.apache.flink.runtime.testingUtils.TestingJobManagerMessages.RequestExecutionGraph)3