use of org.apache.flink.runtime.jobgraph.IntermediateDataSetID in project flink by apache.
the class JobManagerTest method testRequestPartitionState.
/**
* Tests responses to partition state requests.
*/
@Test
public void testRequestPartitionState() throws Exception {
new JavaTestKit(system) {
{
new Within(duration("15 seconds")) {
@Override
protected void run() {
// Setup
TestingCluster cluster = null;
try {
cluster = startTestingCluster(2, 1, DEFAULT_AKKA_ASK_TIMEOUT());
final IntermediateDataSetID rid = new IntermediateDataSetID();
// Create a task
final JobVertex sender = new JobVertex("Sender");
sender.setParallelism(1);
// just block
sender.setInvokableClass(BlockingNoOpInvokable.class);
sender.createAndAddResultDataSet(rid, PIPELINED);
final JobGraph jobGraph = new JobGraph("Blocking test job", sender);
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(JobSubmitSuccess.class);
jobManagerGateway.tell(new WaitForAllVerticesToBeRunningOrFinished(jid), testActorGateway);
expectMsgClass(AllVerticesRunning.class);
// This is the mock execution ID of the task requesting the state of the partition
final ExecutionAttemptID receiver = new ExecutionAttemptID();
// Request the execution graph to get the runtime info
jobManagerGateway.tell(new RequestExecutionGraph(jid), testActorGateway);
final ExecutionGraph eg = (ExecutionGraph) expectMsgClass(ExecutionGraphFound.class).executionGraph();
final ExecutionVertex vertex = eg.getJobVertex(sender.getID()).getTaskVertices()[0];
final IntermediateResultPartition partition = vertex.getProducedPartitions().values().iterator().next();
final ResultPartitionID partitionId = new ResultPartitionID(partition.getPartitionId(), vertex.getCurrentExecutionAttempt().getAttemptId());
// - The test ----------------------------------------------------------------------
// 1. All execution states
RequestPartitionProducerState request = new RequestPartitionProducerState(jid, rid, partitionId);
for (ExecutionState state : ExecutionState.values()) {
ExecutionGraphTestUtils.setVertexState(vertex, state);
Future<ExecutionState> futurePartitionState = jobManagerGateway.ask(request, getRemainingTime()).mapTo(ClassTag$.MODULE$.<ExecutionState>apply(ExecutionState.class));
ExecutionState resp = Await.result(futurePartitionState, getRemainingTime());
assertEquals(state, resp);
}
// 2. Non-existing execution
request = new RequestPartitionProducerState(jid, rid, new ResultPartitionID());
Future<?> futurePartitionState = jobManagerGateway.ask(request, getRemainingTime());
try {
Await.result(futurePartitionState, getRemainingTime());
fail("Did not fail with expected RuntimeException");
} catch (RuntimeException e) {
assertEquals(IllegalArgumentException.class, e.getCause().getClass());
}
// 3. Non-existing job
request = new RequestPartitionProducerState(new JobID(), rid, new ResultPartitionID());
futurePartitionState = jobManagerGateway.ask(request, getRemainingTime());
try {
Await.result(futurePartitionState, getRemainingTime());
fail("Did not fail with expected IllegalArgumentException");
} catch (IllegalArgumentException ignored) {
}
} catch (Exception e) {
e.printStackTrace();
fail(e.getMessage());
} finally {
if (cluster != null) {
cluster.shutdown();
}
}
}
};
}
};
}
use of org.apache.flink.runtime.jobgraph.IntermediateDataSetID 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();
}
}
}
};
}
};
}
use of org.apache.flink.runtime.jobgraph.IntermediateDataSetID in project flink by apache.
the class SingleInputGateTest method testUpdateChannelBeforeRequest.
/**
* Tests that an update channel does not trigger a partition request before the UDF has
* requested any partitions. Otherwise, this can lead to races when registering a listener at
* the gate (e.g. in UnionInputGate), which can result in missed buffer notifications at the
* listener.
*/
@Test
public void testUpdateChannelBeforeRequest() throws Exception {
SingleInputGate inputGate = new SingleInputGate("t1", new JobID(), new IntermediateDataSetID(), ResultPartitionType.PIPELINED, 0, 1, mock(TaskActions.class), new UnregisteredTaskMetricsGroup.DummyTaskIOMetricGroup());
ResultPartitionManager partitionManager = mock(ResultPartitionManager.class);
InputChannel unknown = new UnknownInputChannel(inputGate, 0, new ResultPartitionID(), partitionManager, new TaskEventDispatcher(), new LocalConnectionManager(), 0, 0, new UnregisteredTaskMetricsGroup.DummyTaskIOMetricGroup());
inputGate.setInputChannel(unknown.partitionId.getPartitionId(), unknown);
// Update to a local channel and verify that no request is triggered
inputGate.updateInputChannel(new InputChannelDeploymentDescriptor(unknown.partitionId, ResultPartitionLocation.createLocal()));
verify(partitionManager, never()).createSubpartitionView(any(ResultPartitionID.class), anyInt(), any(BufferProvider.class), any(BufferAvailabilityListener.class));
}
use of org.apache.flink.runtime.jobgraph.IntermediateDataSetID in project flink by apache.
the class UnionInputGateTest method testBasicGetNextLogic.
/**
* Tests basic correctness of buffer-or-event interleaving and correct <code>null</code> return
* value after receiving all end-of-partition events.
*
* <p> For buffer-or-event instances, it is important to verify that they have been set off to
* the correct logical index.
*/
@Test(timeout = 120 * 1000)
public void testBasicGetNextLogic() throws Exception {
// Setup
final String testTaskName = "Test Task";
final SingleInputGate ig1 = new SingleInputGate(testTaskName, new JobID(), new IntermediateDataSetID(), ResultPartitionType.PIPELINED, 0, 3, mock(TaskActions.class), new UnregisteredTaskMetricsGroup.DummyTaskIOMetricGroup());
final SingleInputGate ig2 = new SingleInputGate(testTaskName, new JobID(), new IntermediateDataSetID(), ResultPartitionType.PIPELINED, 0, 5, mock(TaskActions.class), new UnregisteredTaskMetricsGroup.DummyTaskIOMetricGroup());
final UnionInputGate union = new UnionInputGate(new SingleInputGate[] { ig1, ig2 });
assertEquals(ig1.getNumberOfInputChannels() + ig2.getNumberOfInputChannels(), union.getNumberOfInputChannels());
final TestInputChannel[][] inputChannels = new TestInputChannel[][] { TestInputChannel.createInputChannels(ig1, 3), TestInputChannel.createInputChannels(ig2, 5) };
// 0 => 0
inputChannels[0][0].readBuffer();
// 0 => 0
inputChannels[0][0].readEndOfPartitionEvent();
// 2 => 5
inputChannels[1][2].readBuffer();
// 2 => 5
inputChannels[1][2].readEndOfPartitionEvent();
// 0 => 3
inputChannels[1][0].readBuffer();
// 1 => 4
inputChannels[1][1].readBuffer();
// 1 => 1
inputChannels[0][1].readBuffer();
// 3 => 6
inputChannels[1][3].readBuffer();
// 1 => 1
inputChannels[0][1].readEndOfPartitionEvent();
// 3 => 6
inputChannels[1][3].readEndOfPartitionEvent();
// 1 => 2
inputChannels[0][2].readBuffer();
// 1 => 2
inputChannels[0][2].readEndOfPartitionEvent();
// 4 => 7
inputChannels[1][4].readBuffer();
// 4 => 7
inputChannels[1][4].readEndOfPartitionEvent();
// 0 => 3
inputChannels[1][1].readEndOfPartitionEvent();
// 0 => 3
inputChannels[1][0].readEndOfPartitionEvent();
ig1.notifyChannelNonEmpty(inputChannels[0][0].getInputChannel());
ig1.notifyChannelNonEmpty(inputChannels[0][1].getInputChannel());
ig1.notifyChannelNonEmpty(inputChannels[0][2].getInputChannel());
ig2.notifyChannelNonEmpty(inputChannels[1][0].getInputChannel());
ig2.notifyChannelNonEmpty(inputChannels[1][1].getInputChannel());
ig2.notifyChannelNonEmpty(inputChannels[1][2].getInputChannel());
ig2.notifyChannelNonEmpty(inputChannels[1][3].getInputChannel());
ig2.notifyChannelNonEmpty(inputChannels[1][4].getInputChannel());
// gate 1, channel 0
SingleInputGateTest.verifyBufferOrEvent(union, true, 0);
// gate 2, channel 0
SingleInputGateTest.verifyBufferOrEvent(union, true, 3);
// gate 1, channel 1
SingleInputGateTest.verifyBufferOrEvent(union, true, 1);
// gate 2, channel 1
SingleInputGateTest.verifyBufferOrEvent(union, true, 4);
// gate 1, channel 2
SingleInputGateTest.verifyBufferOrEvent(union, true, 2);
// gate 2, channel 1
SingleInputGateTest.verifyBufferOrEvent(union, true, 5);
// gate 1, channel 0
SingleInputGateTest.verifyBufferOrEvent(union, false, 0);
// gate 2, channel 1
SingleInputGateTest.verifyBufferOrEvent(union, true, 6);
// gate 1, channel 1
SingleInputGateTest.verifyBufferOrEvent(union, false, 1);
// gate 2, channel 1
SingleInputGateTest.verifyBufferOrEvent(union, true, 7);
// gate 1, channel 2
SingleInputGateTest.verifyBufferOrEvent(union, false, 2);
// gate 2, channel 0
SingleInputGateTest.verifyBufferOrEvent(union, false, 3);
// gate 2, channel 1
SingleInputGateTest.verifyBufferOrEvent(union, false, 4);
// gate 2, channel 2
SingleInputGateTest.verifyBufferOrEvent(union, false, 5);
// gate 2, channel 3
SingleInputGateTest.verifyBufferOrEvent(union, false, 6);
// gate 2, channel 4
SingleInputGateTest.verifyBufferOrEvent(union, false, 7);
// Return null when the input gate has received all end-of-partition events
assertTrue(union.isFinished());
assertNull(union.getNextBufferOrEvent());
}
use of org.apache.flink.runtime.jobgraph.IntermediateDataSetID in project flink by apache.
the class ExecutionVertex method createDeploymentDescriptor.
/**
* Creates a task deployment descriptor to deploy a subtask to the given target slot.
*
* TODO: This should actually be in the EXECUTION
*/
TaskDeploymentDescriptor createDeploymentDescriptor(ExecutionAttemptID executionId, SimpleSlot targetSlot, TaskStateHandles taskStateHandles, int attemptNumber) throws ExecutionGraphException {
// Produced intermediate results
List<ResultPartitionDeploymentDescriptor> producedPartitions = new ArrayList<>(resultPartitions.size());
// Consumed intermediate results
List<InputGateDeploymentDescriptor> consumedPartitions = new ArrayList<>(inputEdges.length);
boolean lazyScheduling = getExecutionGraph().getScheduleMode().allowLazyDeployment();
for (IntermediateResultPartition partition : resultPartitions.values()) {
List<List<ExecutionEdge>> consumers = partition.getConsumers();
if (consumers.isEmpty()) {
//TODO this case only exists for test, currently there has to be exactly one consumer in real jobs!
producedPartitions.add(ResultPartitionDeploymentDescriptor.from(partition, ExecutionConfig.UPPER_BOUND_MAX_PARALLELISM, lazyScheduling));
} else {
Preconditions.checkState(1 == consumers.size(), "Only one consumer supported in the current implementation! Found: " + consumers.size());
List<ExecutionEdge> consumer = consumers.get(0);
ExecutionJobVertex vertex = consumer.get(0).getTarget().getJobVertex();
int maxParallelism = vertex.getMaxParallelism();
producedPartitions.add(ResultPartitionDeploymentDescriptor.from(partition, maxParallelism, lazyScheduling));
}
}
for (ExecutionEdge[] edges : inputEdges) {
InputChannelDeploymentDescriptor[] partitions = InputChannelDeploymentDescriptor.fromEdges(edges, targetSlot, lazyScheduling);
// If the produced partition has multiple consumers registered, we
// need to request the one matching our sub task index.
// TODO Refactor after removing the consumers from the intermediate result partitions
int numConsumerEdges = edges[0].getSource().getConsumers().get(0).size();
int queueToRequest = subTaskIndex % numConsumerEdges;
IntermediateResult consumedIntermediateResult = edges[0].getSource().getIntermediateResult();
final IntermediateDataSetID resultId = consumedIntermediateResult.getId();
final ResultPartitionType partitionType = consumedIntermediateResult.getResultType();
consumedPartitions.add(new InputGateDeploymentDescriptor(resultId, partitionType, queueToRequest, partitions));
}
SerializedValue<JobInformation> serializedJobInformation = getExecutionGraph().getSerializedJobInformation();
SerializedValue<TaskInformation> serializedJobVertexInformation = null;
try {
serializedJobVertexInformation = jobVertex.getSerializedTaskInformation();
} catch (IOException e) {
throw new ExecutionGraphException("Could not create a serialized JobVertexInformation for " + jobVertex.getJobVertexId(), e);
}
return new TaskDeploymentDescriptor(serializedJobInformation, serializedJobVertexInformation, executionId, targetSlot.getAllocatedSlot().getSlotAllocationId(), subTaskIndex, attemptNumber, targetSlot.getRoot().getSlotNumber(), taskStateHandles, producedPartitions, consumedPartitions);
}
Aggregations