use of org.apache.flink.runtime.testutils.DirectScheduledExecutorService in project flink by apache.
the class CheckpointCoordinatorTest method testMinCheckpointPause.
@Test
public void testMinCheckpointPause() throws Exception {
// will use a different thread to allow checkpoint triggering before exiting from
// receiveAcknowledgeMessage
ScheduledExecutorService executorService = Executors.newSingleThreadScheduledExecutor();
CheckpointCoordinator coordinator = null;
try {
int pause = 1000;
JobVertexID jobVertexId = new JobVertexID();
ExecutionGraph graph = new CheckpointCoordinatorTestingUtils.CheckpointExecutionGraphBuilder().addJobVertex(jobVertexId).setMainThreadExecutor(ComponentMainThreadExecutorServiceAdapter.forSingleThreadExecutor(new DirectScheduledExecutorService())).build();
ExecutionVertex vertex = graph.getJobVertex(jobVertexId).getTaskVertices()[0];
ExecutionAttemptID attemptId = vertex.getCurrentExecutionAttempt().getAttemptId();
coordinator = new CheckpointCoordinatorBuilder().setTimer(new ScheduledExecutorServiceAdapter(executorService)).setCheckpointCoordinatorConfiguration(CheckpointCoordinatorConfiguration.builder().setCheckpointInterval(pause).setCheckpointTimeout(Long.MAX_VALUE).setMaxConcurrentCheckpoints(1).setMinPauseBetweenCheckpoints(pause).build()).setExecutionGraph(graph).build();
coordinator.startCheckpointScheduler();
coordinator.triggerCheckpoint(// trigger, execute, and later complete by receiveAcknowledgeMessage
true);
coordinator.triggerCheckpoint(// enqueue and later see if it gets executed in the middle of
true);
// receiveAcknowledgeMessage
while (coordinator.getNumberOfPendingCheckpoints() == 0) {
// wait for at least 1 request to be fully processed
Thread.sleep(10);
}
coordinator.receiveAcknowledgeMessage(new AcknowledgeCheckpoint(graph.getJobID(), attemptId, 1L), TASK_MANAGER_LOCATION_INFO);
Thread.sleep(pause / 2);
assertEquals(0, coordinator.getNumberOfPendingCheckpoints());
// make sure that the 2nd request is eventually processed
while (coordinator.getNumberOfPendingCheckpoints() == 0) {
Thread.sleep(1);
}
} finally {
if (coordinator != null) {
coordinator.shutdown();
}
executorService.shutdownNow();
}
}
use of org.apache.flink.runtime.testutils.DirectScheduledExecutorService in project flink by apache.
the class DefaultExecutionGraphDeploymentTest method testNoResourceAvailableFailure.
/**
* Tests that a blocking batch job fails if there are not enough resources left to schedule the
* succeeding tasks. This test case is related to [FLINK-4296] where finished producing tasks
* swallow the fail exception when scheduling a consumer task.
*/
@Test
public void testNoResourceAvailableFailure() throws Exception {
JobVertex v1 = new JobVertex("source");
JobVertex v2 = new JobVertex("sink");
int dop1 = 2;
int dop2 = 2;
v1.setParallelism(dop1);
v2.setParallelism(dop2);
v1.setInvokableClass(BatchTask.class);
v2.setInvokableClass(BatchTask.class);
v2.connectNewDataSetAsInput(v1, DistributionPattern.POINTWISE, ResultPartitionType.BLOCKING);
final JobGraph graph = JobGraphTestUtils.batchJobGraph(v1, v2);
DirectScheduledExecutorService directExecutor = new DirectScheduledExecutorService();
// execution graph that executes actions synchronously
final SchedulerBase scheduler = SchedulerTestingUtils.newSchedulerBuilder(graph, ComponentMainThreadExecutorServiceAdapter.forMainThread()).setExecutionSlotAllocatorFactory(SchedulerTestingUtils.newSlotSharingExecutionSlotAllocatorFactory(TestingPhysicalSlotProvider.createWithLimitedAmountOfPhysicalSlots(1))).setFutureExecutor(directExecutor).setBlobWriter(blobWriter).build();
final ExecutionGraph eg = scheduler.getExecutionGraph();
checkJobOffloaded((DefaultExecutionGraph) eg);
// schedule, this triggers mock deployment
scheduler.startScheduling();
ExecutionAttemptID attemptID = eg.getJobVertex(v1.getID()).getTaskVertices()[0].getCurrentExecutionAttempt().getAttemptId();
scheduler.updateTaskExecutionState(new TaskExecutionState(attemptID, ExecutionState.RUNNING));
scheduler.updateTaskExecutionState(new TaskExecutionState(attemptID, ExecutionState.FINISHED, null));
assertEquals(JobStatus.FAILED, eg.getState());
}
use of org.apache.flink.runtime.testutils.DirectScheduledExecutorService in project flink by apache.
the class DefaultExecutionGraphDeploymentTest method testBuildDeploymentDescriptor.
@Test
public void testBuildDeploymentDescriptor() throws Exception {
final JobVertexID jid1 = new JobVertexID();
final JobVertexID jid2 = new JobVertexID();
final JobVertexID jid3 = new JobVertexID();
final JobVertexID jid4 = new JobVertexID();
JobVertex v1 = new JobVertex("v1", jid1);
JobVertex v2 = new JobVertex("v2", jid2);
JobVertex v3 = new JobVertex("v3", jid3);
JobVertex v4 = new JobVertex("v4", jid4);
v1.setParallelism(10);
v2.setParallelism(10);
v3.setParallelism(10);
v4.setParallelism(10);
v1.setInvokableClass(BatchTask.class);
v2.setInvokableClass(BatchTask.class);
v3.setInvokableClass(BatchTask.class);
v4.setInvokableClass(BatchTask.class);
v2.connectNewDataSetAsInput(v1, DistributionPattern.ALL_TO_ALL, ResultPartitionType.PIPELINED);
v3.connectNewDataSetAsInput(v2, DistributionPattern.ALL_TO_ALL, ResultPartitionType.PIPELINED);
v4.connectNewDataSetAsInput(v2, DistributionPattern.ALL_TO_ALL, ResultPartitionType.PIPELINED);
final JobGraph jobGraph = JobGraphTestUtils.batchJobGraph(v1, v2, v3, v4);
final JobID jobId = jobGraph.getJobID();
DirectScheduledExecutorService executor = new DirectScheduledExecutorService();
DefaultExecutionGraph eg = TestingDefaultExecutionGraphBuilder.newBuilder().setJobGraph(jobGraph).setFutureExecutor(executor).setIoExecutor(executor).setBlobWriter(blobWriter).build();
eg.start(ComponentMainThreadExecutorServiceAdapter.forMainThread());
checkJobOffloaded(eg);
ExecutionJobVertex ejv = eg.getAllVertices().get(jid2);
ExecutionVertex vertex = ejv.getTaskVertices()[3];
final SimpleAckingTaskManagerGateway taskManagerGateway = new SimpleAckingTaskManagerGateway();
final CompletableFuture<TaskDeploymentDescriptor> tdd = new CompletableFuture<>();
taskManagerGateway.setSubmitConsumer(FunctionUtils.uncheckedConsumer(taskDeploymentDescriptor -> {
taskDeploymentDescriptor.loadBigData(blobCache);
tdd.complete(taskDeploymentDescriptor);
}));
final LogicalSlot slot = new TestingLogicalSlotBuilder().setTaskManagerGateway(taskManagerGateway).createTestingLogicalSlot();
assertEquals(ExecutionState.CREATED, vertex.getExecutionState());
vertex.getCurrentExecutionAttempt().transitionState(ExecutionState.SCHEDULED);
vertex.getCurrentExecutionAttempt().registerProducedPartitions(slot.getTaskManagerLocation(), true).get();
vertex.deployToSlot(slot);
assertEquals(ExecutionState.DEPLOYING, vertex.getExecutionState());
checkTaskOffloaded(eg, vertex.getJobvertexId());
TaskDeploymentDescriptor descr = tdd.get();
assertNotNull(descr);
JobInformation jobInformation = descr.getSerializedJobInformation().deserializeValue(getClass().getClassLoader());
TaskInformation taskInformation = descr.getSerializedTaskInformation().deserializeValue(getClass().getClassLoader());
assertEquals(jobId, descr.getJobId());
assertEquals(jobId, jobInformation.getJobId());
assertEquals(jid2, taskInformation.getJobVertexId());
assertEquals(3, descr.getSubtaskIndex());
assertEquals(10, taskInformation.getNumberOfSubtasks());
assertEquals(BatchTask.class.getName(), taskInformation.getInvokableClassName());
assertEquals("v2", taskInformation.getTaskName());
Collection<ResultPartitionDeploymentDescriptor> producedPartitions = descr.getProducedPartitions();
Collection<InputGateDeploymentDescriptor> consumedPartitions = descr.getInputGates();
assertEquals(2, producedPartitions.size());
assertEquals(1, consumedPartitions.size());
Iterator<ResultPartitionDeploymentDescriptor> iteratorProducedPartitions = producedPartitions.iterator();
Iterator<InputGateDeploymentDescriptor> iteratorConsumedPartitions = consumedPartitions.iterator();
assertEquals(10, iteratorProducedPartitions.next().getNumberOfSubpartitions());
assertEquals(10, iteratorProducedPartitions.next().getNumberOfSubpartitions());
ShuffleDescriptor[] shuffleDescriptors = iteratorConsumedPartitions.next().getShuffleDescriptors();
assertEquals(10, shuffleDescriptors.length);
Iterator<ConsumedPartitionGroup> iteratorConsumedPartitionGroup = vertex.getAllConsumedPartitionGroups().iterator();
int idx = 0;
for (IntermediateResultPartitionID partitionId : iteratorConsumedPartitionGroup.next()) {
assertEquals(partitionId, shuffleDescriptors[idx++].getResultPartitionID().getPartitionId());
}
}
use of org.apache.flink.runtime.testutils.DirectScheduledExecutorService in project flink by apache.
the class BatchingStateChangeUploaderTest method testClose.
@Test
public void testClose() throws Exception {
TestingStateChangeUploader probe = new TestingStateChangeUploader();
DirectScheduledExecutorService scheduler = new DirectScheduledExecutorService();
DirectScheduledExecutorService retryScheduler = new DirectScheduledExecutorService();
new BatchingStateChangeUploader(0, 0, MAX_BYTES_IN_FLIGHT, RetryPolicy.NONE, probe, scheduler, new RetryingExecutor(retryScheduler, createUnregisteredChangelogStorageMetricGroup().getAttemptsPerUpload()), createUnregisteredChangelogStorageMetricGroup()).close();
assertTrue(probe.isClosed());
assertTrue(scheduler.isShutdown());
assertTrue(retryScheduler.isShutdown());
}
use of org.apache.flink.runtime.testutils.DirectScheduledExecutorService in project flink by apache.
the class BatchingStateChangeUploaderTest method testRetry.
/**
* Test integration with {@link RetryingExecutor}.
*/
@Test
public void testRetry() throws Exception {
final int maxAttempts = 5;
try (BatchingStateChangeUploader store = new BatchingStateChangeUploader(0, 0, MAX_BYTES_IN_FLIGHT, RetryPolicy.fixed(maxAttempts, 0, 0), new TestingStateChangeUploader() {
final AtomicInteger currentAttempt = new AtomicInteger(0);
@Override
public void upload(UploadTask uploadTask) throws IOException {
if (currentAttempt.getAndIncrement() < maxAttempts - 1) {
throw new IOException();
} else {
uploadTask.complete(emptyList());
}
}
}, new DirectScheduledExecutorService(), new RetryingExecutor(new DirectScheduledExecutorService(), createUnregisteredChangelogStorageMetricGroup().getAttemptsPerUpload()), createUnregisteredChangelogStorageMetricGroup())) {
CompletableFuture<List<UploadResult>> completionFuture = new CompletableFuture<>();
store.upload(new UploadTask(getChanges(4), completionFuture::complete, (unused, throwable) -> completionFuture.completeExceptionally(throwable)));
completionFuture.get();
}
}
Aggregations