Search in sources :

Example 21 with AllocationID

use of org.apache.flink.runtime.clusterframework.types.AllocationID in project flink by apache.

the class ExecutionGraphMetricsTest method testExecutionGraphRestartTimeMetric.

/**
	 * This test tests that the restarting time metric correctly displays restarting times.
	 */
@Test
public void testExecutionGraphRestartTimeMetric() throws JobException, IOException, InterruptedException {
    final ScheduledExecutorService executor = Executors.newSingleThreadScheduledExecutor();
    try {
        // setup execution graph with mocked scheduling logic
        int parallelism = 1;
        JobVertex jobVertex = new JobVertex("TestVertex");
        jobVertex.setParallelism(parallelism);
        jobVertex.setInvokableClass(NoOpInvokable.class);
        JobGraph jobGraph = new JobGraph("Test Job", jobVertex);
        Configuration config = new Configuration();
        config.setString(ConfigConstants.METRICS_REPORTERS_LIST, "test");
        config.setString(ConfigConstants.METRICS_REPORTER_PREFIX + "test." + ConfigConstants.METRICS_REPORTER_CLASS_SUFFIX, TestingReporter.class.getName());
        Configuration jobConfig = new Configuration();
        Time timeout = Time.seconds(10L);
        MetricRegistry metricRegistry = new MetricRegistry(MetricRegistryConfiguration.fromConfiguration(config));
        assertTrue(metricRegistry.getReporters().size() == 1);
        MetricReporter reporter = metricRegistry.getReporters().get(0);
        assertTrue(reporter instanceof TestingReporter);
        TestingReporter testingReporter = (TestingReporter) reporter;
        MetricGroup metricGroup = new JobManagerMetricGroup(metricRegistry, "localhost");
        Scheduler scheduler = mock(Scheduler.class);
        ResourceID taskManagerId = ResourceID.generate();
        TaskManagerLocation taskManagerLocation = mock(TaskManagerLocation.class);
        when(taskManagerLocation.getResourceID()).thenReturn(taskManagerId);
        when(taskManagerLocation.getHostname()).thenReturn("localhost");
        TaskManagerGateway taskManagerGateway = mock(TaskManagerGateway.class);
        Instance instance = mock(Instance.class);
        when(instance.getTaskManagerLocation()).thenReturn(taskManagerLocation);
        when(instance.getTaskManagerID()).thenReturn(taskManagerId);
        when(instance.getTaskManagerGateway()).thenReturn(taskManagerGateway);
        Slot rootSlot = mock(Slot.class);
        AllocatedSlot mockAllocatedSlot = mock(AllocatedSlot.class);
        when(mockAllocatedSlot.getSlotAllocationId()).thenReturn(new AllocationID());
        SimpleSlot simpleSlot = mock(SimpleSlot.class);
        when(simpleSlot.isAlive()).thenReturn(true);
        when(simpleSlot.getTaskManagerLocation()).thenReturn(taskManagerLocation);
        when(simpleSlot.getTaskManagerID()).thenReturn(taskManagerId);
        when(simpleSlot.getTaskManagerGateway()).thenReturn(taskManagerGateway);
        when(simpleSlot.setExecutedVertex(Matchers.any(Execution.class))).thenReturn(true);
        when(simpleSlot.getRoot()).thenReturn(rootSlot);
        when(simpleSlot.getAllocatedSlot()).thenReturn(mockAllocatedSlot);
        FlinkCompletableFuture<SimpleSlot> future = new FlinkCompletableFuture<>();
        future.complete(simpleSlot);
        when(scheduler.allocateSlot(any(ScheduledUnit.class), anyBoolean())).thenReturn(future);
        when(rootSlot.getSlotNumber()).thenReturn(0);
        when(taskManagerGateway.submitTask(any(TaskDeploymentDescriptor.class), any(Time.class))).thenReturn(FlinkCompletableFuture.completed(Acknowledge.get()));
        TestingRestartStrategy testingRestartStrategy = new TestingRestartStrategy();
        ExecutionGraph executionGraph = new ExecutionGraph(executor, executor, jobGraph.getJobID(), jobGraph.getName(), jobConfig, new SerializedValue<ExecutionConfig>(null), timeout, testingRestartStrategy, Collections.<BlobKey>emptyList(), Collections.<URL>emptyList(), scheduler, getClass().getClassLoader(), metricGroup);
        // get restarting time metric
        Metric metric = testingReporter.getMetric(ExecutionGraph.RESTARTING_TIME_METRIC_NAME);
        assertNotNull(metric);
        assertTrue(metric instanceof Gauge);
        @SuppressWarnings("unchecked") Gauge<Long> restartingTime = (Gauge<Long>) metric;
        // check that the restarting time is 0 since it's the initial start
        assertTrue(0L == restartingTime.getValue());
        executionGraph.attachJobGraph(jobGraph.getVerticesSortedTopologicallyFromSources());
        // start execution
        executionGraph.scheduleForExecution();
        assertTrue(0L == restartingTime.getValue());
        List<ExecutionAttemptID> executionIDs = new ArrayList<>();
        for (ExecutionVertex executionVertex : executionGraph.getAllExecutionVertices()) {
            executionIDs.add(executionVertex.getCurrentExecutionAttempt().getAttemptId());
        }
        // tell execution graph that the tasks are in state running --> job status switches to state running
        for (ExecutionAttemptID executionID : executionIDs) {
            executionGraph.updateState(new TaskExecutionState(jobGraph.getJobID(), executionID, ExecutionState.RUNNING));
        }
        assertEquals(JobStatus.RUNNING, executionGraph.getState());
        assertTrue(0L == restartingTime.getValue());
        // fail the job so that it goes into state restarting
        for (ExecutionAttemptID executionID : executionIDs) {
            executionGraph.updateState(new TaskExecutionState(jobGraph.getJobID(), executionID, ExecutionState.FAILED, new Exception()));
        }
        assertEquals(JobStatus.RESTARTING, executionGraph.getState());
        long firstRestartingTimestamp = executionGraph.getStatusTimestamp(JobStatus.RESTARTING);
        // wait some time so that the restarting time gauge shows a value different from 0
        Thread.sleep(50);
        long previousRestartingTime = restartingTime.getValue();
        // check that the restarting time is monotonically increasing
        for (int i = 0; i < 10; i++) {
            long currentRestartingTime = restartingTime.getValue();
            assertTrue(currentRestartingTime >= previousRestartingTime);
            previousRestartingTime = currentRestartingTime;
        }
        // check that we have measured some restarting time
        assertTrue(previousRestartingTime > 0);
        // restart job
        testingRestartStrategy.restartExecutionGraph();
        executionIDs.clear();
        for (ExecutionVertex executionVertex : executionGraph.getAllExecutionVertices()) {
            executionIDs.add(executionVertex.getCurrentExecutionAttempt().getAttemptId());
        }
        for (ExecutionAttemptID executionID : executionIDs) {
            executionGraph.updateState(new TaskExecutionState(jobGraph.getJobID(), executionID, ExecutionState.RUNNING));
        }
        assertEquals(JobStatus.RUNNING, executionGraph.getState());
        assertTrue(firstRestartingTimestamp != 0);
        previousRestartingTime = restartingTime.getValue();
        // check that the restarting time does not increase after we've reached the running state
        for (int i = 0; i < 10; i++) {
            long currentRestartingTime = restartingTime.getValue();
            assertTrue(currentRestartingTime == previousRestartingTime);
            previousRestartingTime = currentRestartingTime;
        }
        // fail job again
        for (ExecutionAttemptID executionID : executionIDs) {
            executionGraph.updateState(new TaskExecutionState(jobGraph.getJobID(), executionID, ExecutionState.FAILED, new Exception()));
        }
        assertEquals(JobStatus.RESTARTING, executionGraph.getState());
        long secondRestartingTimestamp = executionGraph.getStatusTimestamp(JobStatus.RESTARTING);
        assertTrue(firstRestartingTimestamp != secondRestartingTimestamp);
        Thread.sleep(50);
        previousRestartingTime = restartingTime.getValue();
        // check that the restarting time is increasing again
        for (int i = 0; i < 10; i++) {
            long currentRestartingTime = restartingTime.getValue();
            assertTrue(currentRestartingTime >= previousRestartingTime);
            previousRestartingTime = currentRestartingTime;
        }
        assertTrue(previousRestartingTime > 0);
        // now lets fail the job while it is in restarting and see whether the restarting time then stops to increase
        // for this to work, we have to use a SuppressRestartException
        executionGraph.fail(new SuppressRestartsException(new Exception()));
        assertEquals(JobStatus.FAILED, executionGraph.getState());
        previousRestartingTime = restartingTime.getValue();
        for (int i = 0; i < 10; i++) {
            long currentRestartingTime = restartingTime.getValue();
            assertTrue(currentRestartingTime == previousRestartingTime);
            previousRestartingTime = currentRestartingTime;
        }
    } finally {
        executor.shutdownNow();
    }
}
Also used : JobManagerMetricGroup(org.apache.flink.runtime.metrics.groups.JobManagerMetricGroup) MetricRegistryConfiguration(org.apache.flink.runtime.metrics.MetricRegistryConfiguration) Configuration(org.apache.flink.configuration.Configuration) Instance(org.apache.flink.runtime.instance.Instance) Scheduler(org.apache.flink.runtime.jobmanager.scheduler.Scheduler) MetricGroup(org.apache.flink.metrics.MetricGroup) JobManagerMetricGroup(org.apache.flink.runtime.metrics.groups.JobManagerMetricGroup) TaskManagerGateway(org.apache.flink.runtime.jobmanager.slots.TaskManagerGateway) ArrayList(java.util.ArrayList) Time(org.apache.flink.api.common.time.Time) ExecutionConfig(org.apache.flink.api.common.ExecutionConfig) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) FlinkCompletableFuture(org.apache.flink.runtime.concurrent.impl.FlinkCompletableFuture) Gauge(org.apache.flink.metrics.Gauge) SuppressRestartsException(org.apache.flink.runtime.execution.SuppressRestartsException) ResourceID(org.apache.flink.runtime.clusterframework.types.ResourceID) TaskDeploymentDescriptor(org.apache.flink.runtime.deployment.TaskDeploymentDescriptor) ScheduledExecutorService(java.util.concurrent.ScheduledExecutorService) AllocatedSlot(org.apache.flink.runtime.jobmanager.slots.AllocatedSlot) TaskManagerLocation(org.apache.flink.runtime.taskmanager.TaskManagerLocation) MetricRegistry(org.apache.flink.runtime.metrics.MetricRegistry) AllocationID(org.apache.flink.runtime.clusterframework.types.AllocationID) ScheduledUnit(org.apache.flink.runtime.jobmanager.scheduler.ScheduledUnit) MetricReporter(org.apache.flink.metrics.reporter.MetricReporter) TaskExecutionState(org.apache.flink.runtime.taskmanager.TaskExecutionState) SuppressRestartsException(org.apache.flink.runtime.execution.SuppressRestartsException) JobException(org.apache.flink.runtime.JobException) IOException(java.io.IOException) JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) Slot(org.apache.flink.runtime.instance.Slot) AllocatedSlot(org.apache.flink.runtime.jobmanager.slots.AllocatedSlot) Metric(org.apache.flink.metrics.Metric) Test(org.junit.Test)

Example 22 with AllocationID

use of org.apache.flink.runtime.clusterframework.types.AllocationID in project flink by apache.

the class AvailableSlotsTest method createAllocatedSlot.

static AllocatedSlot createAllocatedSlot(final ResourceID resourceId) {
    TaskManagerLocation mockTaskManagerLocation = mock(TaskManagerLocation.class);
    when(mockTaskManagerLocation.getResourceID()).thenReturn(resourceId);
    TaskManagerGateway mockTaskManagerGateway = mock(TaskManagerGateway.class);
    return new AllocatedSlot(new AllocationID(), new JobID(), mockTaskManagerLocation, 0, DEFAULT_TESTING_PROFILE, mockTaskManagerGateway);
}
Also used : AllocatedSlot(org.apache.flink.runtime.jobmanager.slots.AllocatedSlot) TaskManagerLocation(org.apache.flink.runtime.taskmanager.TaskManagerLocation) AllocationID(org.apache.flink.runtime.clusterframework.types.AllocationID) TaskManagerGateway(org.apache.flink.runtime.jobmanager.slots.TaskManagerGateway) JobID(org.apache.flink.api.common.JobID)

Example 23 with AllocationID

use of org.apache.flink.runtime.clusterframework.types.AllocationID in project flink by apache.

the class ExecutionVertexDeploymentTest method testTddProducedPartitionsLazyScheduling.

/**
	 * Tests that the lazy scheduling flag is correctly forwarded to the produced partition descriptors.
	 */
@Test
public void testTddProducedPartitionsLazyScheduling() throws Exception {
    TestingUtils.QueuedActionExecutionContext context = TestingUtils.queuedActionExecutionContext();
    ExecutionJobVertex jobVertex = getExecutionVertex(new JobVertexID(), context);
    IntermediateResult result = new IntermediateResult(new IntermediateDataSetID(), jobVertex, 1, ResultPartitionType.PIPELINED);
    ExecutionVertex vertex = new ExecutionVertex(jobVertex, 0, new IntermediateResult[] { result }, Time.minutes(1));
    ExecutionEdge mockEdge = createMockExecutionEdge(1);
    result.getPartitions()[0].addConsumerGroup();
    result.getPartitions()[0].addConsumer(mockEdge, 0);
    AllocatedSlot allocatedSlot = mock(AllocatedSlot.class);
    when(allocatedSlot.getSlotAllocationId()).thenReturn(new AllocationID());
    Slot root = mock(Slot.class);
    when(root.getSlotNumber()).thenReturn(1);
    SimpleSlot slot = mock(SimpleSlot.class);
    when(slot.getRoot()).thenReturn(root);
    when(slot.getAllocatedSlot()).thenReturn(allocatedSlot);
    when(root.getAllocatedSlot()).thenReturn(allocatedSlot);
    for (ScheduleMode mode : ScheduleMode.values()) {
        vertex.getExecutionGraph().setScheduleMode(mode);
        TaskDeploymentDescriptor tdd = vertex.createDeploymentDescriptor(new ExecutionAttemptID(), slot, null, 1);
        Collection<ResultPartitionDeploymentDescriptor> producedPartitions = tdd.getProducedPartitions();
        assertEquals(1, producedPartitions.size());
        ResultPartitionDeploymentDescriptor desc = producedPartitions.iterator().next();
        assertEquals(mode.allowLazyDeployment(), desc.sendScheduleOrUpdateConsumersMessage());
    }
}
Also used : AllocatedSlot(org.apache.flink.runtime.jobmanager.slots.AllocatedSlot) ResultPartitionDeploymentDescriptor(org.apache.flink.runtime.deployment.ResultPartitionDeploymentDescriptor) JobVertexID(org.apache.flink.runtime.jobgraph.JobVertexID) AllocationID(org.apache.flink.runtime.clusterframework.types.AllocationID) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) ExecutionGraphTestUtils.getExecutionVertex(org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.getExecutionVertex) TestingUtils(org.apache.flink.runtime.testingUtils.TestingUtils) ScheduleMode(org.apache.flink.runtime.jobgraph.ScheduleMode) Slot(org.apache.flink.runtime.instance.Slot) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) AllocatedSlot(org.apache.flink.runtime.jobmanager.slots.AllocatedSlot) IntermediateDataSetID(org.apache.flink.runtime.jobgraph.IntermediateDataSetID) TaskDeploymentDescriptor(org.apache.flink.runtime.deployment.TaskDeploymentDescriptor) Test(org.junit.Test)

Example 24 with AllocationID

use of org.apache.flink.runtime.clusterframework.types.AllocationID in project flink by apache.

the class SlotPoolTest method testOfferSlot.

@Test
public void testOfferSlot() throws Exception {
    ResourceID resourceID = new ResourceID("resource");
    slotPool.registerTaskManager(resourceID);
    Future<SimpleSlot> future = slotPool.allocateSlot(mock(ScheduledUnit.class), DEFAULT_TESTING_PROFILE, null);
    assertFalse(future.isDone());
    ArgumentCaptor<SlotRequest> slotRequestArgumentCaptor = ArgumentCaptor.forClass(SlotRequest.class);
    verify(resourceManagerGateway).requestSlot(any(UUID.class), any(UUID.class), slotRequestArgumentCaptor.capture(), any(Time.class));
    final SlotRequest slotRequest = slotRequestArgumentCaptor.getValue();
    // slot from unregistered resource
    AllocatedSlot invalid = createAllocatedSlot(new ResourceID("unregistered"), slotRequest.getAllocationId(), jobId, DEFAULT_TESTING_PROFILE);
    assertFalse(slotPool.offerSlot(invalid));
    AllocatedSlot notRequested = createAllocatedSlot(resourceID, new AllocationID(), jobId, DEFAULT_TESTING_PROFILE);
    // we'll also accept non requested slots
    assertTrue(slotPool.offerSlot(notRequested));
    AllocatedSlot allocatedSlot = createAllocatedSlot(resourceID, slotRequest.getAllocationId(), jobId, DEFAULT_TESTING_PROFILE);
    // accepted slot
    assertTrue(slotPool.offerSlot(allocatedSlot));
    SimpleSlot slot = future.get(1, TimeUnit.SECONDS);
    assertTrue(future.isDone());
    assertTrue(slot.isAlive());
    // duplicated offer with using slot
    assertTrue(slotPool.offerSlot(allocatedSlot));
    assertTrue(future.isDone());
    assertTrue(slot.isAlive());
    // duplicated offer with free slot
    slot.releaseSlot();
    assertTrue(slot.isReleased());
    assertTrue(slotPool.offerSlot(allocatedSlot));
}
Also used : AllocatedSlot(org.apache.flink.runtime.jobmanager.slots.AllocatedSlot) ResourceID(org.apache.flink.runtime.clusterframework.types.ResourceID) AllocationID(org.apache.flink.runtime.clusterframework.types.AllocationID) ScheduledUnit(org.apache.flink.runtime.jobmanager.scheduler.ScheduledUnit) Time(org.apache.flink.api.common.time.Time) SlotRequest(org.apache.flink.runtime.resourcemanager.SlotRequest) UUID(java.util.UUID) Test(org.junit.Test)

Example 25 with AllocationID

use of org.apache.flink.runtime.clusterframework.types.AllocationID in project flink by apache.

the class TaskDeploymentDescriptorTest method testSerialization.

@Test
public void testSerialization() {
    try {
        final JobID jobID = new JobID();
        final JobVertexID vertexID = new JobVertexID();
        final ExecutionAttemptID execId = new ExecutionAttemptID();
        final AllocationID allocationId = new AllocationID();
        final String jobName = "job name";
        final String taskName = "task name";
        final int numberOfKeyGroups = 1;
        final int indexInSubtaskGroup = 0;
        final int currentNumberOfSubtasks = 1;
        final int attemptNumber = 0;
        final Configuration jobConfiguration = new Configuration();
        final Configuration taskConfiguration = new Configuration();
        final Class<? extends AbstractInvokable> invokableClass = BatchTask.class;
        final List<ResultPartitionDeploymentDescriptor> producedResults = new ArrayList<ResultPartitionDeploymentDescriptor>(0);
        final List<InputGateDeploymentDescriptor> inputGates = new ArrayList<InputGateDeploymentDescriptor>(0);
        final List<BlobKey> requiredJars = new ArrayList<BlobKey>(0);
        final List<URL> requiredClasspaths = new ArrayList<URL>(0);
        final SerializedValue<ExecutionConfig> executionConfig = new SerializedValue<>(new ExecutionConfig());
        final SerializedValue<JobInformation> serializedJobInformation = new SerializedValue<>(new JobInformation(jobID, jobName, executionConfig, jobConfiguration, requiredJars, requiredClasspaths));
        final SerializedValue<TaskInformation> serializedJobVertexInformation = new SerializedValue<>(new TaskInformation(vertexID, taskName, currentNumberOfSubtasks, numberOfKeyGroups, invokableClass.getName(), taskConfiguration));
        final int targetSlotNumber = 47;
        final TaskStateHandles taskStateHandles = new TaskStateHandles();
        final TaskDeploymentDescriptor orig = new TaskDeploymentDescriptor(serializedJobInformation, serializedJobVertexInformation, execId, allocationId, indexInSubtaskGroup, attemptNumber, targetSlotNumber, taskStateHandles, producedResults, inputGates);
        final TaskDeploymentDescriptor copy = CommonTestUtils.createCopySerializable(orig);
        assertFalse(orig.getSerializedJobInformation() == copy.getSerializedJobInformation());
        assertFalse(orig.getSerializedTaskInformation() == copy.getSerializedTaskInformation());
        assertFalse(orig.getExecutionAttemptId() == copy.getExecutionAttemptId());
        assertFalse(orig.getTaskStateHandles() == copy.getTaskStateHandles());
        assertFalse(orig.getProducedPartitions() == copy.getProducedPartitions());
        assertFalse(orig.getInputGates() == copy.getInputGates());
        assertEquals(orig.getSerializedJobInformation(), copy.getSerializedJobInformation());
        assertEquals(orig.getSerializedTaskInformation(), copy.getSerializedTaskInformation());
        assertEquals(orig.getExecutionAttemptId(), copy.getExecutionAttemptId());
        assertEquals(orig.getAllocationId(), copy.getAllocationId());
        assertEquals(orig.getSubtaskIndex(), copy.getSubtaskIndex());
        assertEquals(orig.getAttemptNumber(), copy.getAttemptNumber());
        assertEquals(orig.getTargetSlotNumber(), copy.getTargetSlotNumber());
        assertEquals(orig.getTaskStateHandles(), copy.getTaskStateHandles());
        assertEquals(orig.getProducedPartitions(), copy.getProducedPartitions());
        assertEquals(orig.getInputGates(), copy.getInputGates());
    } catch (Exception e) {
        e.printStackTrace();
        fail(e.getMessage());
    }
}
Also used : Configuration(org.apache.flink.configuration.Configuration) JobVertexID(org.apache.flink.runtime.jobgraph.JobVertexID) ArrayList(java.util.ArrayList) ExecutionConfig(org.apache.flink.api.common.ExecutionConfig) URL(java.net.URL) BlobKey(org.apache.flink.runtime.blob.BlobKey) JobInformation(org.apache.flink.runtime.executiongraph.JobInformation) ExecutionAttemptID(org.apache.flink.runtime.executiongraph.ExecutionAttemptID) TaskInformation(org.apache.flink.runtime.executiongraph.TaskInformation) BatchTask(org.apache.flink.runtime.operators.BatchTask) AllocationID(org.apache.flink.runtime.clusterframework.types.AllocationID) SerializedValue(org.apache.flink.util.SerializedValue) TaskStateHandles(org.apache.flink.runtime.state.TaskStateHandles) JobID(org.apache.flink.api.common.JobID) Test(org.junit.Test)

Aggregations

AllocationID (org.apache.flink.runtime.clusterframework.types.AllocationID)45 JobID (org.apache.flink.api.common.JobID)29 Test (org.junit.Test)27 SlotRequest (org.apache.flink.runtime.resourcemanager.SlotRequest)16 ResourceID (org.apache.flink.runtime.clusterframework.types.ResourceID)13 SlotID (org.apache.flink.runtime.clusterframework.types.SlotID)13 Configuration (org.apache.flink.configuration.Configuration)12 BroadcastVariableManager (org.apache.flink.runtime.broadcast.BroadcastVariableManager)12 FileCache (org.apache.flink.runtime.filecache.FileCache)12 IOManager (org.apache.flink.runtime.io.disk.iomanager.IOManager)12 MemoryManager (org.apache.flink.runtime.memory.MemoryManager)12 NetworkEnvironment (org.apache.flink.runtime.io.network.NetworkEnvironment)11 TaskManagerLocation (org.apache.flink.runtime.taskmanager.TaskManagerLocation)10 UUID (java.util.UUID)9 ExecutionConfig (org.apache.flink.api.common.ExecutionConfig)9 Time (org.apache.flink.api.common.time.Time)9 JobInformation (org.apache.flink.runtime.executiongraph.JobInformation)9 TaskInformation (org.apache.flink.runtime.executiongraph.TaskInformation)9 JobVertexID (org.apache.flink.runtime.jobgraph.JobVertexID)9 MetricRegistry (org.apache.flink.runtime.metrics.MetricRegistry)9