Search in sources :

Example 1 with Scheduler

use of org.apache.flink.runtime.jobmanager.scheduler.Scheduler 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 2 with Scheduler

use of org.apache.flink.runtime.jobmanager.scheduler.Scheduler in project flink by apache.

the class ExecutionGraphRestartTest method createExecutionGraph.

private static Tuple2<ExecutionGraph, Instance> createExecutionGraph(RestartStrategy restartStrategy, boolean isSpy) throws Exception {
    Instance instance = ExecutionGraphTestUtils.getInstance(new ActorTaskManagerGateway(new SimpleActorGateway(TestingUtils.directExecutionContext())), NUM_TASKS);
    Scheduler scheduler = new Scheduler(TestingUtils.defaultExecutionContext());
    scheduler.newInstanceAvailable(instance);
    JobVertex sender = newJobVertex("Task", NUM_TASKS, NoOpInvokable.class);
    JobGraph jobGraph = new JobGraph("Pointwise job", sender);
    ExecutionGraph eg = newExecutionGraph(restartStrategy, scheduler);
    if (isSpy) {
        eg = spy(eg);
    }
    eg.attachJobGraph(jobGraph.getVerticesSortedTopologicallyFromSources());
    assertEquals(JobStatus.CREATED, eg.getState());
    eg.scheduleForExecution();
    assertEquals(JobStatus.RUNNING, eg.getState());
    return new Tuple2<>(eg, instance);
}
Also used : JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) Instance(org.apache.flink.runtime.instance.Instance) Scheduler(org.apache.flink.runtime.jobmanager.scheduler.Scheduler) Tuple2(org.apache.flink.api.java.tuple.Tuple2) SimpleActorGateway(org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.SimpleActorGateway) ActorTaskManagerGateway(org.apache.flink.runtime.jobmanager.slots.ActorTaskManagerGateway)

Example 3 with Scheduler

use of org.apache.flink.runtime.jobmanager.scheduler.Scheduler in project flink by apache.

the class ExecutionGraphRestartTest method testConstraintsAfterRestart.

@Test
public void testConstraintsAfterRestart() throws Exception {
    //setting up
    Instance instance = ExecutionGraphTestUtils.getInstance(new ActorTaskManagerGateway(new SimpleActorGateway(TestingUtils.directExecutionContext())), NUM_TASKS);
    Scheduler scheduler = new Scheduler(TestingUtils.defaultExecutionContext());
    scheduler.newInstanceAvailable(instance);
    JobVertex groupVertex = newJobVertex("Task1", NUM_TASKS, NoOpInvokable.class);
    JobVertex groupVertex2 = newJobVertex("Task2", NUM_TASKS, NoOpInvokable.class);
    SlotSharingGroup sharingGroup = new SlotSharingGroup();
    groupVertex.setSlotSharingGroup(sharingGroup);
    groupVertex2.setSlotSharingGroup(sharingGroup);
    groupVertex.setStrictlyCoLocatedWith(groupVertex2);
    //initiate and schedule job
    JobGraph jobGraph = new JobGraph("Pointwise job", groupVertex, groupVertex2);
    ExecutionGraph eg = newExecutionGraph(new FixedDelayRestartStrategy(1, 0L), scheduler);
    eg.attachJobGraph(jobGraph.getVerticesSortedTopologicallyFromSources());
    assertEquals(JobStatus.CREATED, eg.getState());
    eg.scheduleForExecution();
    assertEquals(JobStatus.RUNNING, eg.getState());
    //sanity checks
    validateConstraints(eg);
    //restart automatically
    restartAfterFailure(eg, new FiniteDuration(2, TimeUnit.MINUTES), false);
    //checking execution vertex properties
    validateConstraints(eg);
    haltExecution(eg);
}
Also used : JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) Instance(org.apache.flink.runtime.instance.Instance) FixedDelayRestartStrategy(org.apache.flink.runtime.executiongraph.restart.FixedDelayRestartStrategy) Scheduler(org.apache.flink.runtime.jobmanager.scheduler.Scheduler) FiniteDuration(scala.concurrent.duration.FiniteDuration) SimpleActorGateway(org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.SimpleActorGateway) SlotSharingGroup(org.apache.flink.runtime.jobmanager.scheduler.SlotSharingGroup) ActorTaskManagerGateway(org.apache.flink.runtime.jobmanager.slots.ActorTaskManagerGateway) Test(org.junit.Test)

Example 4 with Scheduler

use of org.apache.flink.runtime.jobmanager.scheduler.Scheduler in project flink by apache.

the class ExecutionGraphRestartTest method testFailExecutionGraphAfterCancel.

/**
	 * Tests that it is possible to fail a graph via a call to
	 * {@link ExecutionGraph#fail(Throwable)} after cancellation.
	 */
@Test
public void testFailExecutionGraphAfterCancel() throws Exception {
    Instance instance = ExecutionGraphTestUtils.getInstance(new ActorTaskManagerGateway(new SimpleActorGateway(TestingUtils.directExecutionContext())), 2);
    Scheduler scheduler = new Scheduler(TestingUtils.defaultExecutionContext());
    scheduler.newInstanceAvailable(instance);
    JobVertex vertex = newJobVertex("Test Vertex", 1, NoOpInvokable.class);
    ExecutionConfig executionConfig = new ExecutionConfig();
    executionConfig.setRestartStrategy(RestartStrategies.fixedDelayRestart(Integer.MAX_VALUE, Integer.MAX_VALUE));
    JobGraph jobGraph = new JobGraph("Test Job", vertex);
    jobGraph.setExecutionConfig(executionConfig);
    ExecutionGraph eg = newExecutionGraph(new InfiniteDelayRestartStrategy(), scheduler);
    eg.attachJobGraph(jobGraph.getVerticesSortedTopologicallyFromSources());
    assertEquals(JobStatus.CREATED, eg.getState());
    eg.scheduleForExecution();
    assertEquals(JobStatus.RUNNING, eg.getState());
    // Fail right after cancel (for example with concurrent slot release)
    eg.cancel();
    assertEquals(JobStatus.CANCELLING, eg.getState());
    eg.fail(new Exception("Test Exception"));
    assertEquals(JobStatus.FAILING, eg.getState());
    Execution execution = eg.getAllExecutionVertices().iterator().next().getCurrentExecutionAttempt();
    execution.cancelingComplete();
    assertEquals(JobStatus.RESTARTING, eg.getState());
}
Also used : JobGraph(org.apache.flink.runtime.jobgraph.JobGraph) JobVertex(org.apache.flink.runtime.jobgraph.JobVertex) InfiniteDelayRestartStrategy(org.apache.flink.runtime.executiongraph.restart.InfiniteDelayRestartStrategy) Instance(org.apache.flink.runtime.instance.Instance) Scheduler(org.apache.flink.runtime.jobmanager.scheduler.Scheduler) ExecutionConfig(org.apache.flink.api.common.ExecutionConfig) SimpleActorGateway(org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.SimpleActorGateway) SuppressRestartsException(org.apache.flink.runtime.execution.SuppressRestartsException) IOException(java.io.IOException) ActorTaskManagerGateway(org.apache.flink.runtime.jobmanager.slots.ActorTaskManagerGateway) Test(org.junit.Test)

Example 5 with Scheduler

use of org.apache.flink.runtime.jobmanager.scheduler.Scheduler in project flink by apache.

the class ExecutionVertexSchedulingTest method testSlotReleasedWhenScheduledQueued.

@Test
public void testSlotReleasedWhenScheduledQueued() {
    try {
        final ExecutionJobVertex ejv = getExecutionVertex(new JobVertexID());
        final ExecutionVertex vertex = new ExecutionVertex(ejv, 0, new IntermediateResult[0], AkkaUtils.getDefaultTimeout());
        // a slot than cannot be deployed to
        final Instance instance = getInstance(new ActorTaskManagerGateway(DummyActorGateway.INSTANCE));
        final SimpleSlot slot = instance.allocateSimpleSlot(ejv.getJobId());
        slot.releaseSlot();
        assertTrue(slot.isReleased());
        final FlinkCompletableFuture<SimpleSlot> future = new FlinkCompletableFuture<>();
        Scheduler scheduler = mock(Scheduler.class);
        when(scheduler.allocateSlot(Matchers.any(ScheduledUnit.class), anyBoolean())).thenReturn(future);
        assertEquals(ExecutionState.CREATED, vertex.getExecutionState());
        // try to deploy to the slot
        vertex.scheduleForExecution(scheduler, true);
        // future has not yet a slot
        assertEquals(ExecutionState.SCHEDULED, vertex.getExecutionState());
        future.complete(slot);
        // will have failed
        assertEquals(ExecutionState.FAILED, vertex.getExecutionState());
    } catch (Exception e) {
        e.printStackTrace();
        fail(e.getMessage());
    }
}
Also used : Instance(org.apache.flink.runtime.instance.Instance) ExecutionGraphTestUtils.getInstance(org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.getInstance) Scheduler(org.apache.flink.runtime.jobmanager.scheduler.Scheduler) JobVertexID(org.apache.flink.runtime.jobgraph.JobVertexID) ScheduledUnit(org.apache.flink.runtime.jobmanager.scheduler.ScheduledUnit) SimpleSlot(org.apache.flink.runtime.instance.SimpleSlot) FlinkCompletableFuture(org.apache.flink.runtime.concurrent.impl.FlinkCompletableFuture) ExecutionGraphTestUtils.getExecutionVertex(org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.getExecutionVertex) ActorTaskManagerGateway(org.apache.flink.runtime.jobmanager.slots.ActorTaskManagerGateway) Test(org.junit.Test)

Aggregations

Scheduler (org.apache.flink.runtime.jobmanager.scheduler.Scheduler)29 JobVertex (org.apache.flink.runtime.jobgraph.JobVertex)23 Test (org.junit.Test)23 JobID (org.apache.flink.api.common.JobID)20 ExecutionConfig (org.apache.flink.api.common.ExecutionConfig)19 Configuration (org.apache.flink.configuration.Configuration)19 NoRestartStrategy (org.apache.flink.runtime.executiongraph.restart.NoRestartStrategy)16 ActorTaskManagerGateway (org.apache.flink.runtime.jobmanager.slots.ActorTaskManagerGateway)14 Instance (org.apache.flink.runtime.instance.Instance)12 ArrayList (java.util.ArrayList)9 JobException (org.apache.flink.runtime.JobException)9 JobVertexID (org.apache.flink.runtime.jobgraph.JobVertexID)9 SimpleSlot (org.apache.flink.runtime.instance.SimpleSlot)8 JobGraph (org.apache.flink.runtime.jobgraph.JobGraph)8 IOException (java.io.IOException)7 SimpleActorGateway (org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.SimpleActorGateway)5 FlinkCompletableFuture (org.apache.flink.runtime.concurrent.impl.FlinkCompletableFuture)4 SuppressRestartsException (org.apache.flink.runtime.execution.SuppressRestartsException)4 ExecutionGraphTestUtils.getInstance (org.apache.flink.runtime.executiongraph.ExecutionGraphTestUtils.getInstance)4 IntermediateDataSet (org.apache.flink.runtime.jobgraph.IntermediateDataSet)4