Examples with ExecutionConfig org.apache.flink.api.common.ExecutionConfig used on opensource projects

Example 76 with ExecutionConfig

use of org.apache.flink.api.common.ExecutionConfig 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();
    }
}

Example 77 with ExecutionConfig

use of org.apache.flink.api.common.ExecutionConfig 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());
}

Example 78 with ExecutionConfig

use of org.apache.flink.api.common.ExecutionConfig in project flink by apache.

the class PointwisePatternTest method testLowToHigh.

private void testLowToHigh(int lowDop, int highDop) throws Exception {
    if (highDop < lowDop) {
        throw new IllegalArgumentException();
    }
    final int factor = highDop / lowDop;
    final int delta = highDop % lowDop == 0 ? 0 : 1;
    JobVertex v1 = new JobVertex("vertex1");
    JobVertex v2 = new JobVertex("vertex2");
    v1.setParallelism(lowDop);
    v2.setParallelism(highDop);
    v1.setInvokableClass(AbstractInvokable.class);
    v2.setInvokableClass(AbstractInvokable.class);
    v2.connectNewDataSetAsInput(v1, DistributionPattern.POINTWISE, ResultPartitionType.PIPELINED);
    List<JobVertex> ordered = new ArrayList<JobVertex>(Arrays.asList(v1, v2));
    ExecutionGraph eg = new ExecutionGraph(TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), jobId, jobName, cfg, new SerializedValue<>(new ExecutionConfig()), AkkaUtils.getDefaultTimeout(), new NoRestartStrategy(), new Scheduler(TestingUtils.defaultExecutionContext()));
    try {
        eg.attachJobGraph(ordered);
    } catch (JobException e) {
        e.printStackTrace();
        fail("Job failed with exception: " + e.getMessage());
    }
    ExecutionJobVertex target = eg.getAllVertices().get(v2.getID());
    int[] timesUsed = new int[lowDop];
    for (ExecutionVertex ev : target.getTaskVertices()) {
        assertEquals(1, ev.getNumberOfInputs());
        ExecutionEdge[] inEdges = ev.getInputEdges(0);
        assertEquals(1, inEdges.length);
        timesUsed[inEdges[0].getSource().getPartitionNumber()]++;
    }
    for (int used : timesUsed) {
        assertTrue(used >= factor && used <= factor + delta);
    }
}

Example 79 with ExecutionConfig

use of org.apache.flink.api.common.ExecutionConfig in project flink by apache.

the class PointwisePatternTest method testNTo7N.

@Test
public void testNTo7N() throws Exception {
    final int N = 11;
    JobVertex v1 = new JobVertex("vertex1");
    JobVertex v2 = new JobVertex("vertex2");
    v1.setParallelism(N);
    v2.setParallelism(7 * N);
    v1.setInvokableClass(AbstractInvokable.class);
    v2.setInvokableClass(AbstractInvokable.class);
    v2.connectNewDataSetAsInput(v1, DistributionPattern.POINTWISE, ResultPartitionType.PIPELINED);
    List<JobVertex> ordered = new ArrayList<JobVertex>(Arrays.asList(v1, v2));
    ExecutionGraph eg = new ExecutionGraph(TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), jobId, jobName, cfg, new SerializedValue<>(new ExecutionConfig()), AkkaUtils.getDefaultTimeout(), new NoRestartStrategy(), new Scheduler(TestingUtils.defaultExecutionContext()));
    try {
        eg.attachJobGraph(ordered);
    } catch (JobException e) {
        e.printStackTrace();
        fail("Job failed with exception: " + e.getMessage());
    }
    ExecutionJobVertex target = eg.getAllVertices().get(v2.getID());
    for (ExecutionVertex ev : target.getTaskVertices()) {
        assertEquals(1, ev.getNumberOfInputs());
        ExecutionEdge[] inEdges = ev.getInputEdges(0);
        assertEquals(1, inEdges.length);
        assertEquals(ev.getParallelSubtaskIndex() / 7, inEdges[0].getSource().getPartitionNumber());
    }
}

Example 80 with ExecutionConfig

use of org.apache.flink.api.common.ExecutionConfig in project flink by apache.

the class PointwisePatternTest method test2NToN.

@Test
public void test2NToN() throws Exception {
    final int N = 17;
    JobVertex v1 = new JobVertex("vertex1");
    JobVertex v2 = new JobVertex("vertex2");
    v1.setParallelism(2 * N);
    v2.setParallelism(N);
    v1.setInvokableClass(AbstractInvokable.class);
    v2.setInvokableClass(AbstractInvokable.class);
    v2.connectNewDataSetAsInput(v1, DistributionPattern.POINTWISE, ResultPartitionType.PIPELINED);
    List<JobVertex> ordered = new ArrayList<JobVertex>(Arrays.asList(v1, v2));
    ExecutionGraph eg = new ExecutionGraph(TestingUtils.defaultExecutor(), TestingUtils.defaultExecutor(), jobId, jobName, cfg, new SerializedValue<>(new ExecutionConfig()), AkkaUtils.getDefaultTimeout(), new NoRestartStrategy(), new Scheduler(TestingUtils.defaultExecutionContext()));
    try {
        eg.attachJobGraph(ordered);
    } catch (JobException e) {
        e.printStackTrace();
        fail("Job failed with exception: " + e.getMessage());
    }
    ExecutionJobVertex target = eg.getAllVertices().get(v2.getID());
    for (ExecutionVertex ev : target.getTaskVertices()) {
        assertEquals(1, ev.getNumberOfInputs());
        ExecutionEdge[] inEdges = ev.getInputEdges(0);
        assertEquals(2, inEdges.length);
        assertEquals(ev.getParallelSubtaskIndex() * 2, inEdges[0].getSource().getPartitionNumber());
        assertEquals(ev.getParallelSubtaskIndex() * 2 + 1, inEdges[1].getSource().getPartitionNumber());
    }
}