Examples with MetricGroup org.apache.flink.metrics.MetricGroup used on opensource projects

Example 6 with MetricGroup

use of org.apache.flink.metrics.MetricGroup 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 7 with MetricGroup

use of org.apache.flink.metrics.MetricGroup in project flink by apache.

the class MetricUtils method instantiateGarbageCollectorMetrics.

private static void instantiateGarbageCollectorMetrics(MetricGroup metrics) {
    List<GarbageCollectorMXBean> garbageCollectors = ManagementFactory.getGarbageCollectorMXBeans();
    for (final GarbageCollectorMXBean garbageCollector : garbageCollectors) {
        MetricGroup gcGroup = metrics.addGroup(garbageCollector.getName());
        gcGroup.gauge("Count", new Gauge<Long>() {

            @Override
            public Long getValue() {
                return garbageCollector.getCollectionCount();
            }
        });
        gcGroup.gauge("Time", new Gauge<Long>() {

            @Override
            public Long getValue() {
                return garbageCollector.getCollectionTime();
            }
        });
    }
}

Example 8 with MetricGroup

use of org.apache.flink.metrics.MetricGroup in project flink by apache.

the class MetricUtils method instantiateStatusMetrics.

public static void instantiateStatusMetrics(MetricGroup metrics) {
    MetricGroup status = metrics.addGroup(METRIC_GROUP_STATUS_NAME);
    MetricGroup jvm = status.addGroup("JVM");
    instantiateClassLoaderMetrics(jvm.addGroup("ClassLoader"));
    instantiateGarbageCollectorMetrics(jvm.addGroup("GarbageCollector"));
    instantiateMemoryMetrics(jvm.addGroup("Memory"));
    instantiateThreadMetrics(jvm.addGroup("Threads"));
    instantiateCPUMetrics(jvm.addGroup("CPU"));
}

Example 9 with MetricGroup

use of org.apache.flink.metrics.MetricGroup in project flink by apache.

the class MetricUtils method instantiateMemoryMetrics.

private static void instantiateMemoryMetrics(MetricGroup metrics) {
    final MemoryMXBean mxBean = ManagementFactory.getMemoryMXBean();
    MetricGroup heap = metrics.addGroup("Heap");
    heap.gauge("Used", new Gauge<Long>() {

        @Override
        public Long getValue() {
            return mxBean.getHeapMemoryUsage().getUsed();
        }
    });
    heap.gauge("Committed", new Gauge<Long>() {

        @Override
        public Long getValue() {
            return mxBean.getHeapMemoryUsage().getCommitted();
        }
    });
    heap.gauge("Max", new Gauge<Long>() {

        @Override
        public Long getValue() {
            return mxBean.getHeapMemoryUsage().getMax();
        }
    });
    MetricGroup nonHeap = metrics.addGroup("NonHeap");
    nonHeap.gauge("Used", new Gauge<Long>() {

        @Override
        public Long getValue() {
            return mxBean.getNonHeapMemoryUsage().getUsed();
        }
    });
    nonHeap.gauge("Committed", new Gauge<Long>() {

        @Override
        public Long getValue() {
            return mxBean.getNonHeapMemoryUsage().getCommitted();
        }
    });
    nonHeap.gauge("Max", new Gauge<Long>() {

        @Override
        public Long getValue() {
            return mxBean.getNonHeapMemoryUsage().getMax();
        }
    });
    List<BufferPoolMXBean> bufferMxBeans = ManagementFactory.getPlatformMXBeans(BufferPoolMXBean.class);
    for (final BufferPoolMXBean bufferMxBean : bufferMxBeans) {
        MetricGroup bufferGroup = metrics.addGroup(WordUtils.capitalize(bufferMxBean.getName()));
        bufferGroup.gauge("Count", new Gauge<Long>() {

            @Override
            public Long getValue() {
                return bufferMxBean.getCount();
            }
        });
        bufferGroup.gauge("MemoryUsed", new Gauge<Long>() {

            @Override
            public Long getValue() {
                return bufferMxBean.getMemoryUsed();
            }
        });
        bufferGroup.gauge("TotalCapacity", new Gauge<Long>() {

            @Override
            public Long getValue() {
                return bufferMxBean.getTotalCapacity();
            }
        });
    }
}

Example 10 with MetricGroup

use of org.apache.flink.metrics.MetricGroup in project flink by apache.

the class TaskIOMetricGroup method initializeBufferMetrics.

// ============================================================================================
// Buffer metrics
// ============================================================================================
/**
	 * Initialize Buffer Metrics for a task
	 */
public void initializeBufferMetrics(Task task) {
    final MetricGroup buffers = addGroup("buffers");
    buffers.gauge("inputQueueLength", new InputBuffersGauge(task));
    buffers.gauge("outputQueueLength", new OutputBuffersGauge(task));
    buffers.gauge("inPoolUsage", new InputBufferPoolUsageGauge(task));
    buffers.gauge("outPoolUsage", new OutputBufferPoolUsageGauge(task));
}