use of org.apache.flink.metrics.Gauge in project flink by apache.
the class ScheduledDropwizardReporterTest method testMetricCleanup.
/**
* This test verifies that metrics are properly added and removed to/from the ScheduledDropwizardReporter and
* the underlying Dropwizard MetricRegistry.
*/
@Test
public void testMetricCleanup() {
TestingScheduledDropwizardReporter rep = new TestingScheduledDropwizardReporter();
MetricGroup mp = new UnregisteredMetricsGroup();
Counter c = new SimpleCounter();
Meter m = new Meter() {
@Override
public void markEvent() {
}
@Override
public void markEvent(long n) {
}
@Override
public double getRate() {
return 0;
}
@Override
public long getCount() {
return 0;
}
};
Histogram h = new Histogram() {
@Override
public void update(long value) {
}
@Override
public long getCount() {
return 0;
}
@Override
public HistogramStatistics getStatistics() {
return null;
}
};
Gauge g = new Gauge() {
@Override
public Object getValue() {
return null;
}
};
rep.notifyOfAddedMetric(c, "counter", mp);
assertEquals(1, rep.getCounters().size());
assertEquals(1, rep.registry.getCounters().size());
rep.notifyOfAddedMetric(m, "meter", mp);
assertEquals(1, rep.getMeters().size());
assertEquals(1, rep.registry.getMeters().size());
rep.notifyOfAddedMetric(h, "histogram", mp);
assertEquals(1, rep.getHistograms().size());
assertEquals(1, rep.registry.getHistograms().size());
rep.notifyOfAddedMetric(g, "gauge", mp);
assertEquals(1, rep.getGauges().size());
assertEquals(1, rep.registry.getGauges().size());
rep.notifyOfRemovedMetric(c, "counter", mp);
assertEquals(0, rep.getCounters().size());
assertEquals(0, rep.registry.getCounters().size());
rep.notifyOfRemovedMetric(m, "meter", mp);
assertEquals(0, rep.getMeters().size());
assertEquals(0, rep.registry.getMeters().size());
rep.notifyOfRemovedMetric(h, "histogram", mp);
assertEquals(0, rep.getHistograms().size());
assertEquals(0, rep.registry.getHistograms().size());
rep.notifyOfRemovedMetric(g, "gauge", mp);
assertEquals(0, rep.getGauges().size());
assertEquals(0, rep.registry.getGauges().size());
}
use of org.apache.flink.metrics.Gauge in project flink by apache.
the class TaskExecutorMetricsInitializer 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.<Long, Gauge<Long>>gauge("Count", new Gauge<Long>() {
@Override
public Long getValue() {
return garbageCollector.getCollectionCount();
}
});
gcGroup.<Long, Gauge<Long>>gauge("Time", new Gauge<Long>() {
@Override
public Long getValue() {
return garbageCollector.getCollectionTime();
}
});
}
}
use of org.apache.flink.metrics.Gauge in project flink by apache.
the class TaskExecutorMetricsInitializer method instantiateThreadMetrics.
private static void instantiateThreadMetrics(MetricGroup metrics) {
final ThreadMXBean mxBean = ManagementFactory.getThreadMXBean();
metrics.<Integer, Gauge<Integer>>gauge("Count", new Gauge<Integer>() {
@Override
public Integer getValue() {
return mxBean.getThreadCount();
}
});
}
use of org.apache.flink.metrics.Gauge in project flink by apache.
the class MetricDumpSerializerTest method testSerialization.
@Test
public void testSerialization() throws IOException {
MetricDumpSerialization.MetricDumpSerializer serializer = new MetricDumpSerialization.MetricDumpSerializer();
MetricDumpSerialization.MetricDumpDeserializer deserializer = new MetricDumpSerialization.MetricDumpDeserializer();
Map<Counter, Tuple2<QueryScopeInfo, String>> counters = new HashMap<>();
Map<Gauge<?>, Tuple2<QueryScopeInfo, String>> gauges = new HashMap<>();
Map<Histogram, Tuple2<QueryScopeInfo, String>> histograms = new HashMap<>();
Map<Meter, Tuple2<QueryScopeInfo, String>> meters = new HashMap<>();
SimpleCounter c1 = new SimpleCounter();
SimpleCounter c2 = new SimpleCounter();
c1.inc(1);
c2.inc(2);
Gauge<Integer> g1 = new Gauge<Integer>() {
@Override
public Integer getValue() {
return 4;
}
};
Histogram h1 = new TestingHistogram();
Meter m1 = new Meter() {
@Override
public void markEvent() {
}
@Override
public void markEvent(long n) {
}
@Override
public double getRate() {
return 5;
}
@Override
public long getCount() {
return 10;
}
};
counters.put(c1, new Tuple2<QueryScopeInfo, String>(new QueryScopeInfo.JobManagerQueryScopeInfo("A"), "c1"));
counters.put(c2, new Tuple2<QueryScopeInfo, String>(new QueryScopeInfo.TaskManagerQueryScopeInfo("tmid", "B"), "c2"));
meters.put(m1, new Tuple2<QueryScopeInfo, String>(new QueryScopeInfo.JobQueryScopeInfo("jid", "C"), "c3"));
gauges.put(g1, new Tuple2<QueryScopeInfo, String>(new QueryScopeInfo.TaskQueryScopeInfo("jid", "vid", 2, "D"), "g1"));
histograms.put(h1, new Tuple2<QueryScopeInfo, String>(new QueryScopeInfo.OperatorQueryScopeInfo("jid", "vid", 2, "opname", "E"), "h1"));
MetricDumpSerialization.MetricSerializationResult serialized = serializer.serialize(counters, gauges, histograms, meters);
List<MetricDump> deserialized = deserializer.deserialize(serialized);
// ===== Counters ==============================================================================================
assertEquals(5, deserialized.size());
for (MetricDump metric : deserialized) {
switch(metric.getCategory()) {
case METRIC_CATEGORY_COUNTER:
MetricDump.CounterDump counterDump = (MetricDump.CounterDump) metric;
switch((byte) counterDump.count) {
case 1:
assertTrue(counterDump.scopeInfo instanceof QueryScopeInfo.JobManagerQueryScopeInfo);
assertEquals("A", counterDump.scopeInfo.scope);
assertEquals("c1", counterDump.name);
counters.remove(c1);
break;
case 2:
assertTrue(counterDump.scopeInfo instanceof QueryScopeInfo.TaskManagerQueryScopeInfo);
assertEquals("B", counterDump.scopeInfo.scope);
assertEquals("c2", counterDump.name);
assertEquals("tmid", ((QueryScopeInfo.TaskManagerQueryScopeInfo) counterDump.scopeInfo).taskManagerID);
counters.remove(c2);
break;
default:
fail();
}
break;
case METRIC_CATEGORY_GAUGE:
MetricDump.GaugeDump gaugeDump = (MetricDump.GaugeDump) metric;
assertEquals("4", gaugeDump.value);
assertEquals("g1", gaugeDump.name);
assertTrue(gaugeDump.scopeInfo instanceof QueryScopeInfo.TaskQueryScopeInfo);
QueryScopeInfo.TaskQueryScopeInfo taskInfo = (QueryScopeInfo.TaskQueryScopeInfo) gaugeDump.scopeInfo;
assertEquals("D", taskInfo.scope);
assertEquals("jid", taskInfo.jobID);
assertEquals("vid", taskInfo.vertexID);
assertEquals(2, taskInfo.subtaskIndex);
gauges.remove(g1);
break;
case METRIC_CATEGORY_HISTOGRAM:
MetricDump.HistogramDump histogramDump = (MetricDump.HistogramDump) metric;
assertEquals("h1", histogramDump.name);
assertEquals(0.5, histogramDump.median, 0.1);
assertEquals(0.75, histogramDump.p75, 0.1);
assertEquals(0.90, histogramDump.p90, 0.1);
assertEquals(0.95, histogramDump.p95, 0.1);
assertEquals(0.98, histogramDump.p98, 0.1);
assertEquals(0.99, histogramDump.p99, 0.1);
assertEquals(0.999, histogramDump.p999, 0.1);
assertEquals(4, histogramDump.mean, 0.1);
assertEquals(5, histogramDump.stddev, 0.1);
assertEquals(6, histogramDump.max);
assertEquals(7, histogramDump.min);
assertTrue(histogramDump.scopeInfo instanceof QueryScopeInfo.OperatorQueryScopeInfo);
QueryScopeInfo.OperatorQueryScopeInfo opInfo = (QueryScopeInfo.OperatorQueryScopeInfo) histogramDump.scopeInfo;
assertEquals("E", opInfo.scope);
assertEquals("jid", opInfo.jobID);
assertEquals("vid", opInfo.vertexID);
assertEquals(2, opInfo.subtaskIndex);
assertEquals("opname", opInfo.operatorName);
histograms.remove(h1);
break;
case METRIC_CATEGORY_METER:
MetricDump.MeterDump meterDump = (MetricDump.MeterDump) metric;
assertEquals(5.0, meterDump.rate, 0.1);
assertTrue(meterDump.scopeInfo instanceof QueryScopeInfo.JobQueryScopeInfo);
assertEquals("C", meterDump.scopeInfo.scope);
assertEquals("c3", meterDump.name);
assertEquals("jid", ((QueryScopeInfo.JobQueryScopeInfo) meterDump.scopeInfo).jobID);
break;
default:
fail();
}
}
assertTrue(counters.isEmpty());
assertTrue(gauges.isEmpty());
assertTrue(histograms.isEmpty());
}
use of org.apache.flink.metrics.Gauge 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();
}
}
Aggregations