@Test
@InSequence(1)
public void countedMethodNotCalledYet() {
assertThat("Counter is not registered correctly", registry.getCounters(), hasKey(COUNTER_NAME));
Counter counter = registry.getCounters().get(COUNTER_NAME);
// Make sure that the counter hasn't been called yet
assertThat("Counter count is incorrect", counter.getCount(), is(equalTo(COUNTER_COUNT.get())));
}
@Test
@InSequence(2)
public void countedMethodNotCalledYet(
@Metric(name = "countedMethod", absolute = true) Counter instance) {
assertThat("Counter is not registered correctly", registry.getCounters(), hasKey(COUNTER_NAME));
Counter counter = registry.getCounters().get(COUNTER_NAME);
// Make sure that the counter registered and the bean instance are the same
assertThat("Counter and bean instance are not equal", instance, is(equalTo(counter)));
}
/**
* @summary get the worker metrics
* @return the response object
*/
@GET
@Path(GET_METRICS)
@ReturnType("java.util.SortedMap<String, Long>")
public Response getMetrics() {
MetricRegistry metricRegistry = mWorker.getWorkerMetricsSystem().getMetricRegistry();
// Get all counters.
Map<String, Counter> counters = metricRegistry.getCounters();
// Only the gauge for cached blocks is retrieved here, other gauges are statistics of free/used
// spaces, those statistics can be gotten via other REST apis.
String blocksCachedProperty =
CommonUtils.argsToString(
".", WorkerContext.getWorkerSource().getName(), WorkerSource.BLOCKS_CACHED);
@SuppressWarnings("unchecked")
Gauge<Integer> blocksCached =
(Gauge<Integer>) metricRegistry.getGauges().get(blocksCachedProperty);
// Get values of the counters and gauges and put them into a metrics map.
SortedMap<String, Long> metrics = new TreeMap<>();
for (Map.Entry<String, Counter> counter : counters.entrySet()) {
metrics.put(counter.getKey(), counter.getValue().getCount());
}
metrics.put(blocksCachedProperty, blocksCached.getValue().longValue());
return RestUtils.createResponse(metrics);
}
private void updateQueueMetrics(String queue, int releasedMemory, int releasedVCores) {
// update queue counters
SortedMap<String, Counter> counterMap = metrics.getCounters();
if (releasedMemory != 0) {
String name = "counter.queue." + queue + ".allocated.memory";
if (!counterMap.containsKey(name)) {
metrics.counter(name);
counterMap = metrics.getCounters();
}
counterMap.get(name).inc(-releasedMemory);
}
if (releasedVCores != 0) {
String name = "counter.queue." + queue + ".allocated.cores";
if (!counterMap.containsKey(name)) {
metrics.counter(name);
counterMap = metrics.getCounters();
}
counterMap.get(name).inc(-releasedVCores);
}
}
@Test
@InSequence(4)
public void removeCounterFromRegistry() {
assertThat("Counter is not registered correctly", registry.getCounters(), hasKey(COUNTER_NAME));
Counter counter = registry.getCounters().get(COUNTER_NAME);
// Remove the counter from metrics registry
registry.remove(COUNTER_NAME);
try {
// Call the counted method and assert an exception is thrown
bean.countedMethod(
new Callable<Long>() {
@Override
public Long call() throws Exception {
return null;
}
});
} catch (Exception cause) {
assertThat(cause, is(instanceOf(IllegalStateException.class)));
assertThat(
cause.getMessage(),
is(
equalTo(
"No counter with name ["
+ COUNTER_NAME
+ "] found in registry ["
+ registry
+ "]")));
// Make sure that the counter hasn't been called
assertThat(
"Counter count is incorrect", counter.getCount(), is(equalTo(COUNTER_COUNT.get())));
return;
}
fail("No exception has been re-thrown!");
}
/** Resets all the counters to 0 for testing. */
public static void resetAllCounters() {
for (Map.Entry<String, Counter> entry : METRIC_REGISTRY.getCounters().entrySet()) {
entry.getValue().dec(entry.getValue().getCount());
}
}
@Test
@InSequence(3)
public void callCountedMethodOnce() throws InterruptedException, TimeoutException {
assertThat("Counter is not registered correctly", registry.getCounters(), hasKey(COUNTER_NAME));
Counter counter = registry.getCounters().get(COUNTER_NAME);
// Call the counted method, block and assert it's been counted
final Exchanger<Long> exchanger = new Exchanger<>();
Thread thread =
new Thread(
new Runnable() {
@Override
public void run() {
try {
exchanger.exchange(
bean.countedMethod(
new Callable<Long>() {
@Override
public Long call() throws Exception {
exchanger.exchange(0L);
return exchanger.exchange(0L);
}
}));
} catch (InterruptedException cause) {
throw new RuntimeException(cause);
}
}
});
final AtomicInteger uncaught = new AtomicInteger();
thread.setUncaughtExceptionHandler(
new Thread.UncaughtExceptionHandler() {
@Override
public void uncaughtException(Thread t, Throwable e) {
uncaught.incrementAndGet();
}
});
thread.start();
// Wait until the method is executing and make sure that the counter has been incremented
exchanger.exchange(0L, 5L, TimeUnit.SECONDS);
assertThat(
"Counter count is incorrect",
counter.getCount(),
is(equalTo(COUNTER_COUNT.incrementAndGet())));
// Exchange the result and unblock the method execution
Long random = 1 + Math.round(Math.random() * (Long.MAX_VALUE - 1));
exchanger.exchange(random, 5L, TimeUnit.SECONDS);
// Wait until the method has returned
assertThat(
"Counted method return value is incorrect", exchanger.exchange(0L), is(equalTo(random)));
// Then make sure that the counter has been decremented
assertThat(
"Counter count is incorrect",
counter.getCount(),
is(equalTo(COUNTER_COUNT.decrementAndGet())));
// Finally make sure calling thread is returns correctly
thread.join();
assertThat("Exception thrown in method call thread", uncaught.get(), is(equalTo(0)));
}
public String generateRealTimeTrackingMetrics() {
// JVM
double jvmFreeMemoryGB, jvmMaxMemoryGB, jvmTotalMemoryGB;
if (jvmFreeMemoryGauge == null && metrics.getGauges().containsKey("variable.jvm.free.memory")) {
jvmFreeMemoryGauge = metrics.getGauges().get("variable.jvm.free.memory");
}
if (jvmMaxMemoryGauge == null && metrics.getGauges().containsKey("variable.jvm.max.memory")) {
jvmMaxMemoryGauge = metrics.getGauges().get("variable.jvm.max.memory");
}
if (jvmTotalMemoryGauge == null
&& metrics.getGauges().containsKey("variable.jvm.total.memory")) {
jvmTotalMemoryGauge = metrics.getGauges().get("variable.jvm.total.memory");
}
jvmFreeMemoryGB =
jvmFreeMemoryGauge == null
? 0
: Double.parseDouble(jvmFreeMemoryGauge.getValue().toString()) / 1024 / 1024 / 1024;
jvmMaxMemoryGB =
jvmMaxMemoryGauge == null
? 0
: Double.parseDouble(jvmMaxMemoryGauge.getValue().toString()) / 1024 / 1024 / 1024;
jvmTotalMemoryGB =
jvmTotalMemoryGauge == null
? 0
: Double.parseDouble(jvmTotalMemoryGauge.getValue().toString()) / 1024 / 1024 / 1024;
// number of running applications/containers
String numRunningApps, numRunningContainers;
if (numRunningAppsGauge == null
&& metrics.getGauges().containsKey("variable.running.application")) {
numRunningAppsGauge = metrics.getGauges().get("variable.running.application");
}
if (numRunningContainersGauge == null
&& metrics.getGauges().containsKey("variable.running.container")) {
numRunningContainersGauge = metrics.getGauges().get("variable.running.container");
}
numRunningApps = numRunningAppsGauge == null ? "0" : numRunningAppsGauge.getValue().toString();
numRunningContainers =
numRunningContainersGauge == null ? "0" : numRunningContainersGauge.getValue().toString();
// cluster available/allocate resource
double allocatedMemoryGB, allocatedVCoresGB, availableMemoryGB, availableVCoresGB;
if (allocatedMemoryGauge == null
&& metrics.getGauges().containsKey("variable.cluster.allocated.memory")) {
allocatedMemoryGauge = metrics.getGauges().get("variable.cluster.allocated.memory");
}
if (allocatedVCoresGauge == null
&& metrics.getGauges().containsKey("variable.cluster.allocated.vcores")) {
allocatedVCoresGauge = metrics.getGauges().get("variable.cluster.allocated.vcores");
}
if (availableMemoryGauge == null
&& metrics.getGauges().containsKey("variable.cluster.available.memory")) {
availableMemoryGauge = metrics.getGauges().get("variable.cluster.available.memory");
}
if (availableVCoresGauge == null
&& metrics.getGauges().containsKey("variable.cluster.available.vcores")) {
availableVCoresGauge = metrics.getGauges().get("variable.cluster.available.vcores");
}
allocatedMemoryGB =
allocatedMemoryGauge == null
? 0
: Double.parseDouble(allocatedMemoryGauge.getValue().toString()) / 1024;
allocatedVCoresGB =
allocatedVCoresGauge == null
? 0
: Double.parseDouble(allocatedVCoresGauge.getValue().toString());
availableMemoryGB =
availableMemoryGauge == null
? 0
: Double.parseDouble(availableMemoryGauge.getValue().toString()) / 1024;
availableVCoresGB =
availableVCoresGauge == null
? 0
: Double.parseDouble(availableVCoresGauge.getValue().toString());
// scheduler operation
double allocateTimecost, handleTimecost;
if (allocateTimecostHistogram == null
&& metrics.getHistograms().containsKey("sampler.scheduler.operation.allocate.timecost")) {
allocateTimecostHistogram =
metrics.getHistograms().get("sampler.scheduler.operation.allocate.timecost");
}
if (handleTimecostHistogram == null
&& metrics.getHistograms().containsKey("sampler.scheduler.operation.handle.timecost")) {
handleTimecostHistogram =
metrics.getHistograms().get("sampler.scheduler.operation.handle.timecost");
}
allocateTimecost =
allocateTimecostHistogram == null
? 0.0
: allocateTimecostHistogram.getSnapshot().getMean() / 1000000;
handleTimecost =
handleTimecostHistogram == null
? 0.0
: handleTimecostHistogram.getSnapshot().getMean() / 1000000;
// various handle operation
Map<SchedulerEventType, Double> handleOperTimecostMap =
new HashMap<SchedulerEventType, Double>();
for (SchedulerEventType e : SchedulerEventType.values()) {
String key = "sampler.scheduler.operation.handle." + e + ".timecost";
if (!handleOperTimecostHistogramMap.containsKey(e)
&& metrics.getHistograms().containsKey(key)) {
handleOperTimecostHistogramMap.put(e, metrics.getHistograms().get(key));
}
double timecost =
handleOperTimecostHistogramMap.containsKey(e)
? handleOperTimecostHistogramMap.get(e).getSnapshot().getMean() / 1000000
: 0;
handleOperTimecostMap.put(e, timecost);
}
// allocated resource for each queue
Map<String, Double> queueAllocatedMemoryMap = new HashMap<String, Double>();
Map<String, Long> queueAllocatedVCoresMap = new HashMap<String, Long>();
for (String queue : wrapper.getQueueSet()) {
// memory
String key = "counter.queue." + queue + ".allocated.memory";
if (!queueAllocatedMemoryCounterMap.containsKey(queue)
&& metrics.getCounters().containsKey(key)) {
queueAllocatedMemoryCounterMap.put(queue, metrics.getCounters().get(key));
}
double queueAllocatedMemoryGB =
queueAllocatedMemoryCounterMap.containsKey(queue)
? queueAllocatedMemoryCounterMap.get(queue).getCount() / 1024.0
: 0;
queueAllocatedMemoryMap.put(queue, queueAllocatedMemoryGB);
// vCores
key = "counter.queue." + queue + ".allocated.cores";
if (!queueAllocatedVCoresCounterMap.containsKey(queue)
&& metrics.getCounters().containsKey(key)) {
queueAllocatedVCoresCounterMap.put(queue, metrics.getCounters().get(key));
}
long queueAllocatedVCores =
queueAllocatedVCoresCounterMap.containsKey(queue)
? queueAllocatedVCoresCounterMap.get(queue).getCount()
: 0;
queueAllocatedVCoresMap.put(queue, queueAllocatedVCores);
}
// package results
StringBuilder sb = new StringBuilder();
sb.append("{");
sb.append("\"time\":")
.append(System.currentTimeMillis())
.append(",\"jvm.free.memory\":")
.append(jvmFreeMemoryGB)
.append(",\"jvm.max.memory\":")
.append(jvmMaxMemoryGB)
.append(",\"jvm.total.memory\":")
.append(jvmTotalMemoryGB)
.append(",\"running.applications\":")
.append(numRunningApps)
.append(",\"running.containers\":")
.append(numRunningContainers)
.append(",\"cluster.allocated.memory\":")
.append(allocatedMemoryGB)
.append(",\"cluster.allocated.vcores\":")
.append(allocatedVCoresGB)
.append(",\"cluster.available.memory\":")
.append(availableMemoryGB)
.append(",\"cluster.available.vcores\":")
.append(availableVCoresGB);
for (String queue : wrapper.getQueueSet()) {
sb.append(",\"queue.")
.append(queue)
.append(".allocated.memory\":")
.append(queueAllocatedMemoryMap.get(queue));
sb.append(",\"queue.")
.append(queue)
.append(".allocated.vcores\":")
.append(queueAllocatedVCoresMap.get(queue));
}
// scheduler allocate & handle
sb.append(",\"scheduler.allocate.timecost\":").append(allocateTimecost);
sb.append(",\"scheduler.handle.timecost\":").append(handleTimecost);
for (SchedulerEventType e : SchedulerEventType.values()) {
sb.append(",\"scheduler.handle-")
.append(e)
.append(".timecost\":")
.append(handleOperTimecostMap.get(e));
}
sb.append("}");
return sb.toString();
}
private void updateQueueWithAllocateRequest(
Allocation allocation,
ApplicationAttemptId attemptId,
List<ResourceRequest> resourceRequests,
List<ContainerId> containerIds)
throws IOException {
// update queue information
Resource pendingResource = Resources.createResource(0, 0);
Resource allocatedResource = Resources.createResource(0, 0);
String queueName = appQueueMap.get(attemptId.getApplicationId());
// container requested
for (ResourceRequest request : resourceRequests) {
if (request.getResourceName().equals(ResourceRequest.ANY)) {
Resources.addTo(
pendingResource,
Resources.multiply(request.getCapability(), request.getNumContainers()));
}
}
// container allocated
for (Container container : allocation.getContainers()) {
Resources.addTo(allocatedResource, container.getResource());
Resources.subtractFrom(pendingResource, container.getResource());
}
// container released from AM
SchedulerAppReport report = scheduler.getSchedulerAppInfo(attemptId);
for (ContainerId containerId : containerIds) {
Container container = null;
for (RMContainer c : report.getLiveContainers()) {
if (c.getContainerId().equals(containerId)) {
container = c.getContainer();
break;
}
}
if (container != null) {
// released allocated containers
Resources.subtractFrom(allocatedResource, container.getResource());
} else {
for (RMContainer c : report.getReservedContainers()) {
if (c.getContainerId().equals(containerId)) {
container = c.getContainer();
break;
}
}
if (container != null) {
// released reserved containers
Resources.subtractFrom(pendingResource, container.getResource());
}
}
}
// containers released/preemption from scheduler
Set<ContainerId> preemptionContainers = new HashSet<ContainerId>();
if (allocation.getContainerPreemptions() != null) {
preemptionContainers.addAll(allocation.getContainerPreemptions());
}
if (allocation.getStrictContainerPreemptions() != null) {
preemptionContainers.addAll(allocation.getStrictContainerPreemptions());
}
if (!preemptionContainers.isEmpty()) {
for (ContainerId containerId : preemptionContainers) {
if (!preemptionContainerMap.containsKey(containerId)) {
Container container = null;
for (RMContainer c : report.getLiveContainers()) {
if (c.getContainerId().equals(containerId)) {
container = c.getContainer();
break;
}
}
if (container != null) {
preemptionContainerMap.put(containerId, container.getResource());
}
}
}
}
// update metrics
SortedMap<String, Counter> counterMap = metrics.getCounters();
String names[] =
new String[] {
"counter.queue." + queueName + ".pending.memory",
"counter.queue." + queueName + ".pending.cores",
"counter.queue." + queueName + ".allocated.memory",
"counter.queue." + queueName + ".allocated.cores"
};
int values[] =
new int[] {
pendingResource.getMemory(),
pendingResource.getVirtualCores(),
allocatedResource.getMemory(),
allocatedResource.getVirtualCores()
};
for (int i = names.length - 1; i >= 0; i--) {
if (!counterMap.containsKey(names[i])) {
metrics.counter(names[i]);
counterMap = metrics.getCounters();
}
counterMap.get(names[i]).inc(values[i]);
}
queueLock.lock();
try {
if (!schedulerMetrics.isTracked(queueName)) {
schedulerMetrics.trackQueue(queueName);
}
} finally {
queueLock.unlock();
}
}