While I was reviewing #7083, I saw that if I use ExecutorServiceMetrics.monitor for a ThreadPoolExecutor (Executors.newFixedThreadPool) and a ForkJoinPool I get a few unexpected differences in the outputs.
executor.active and executor.queued have units (threads and tasks) for ThreadPoolExecutor while for ForkJoinPool they do not, this means that changing this:
- ExecutorServiceMetrics.monitor(registry, Executors.newFixedThreadPool(2), name, tags);
+ ExecutorServiceMetrics.monitor(registry, new ForkJoinPool(2), name, tags);
will result in different time series (executor_active_threads vs. executor_active) in the Prometheus output:
- executor_active_threads{name="test"} 0.0
+ executor_active{name="test"} 0.0
- executor_queued_tasks{name="test"} 0.0
+ executor_queued{name="test"} 0.0(Fixing this is a breaking change.)
Click to reveal demo code
public class Demo {
public static void main(String[] args) {
generateAndPrintMetricsFor(Executors.newFixedThreadPool(2));
generateAndPrintMetricsFor(new ForkJoinPool(2));
}
private static void generateAndPrintMetricsFor(ExecutorService executorServiceToMonitor) {
PrometheusMeterRegistry registry = new PrometheusMeterRegistry(PrometheusConfig.DEFAULT);
ExecutorService executorService = ExecutorServiceMetrics.monitor(registry, executorServiceToMonitor, "test", Tags.empty());
List<Future<?>> futures = new ArrayList<>();
for (int i = 0; i < 5; i++) {
futures.add(executorService.submit(Demo::sleep));
}
for (Future<?> future : futures) {
try {
future.get(1_000, TimeUnit.MILLISECONDS);
}
catch (Exception e) {
System.out.println(e.getMessage());
}
finally {
future.cancel(true);
}
}
System.out.println("===== Metrics for " + executorService.getClass().getSimpleName());
System.out.println(registry.scrape());
executorService.shutdown();
}
private static void sleep() {
try {
Thread.sleep(100);
}
catch (InterruptedException e) {
// noop
}
}
} 
While I was reviewing #7083, I saw that if I use
ExecutorServiceMetrics.monitorfor aThreadPoolExecutor(Executors.newFixedThreadPool) and aForkJoinPoolI get a few unexpected differences in the outputs.executor.activeandexecutor.queuedhave units (threadsandtasks) forThreadPoolExecutorwhile forForkJoinPoolthey do not, this means that changing this:will result in different time series (
executor_active_threadsvs.executor_active) in the Prometheus output:(Fixing this is a breaking change.)
Click to reveal demo code