前往小程序,Get更优阅读体验!
立即前往
首页
学习
活动
专区
工具
TVP
发布
社区首页 >专栏 >聊聊ExecutorService的监控

聊聊ExecutorService的监控

作者头像
code4it
发布2018-10-18 15:15:47
1.2K0
发布2018-10-18 15:15:47
举报
文章被收录于专栏:码匠的流水账码匠的流水账

本文主要研究一下ExecutorService的监控

InstrumentedExecutorService

metrics-core-4.0.2-sources.jar!/com/codahale/metrics/InstrumentedExecutorService.java

代码语言:javascript
复制
/**
 * An {@link ExecutorService} that monitors the number of tasks submitted, running,
 * completed and also keeps a {@link Timer} for the task duration.
 * <p/>
 * It will register the metrics using the given (or auto-generated) name as classifier, e.g:
 * "your-executor-service.submitted", "your-executor-service.running", etc.
 */
public class InstrumentedExecutorService implements ExecutorService {
    private static final AtomicLong NAME_COUNTER = new AtomicLong();

    private final ExecutorService delegate;
    private final Meter submitted;
    private final Counter running;
    private final Meter completed;
    private final Timer idle;
    private final Timer duration;

    /**
     * Wraps an {@link ExecutorService} uses an auto-generated default name.
     *
     * @param delegate {@link ExecutorService} to wrap.
     * @param registry {@link MetricRegistry} that will contain the metrics.
     */
    public InstrumentedExecutorService(ExecutorService delegate, MetricRegistry registry) {
        this(delegate, registry, "instrumented-delegate-" + NAME_COUNTER.incrementAndGet());
    }

    /**
     * Wraps an {@link ExecutorService} with an explicit name.
     *
     * @param delegate {@link ExecutorService} to wrap.
     * @param registry {@link MetricRegistry} that will contain the metrics.
     * @param name     name for this executor service.
     */
    public InstrumentedExecutorService(ExecutorService delegate, MetricRegistry registry, String name) {
        this.delegate = delegate;
        this.submitted = registry.meter(MetricRegistry.name(name, "submitted"));
        this.running = registry.counter(MetricRegistry.name(name, "running"));
        this.completed = registry.meter(MetricRegistry.name(name, "completed"));
        this.idle = registry.timer(MetricRegistry.name(name, "idle"));
        this.duration = registry.timer(MetricRegistry.name(name, "duration"));
    }

    @Override
    public void execute(Runnable runnable) {
        submitted.mark();
        delegate.execute(new InstrumentedRunnable(runnable));
    }

    @Override
    public Future<?> submit(Runnable runnable) {
        submitted.mark();
        return delegate.submit(new InstrumentedRunnable(runnable));
    }

    @Override
    public <T> Future<T> submit(Runnable runnable, T result) {
        submitted.mark();
        return delegate.submit(new InstrumentedRunnable(runnable), result);
    }

    @Override
    public <T> Future<T> submit(Callable<T> task) {
        submitted.mark();
        return delegate.submit(new InstrumentedCallable<>(task));
    }

    @Override
    public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) throws InterruptedException {
        submitted.mark(tasks.size());
        Collection<? extends Callable<T>> instrumented = instrument(tasks);
        return delegate.invokeAll(instrumented);
    }

    @Override
    public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks, long timeout, TimeUnit unit) throws InterruptedException {
        submitted.mark(tasks.size());
        Collection<? extends Callable<T>> instrumented = instrument(tasks);
        return delegate.invokeAll(instrumented, timeout, unit);
    }

    @Override
    public <T> T invokeAny(Collection<? extends Callable<T>> tasks) throws ExecutionException, InterruptedException {
        submitted.mark(tasks.size());
        Collection<? extends Callable<T>> instrumented = instrument(tasks);
        return delegate.invokeAny(instrumented);
    }

    @Override
    public <T> T invokeAny(Collection<? extends Callable<T>> tasks, long timeout, TimeUnit unit) throws ExecutionException, InterruptedException, TimeoutException {
        submitted.mark(tasks.size());
        Collection<? extends Callable<T>> instrumented = instrument(tasks);
        return delegate.invokeAny(instrumented, timeout, unit);
    }

    private <T> Collection<? extends Callable<T>> instrument(Collection<? extends Callable<T>> tasks) {
        final List<InstrumentedCallable<T>> instrumented = new ArrayList<>(tasks.size());
        for (Callable<T> task : tasks) {
            instrumented.add(new InstrumentedCallable<>(task));
        }
        return instrumented;
    }

    @Override
    public void shutdown() {
        delegate.shutdown();
    }

    @Override
    public List<Runnable> shutdownNow() {
        return delegate.shutdownNow();
    }

    @Override
    public boolean isShutdown() {
        return delegate.isShutdown();
    }

    @Override
    public boolean isTerminated() {
        return delegate.isTerminated();
    }

    @Override
    public boolean awaitTermination(long l, TimeUnit timeUnit) throws InterruptedException {
        return delegate.awaitTermination(l, timeUnit);
    }
    //......
}
  • InstrumentedExecutorService实现了ExecutorService,对jdk原始的ExecutorService进行了包装,对相应的方法织入指标统计
  • 主要统计了已提交的任务submitted(Meter),运行中的任务running(Counter),完成的任务completed(Meter),空闲时长idle(Timer),运行时长duration(Timer)
  • 为了统计后面几个指标,需要对Runnable以及Callable进行织入,因而引入了InstrumentedRunnable、InstrumentedCallable

InstrumentedRunnable

代码语言:javascript
复制
    private class InstrumentedRunnable implements Runnable {
        private final Runnable task;
        private final Timer.Context idleContext;

        InstrumentedRunnable(Runnable task) {
            this.task = task;
            this.idleContext = idle.time();
        }

        @Override
        public void run() {
            idleContext.stop();
            running.inc();
            final Timer.Context durationContext = duration.time();
            try {
                task.run();
            } finally {
                durationContext.stop();
                running.dec();
                completed.mark();
            }
        }
    }
  • 织入了对idle、duration、running、completed的统计

InstrumentedCallable

代码语言:javascript
复制
    private class InstrumentedCallable<T> implements Callable<T> {
        private final Callable<T> callable;

        InstrumentedCallable(Callable<T> callable) {
            this.callable = callable;
        }

        @Override
        public T call() throws Exception {
            running.inc();
            final Timer.Context context = duration.time();
            try {
                return callable.call();
            } finally {
                context.stop();
                running.dec();
                completed.mark();
            }
        }
    }
  • 织入了对duration、running、completed的统计

ExecutorServiceMetrics

micrometer-core-1.0.3-sources.jar!/io/micrometer/core/instrument/binder/jvm/ExecutorServiceMetrics.java

代码语言:javascript
复制
/**
 * Monitors the status of executor service pools. Does not record timings on operations executed in the {@link ExecutorService},
 * as this requires the instance to be wrapped. Timings are provided separately by wrapping the executor service
 * with {@link TimedExecutorService}.
 *
 * @author Jon Schneider
 * @author Clint Checketts
 */
@NonNullApi
@NonNullFields
public class ExecutorServiceMetrics implements MeterBinder {
    @Nullable
    private final ExecutorService executorService;

    private final Iterable<Tag> tags;

    public ExecutorServiceMetrics(@Nullable ExecutorService executorService, String executorServiceName, Iterable<Tag> tags) {
        this.executorService = executorService;
        this.tags = Tags.concat(tags, "name", executorServiceName);
    }
    //......
    /**
     * Record metrics on the use of an {@link Executor}.
     *
     * @param registry     The registry to bind metrics to.
     * @param executor     The executor to instrument.
     * @param executorName Will be used to tag metrics with "name".
     * @param tags         Tags to apply to all recorded metrics.
     * @return The instrumented executor, proxied.
     */
    public static Executor monitor(MeterRegistry registry, Executor executor, String executorName, Iterable<Tag> tags) {
        if (executor instanceof ExecutorService) {
            return monitor(registry, (ExecutorService) executor, executorName, tags);
        }
        return new TimedExecutor(registry, executor, executorName, tags);
    }

    /**
     * Record metrics on the use of an {@link ExecutorService}.
     *
     * @param registry            The registry to bind metrics to.
     * @param executor            The executor to instrument.
     * @param executorServiceName Will be used to tag metrics with "name".
     * @param tags                Tags to apply to all recorded metrics.
     * @return The instrumented executor, proxied.
     */
    public static ExecutorService monitor(MeterRegistry registry, ExecutorService executor, String executorServiceName, Iterable<Tag> tags) {
        new ExecutorServiceMetrics(executor, executorServiceName, tags).bindTo(registry);
        return new TimedExecutorService(registry, executor, executorServiceName, tags);
    }

    @Override
    public void bindTo(MeterRegistry registry) {
        if (executorService == null) {
            return;
        }

        String className = executorService.getClass().getName();

        if (executorService instanceof ThreadPoolExecutor) {
            monitor(registry, (ThreadPoolExecutor) executorService);
        } else if (className.equals("java.util.concurrent.Executors$DelegatedScheduledExecutorService")) {
            monitor(registry, unwrapThreadPoolExecutor(executorService, executorService.getClass()));
        } else if (className.equals("java.util.concurrent.Executors$FinalizableDelegatedExecutorService")) {
            monitor(registry, unwrapThreadPoolExecutor(executorService, executorService.getClass().getSuperclass()));
        } else if (executorService instanceof ForkJoinPool) {
            monitor(registry, (ForkJoinPool) executorService);
        }
    }

    private void monitor(MeterRegistry registry, @Nullable ThreadPoolExecutor tp) {
        if (tp == null) {
            return;
        }

        FunctionCounter.builder("executor.completed", tp, ThreadPoolExecutor::getCompletedTaskCount)
                .tags(tags)
                .description("The approximate total number of tasks that have completed execution")
                .baseUnit("tasks")
                .register(registry);

        Gauge.builder("executor.active", tp, ThreadPoolExecutor::getActiveCount)
                .tags(tags)
                .description("The approximate number of threads that are actively executing tasks")
                .baseUnit("threads")
                .register(registry);

        Gauge.builder("executor.queued", tp, tpRef -> tpRef.getQueue().size())
                .tags(tags)
                .description("The approximate number of threads that are queued for execution")
                .baseUnit("threads")
                .register(registry);

        Gauge.builder("executor.pool.size", tp, ThreadPoolExecutor::getPoolSize)
                .tags(tags)
                .description("The current number of threads in the pool")
                .baseUnit("threads")
                .register(registry);
    }
    //......
}
  • ExecutorServiceMetrics实现了MeterBinder接口,另外提供了静态方法来创建带有监控指标的ExecutorService,该静态方法命名为monitor,非常形象
  • monitor方法首先创建ExecutorServiceMetrics,并bindTo了MeterRegistry,然后返回TimedExecutorService
  • bindTo方法上报了executor.completed(FunctionCounter),executor.active(Gauge),executor.queued(Gauge),executor.pool.size(Gauge)这几个指标

TimedExecutorService

micrometer-core-1.0.3-sources.jar!/io/micrometer/core/instrument/internal/TimedExecutorService.java

代码语言:javascript
复制
/**
 * An {@link java.util.concurrent.ExecutorService} that is timed
 *
 * @author Jon Schneider
 */
public class TimedExecutorService implements ExecutorService {
    private final ExecutorService delegate;
    private final Timer timer;

    public TimedExecutorService(MeterRegistry registry, ExecutorService delegate, String executorServiceName, Iterable<Tag> tags) {
        this.delegate = delegate;
        this.timer = registry.timer("executor", Tags.concat(tags ,"name", executorServiceName));
    }

    @Override
    public void shutdown() {
        delegate.shutdown();
    }

    @Override
    public List<Runnable> shutdownNow() {
        return delegate.shutdownNow();
    }

    @Override
    public boolean isShutdown() {
        return delegate.isShutdown();
    }

    @Override
    public boolean isTerminated() {
        return delegate.isTerminated();
    }

    @Override
    public boolean awaitTermination(long timeout, TimeUnit unit) throws InterruptedException {
        return delegate.awaitTermination(timeout, unit);
    }

    @Override
    public <T> Future<T> submit(Callable<T> task) {
        return delegate.submit(timer.wrap(task));
    }

    @Override
    public <T> Future<T> submit(Runnable task, T result) {
        return delegate.submit(() -> timer.record(task), result);
    }

    @Override
    public Future<?> submit(Runnable task) {
        return delegate.submit(() -> timer.record(task));
    }

    @Override
    public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks) throws InterruptedException {
        return delegate.invokeAll(wrapAll(tasks));
    }

    @Override
    public <T> List<Future<T>> invokeAll(Collection<? extends Callable<T>> tasks, long timeout, TimeUnit unit) throws InterruptedException {
        return delegate.invokeAll(wrapAll(tasks), timeout, unit);
    }

    @Override
    public <T> T invokeAny(Collection<? extends Callable<T>> tasks) throws InterruptedException, ExecutionException {
        return delegate.invokeAny(wrapAll(tasks));
    }

    @Override
    public <T> T invokeAny(Collection<? extends Callable<T>> tasks, long timeout, TimeUnit unit) throws InterruptedException, ExecutionException, TimeoutException {
        return delegate.invokeAny(wrapAll(tasks), timeout, unit);
    }

    @Override
    public void execute(Runnable command) {
        delegate.execute(timer.wrap(command));
    }

    private <T> Collection<? extends Callable<T>> wrapAll(Collection<? extends Callable<T>> tasks) {
        return tasks.stream().map(timer::wrap).collect(toList());
    }
}
  • 对ExecutorService进行包装,增加了

Timer.record

micrometer-core-1.0.3-sources.jar!/io/micrometer/core/instrument/Timer.java

代码语言:javascript
复制
    /**
     * Executes the runnable `f` and records the time taken.
     *
     * @param f Function to execute and measure the execution time.
     */
    void record(Runnable f);

    /**
     * Wrap a {@link Runnable} so that it is timed when invoked.
     *
     * @param f The Runnable to time when it is invoked.
     * @return The wrapped Runnable.
     */
    default Runnable wrap(Runnable f) {
        return () -> record(f);
    }

    /**
     * Wrap a {@link Callable} so that it is timed when invoked.
     *
     * @param f   The Callable to time when it is invoked.
     * @param <T> The return type of the callable.
     * @return The wrapped callable.
     */
    default <T> Callable<T> wrap(Callable<T> f) {
        return () -> recordCallable(f);
    }
  • warp方法主要是包装调用record方法,而record由实现类去实现

AbstractTimer

micrometer-core-1.0.3-sources.jar!/io/micrometer/core/instrument/AbstractTimer.java

代码语言:javascript
复制
    @Override
    public void record(Runnable f) {
        final long s = clock.monotonicTime();
        try {
            f.run();
        } finally {
            final long e = clock.monotonicTime();
            record(e - s, TimeUnit.NANOSECONDS);
        }
    }

    @Override
    public final void record(long amount, TimeUnit unit) {
        if (amount >= 0) {
            histogram.recordLong(TimeUnit.NANOSECONDS.convert(amount, unit));
            recordNonNegative(amount, unit);

            if (intervalEstimator != null) {
                intervalEstimator.recordInterval(clock.monotonicTime());
            }
        }
    }
  • record采用histogram进行统计

小结

dropwizard及micrometer均提供了对ExecutorService的指标统计的包装,micrometer则更近一步提供了静态方法来直接创建,非常方便。

doc

  • InstrumentedExecutorService
  • ExecutorServiceMetrics
本文参与 腾讯云自媒体分享计划,分享自微信公众号。
原始发表:2018-10-07,如有侵权请联系 cloudcommunity@tencent.com 删除

本文分享自 码匠的流水账 微信公众号,前往查看

如有侵权,请联系 cloudcommunity@tencent.com 删除。

本文参与 腾讯云自媒体分享计划  ,欢迎热爱写作的你一起参与!

评论
登录后参与评论
0 条评论
热度
最新
推荐阅读
目录
  • InstrumentedExecutorService
    • InstrumentedRunnable
      • InstrumentedCallable
      • ExecutorServiceMetrics
        • TimedExecutorService
          • Timer.record
            • AbstractTimer
            • 小结
            • doc
            领券
            问题归档专栏文章快讯文章归档关键词归档开发者手册归档开发者手册 Section 归档