如何创建异步堆栈跟踪? [英] How to create async stacktraces?

查看:168
本文介绍了如何创建异步堆栈跟踪?的处理方法,对大家解决问题具有一定的参考价值,需要的朋友们下面随着小编来一起学习吧!

问题描述

更新:最新版本的Intellij IDEA 实现正是我正在寻找的东西。问题是如何在IDE之外实现这一点(所以我可以将异步堆栈跟踪转储到日志文件),理想情况下不使用检测代理。

UPDATE: The latest version of Intellij IDEA implements exactly what I'm looking for. The question is how to implement this outside of the IDE (so I can to dump async stack traces to log files), ideally without the use of an instrumenting agent.

自从我将应用程序从同步模型转换为异步模型后,我遇到了调试失败的问题。

Ever since I converted my application from a synchronous to asynchronous model I am having problems debugging failures.

当我使用同步API时,我总是在异常堆栈跟踪中找到我的类,所以我知道从哪里开始查找是否出错。使用异步API,我得到的堆栈跟踪不会引用我的类,也不会指示哪个请求触发了失败。

When I use synchronous APIs, I always find my classes in exception stacktraces so I know where to begin looking if something goes wrong. With asynchronous APIs, I am getting stacktraces that do not reference my classes nor indicate what request triggered the failure.

我会给你一个具体的例子,但我是感兴趣的是这种问题的一般解决方案。

I'll give you a concrete example, but I'm interested in a general solution to this kind of problem.

我使用<发出HTTP请求a href =https://jersey.github.io/ =nofollow noreferrer>泽西岛:

new Client().target("http://test.com/").request().rx().get(JsonNode.class);

其中 rx()表示请求应该异步进行,直接返回 CompletionStage< JsonNode> 而不是 JsonNode 。如果此调用失败,我会得到此堆栈跟踪:

where rx() indicates that the request should take place asynchronously, returning a CompletionStage<JsonNode> instead of a JsonNode directly. If this call fails, I get this stacktrace:

javax.ws.rs.ForbiddenException: HTTP 403 Authentication Failed
    at org.glassfish.jersey.client.JerseyInvocation.convertToException(JerseyInvocation.java:1083)
    at org.glassfish.jersey.client.JerseyInvocation.translate(JerseyInvocation.java:883)
    at org.glassfish.jersey.client.JerseyInvocation.lambda$invoke$1(JerseyInvocation.java:767)
    at org.glassfish.jersey.internal.Errors.process(Errors.java:316)
    at org.glassfish.jersey.internal.Errors.process(Errors.java:298)
    at org.glassfish.jersey.internal.Errors.process(Errors.java:229)
    at org.glassfish.jersey.process.internal.RequestScope.runInScope(RequestScope.java:414)
    at org.glassfish.jersey.client.JerseyInvocation.invoke(JerseyInvocation.java:765)
    at org.glassfish.jersey.client.JerseyInvocation$Builder.method(JerseyInvocation.java:456)
    at org.glassfish.jersey.client.JerseyCompletionStageRxInvoker.lambda$method$1(JerseyCompletionStageRxInvoker.java:70)
    at java.util.concurrent.CompletableFuture$AsyncSupply.run(CompletableFuture.java:1590)

注意:


  • 堆栈跟踪不引用用户代码。

  • 异常消息不包含有关触发错误的HTTP请求的上下文信息(HTTP方法,URI等)。

因此,我无法将异常追溯到其来源。

As a result, I have no way of tracking the exception back to its source.

如果你深入挖掘,你会发现泽西岛正在调用

If you dig under the hood, you will discover that Jersey is invoking:

CompletableFuture.supplyAsync(() -> getSyncInvoker().method(name, entity, responseType))

rx ()调用。因为供应商是由泽西岛建造的,所以没有参考用户代码。

for rx() invocations. Because the supplier is constructed by Jersey, there is no reference back to user code.

我试图提交针对Jetty的错误报告不相关的异步示例,随后因安全原因被拒绝。

I tried filing a bug report against Jetty for an unrelated async example, and was subsequently turned down on security grounds.

相反,我一直在添加上下文信息如下:

Instead, I've been adding contextual information as follows:

makeHttpRequest().exceptionally(e ->
{
    throw new RuntimeException(e);
});

意思是,我手动添加例外()在我的代码中的每个HTTP请求之后。 Jersey抛出的任何异常都包含在引用我的代码的辅助异常中。生成的堆栈跟踪如下所示:

Meaning, I am manually adding exceptionally() after every single HTTP request in my code. Any exceptions thrown by Jersey are wrapped in a secondary exception that references my code. The resulting stacktrace looks like this:

java.lang.RuntimeException: javax.ws.rs.ForbiddenException: HTTP 403 Authentication Failed
    at my.user.code.Testcase.lambda$null$1(Testcase.java:25)
    at java.util.concurrent.CompletableFuture.uniExceptionally(CompletableFuture.java:870)
    ... 6 common frames omitted
Caused by: javax.ws.rs.ForbiddenException: HTTP 403 Authentication Failed
    at org.glassfish.jersey.client.JerseyInvocation.convertToException(JerseyInvocation.java:1083)
    at org.glassfish.jersey.client.JerseyInvocation.translate(JerseyInvocation.java:883)
    at org.glassfish.jersey.client.JerseyInvocation.lambda$invoke$1(JerseyInvocation.java:767)
    at org.glassfish.jersey.internal.Errors.process(Errors.java:316)
    at org.glassfish.jersey.internal.Errors.process(Errors.java:298)
    at org.glassfish.jersey.internal.Errors.process(Errors.java:229)
    at org.glassfish.jersey.process.internal.RequestScope.runInScope(RequestScope.java:414)
    at org.glassfish.jersey.client.JerseyInvocation.invoke(JerseyInvocation.java:765)
    at org.glassfish.jersey.client.JerseyInvocation$Builder.method(JerseyInvocation.java:456)
    at org.glassfish.jersey.client.JerseyCompletionStageRxInvoker.lambda$method$1(JerseyCompletionStageRxInvoker.java:70)
    at java.util.concurrent.CompletableFuture$AsyncSupply.run(CompletableFuture.java:1590)
    ... 3 common frames omitted

I不喜欢这种方法,因为它容易出错并降低代码的可读性。如果我错误地忽略了一些HTTP请求,我将最终得到一个模糊的堆栈跟踪并花费大量时间跟踪它。

I don't like this approach because it is error prone and decreases the readability of the code. If I mistakenly omit this for some HTTP request I will end up with a vague stacktrace and spend a lot time tracking it down.

此外,如果我想隐藏这个技巧在实用程序类后面,我必须在 CompletionStage 之外实例化一个异常;否则,实用程序类将显示在堆栈跟踪而不是实际的调用站点中。在 CompletionStage 之外实例化异常是非常昂贵的,因为即使异步调用没有抛出任何异常,此代码也会运行。

Further, if I want to hide this trick behind a utility class then I have to instantiate an exception outside of a CompletionStage; otherwise, the utility class will show up in the stacktrace instead of the actual call site. Instantiating an exception outside of a CompletionStage is extremely expensive because this code runs even if no exception is ever thrown by the async call.

是否有一种强大的,易于维护的方法来向异步调用添加上下文信息?

Is there a robust, easy-to-maintain approach to add contextual information to asynchronous calls?

或者,是否有一种有效的方法可以在没有此上下文信息的情况下将堆栈跟踪追溯到其源?

Alternatively, is there an efficient approach to track stacktraces back to their source without this contextual information?

推荐答案

看到由于这个问题在近一个月内没有得到任何答案,我将发布迄今为止我发现的最佳解决方案:

Seeing as this question has not received any answers in almost a month, I'm going to post the best solution I've found to date:

DebugCompletableFuture.java

/**
 * A {@link CompletableFuture} that eases debugging.
 *
 * @param <T> the type of value returned by the future
 */
public final class DebugCompletableFuture<T> extends CompletableFuture<T>
{
    private static RunMode RUN_MODE = RunMode.DEBUG;
    private static final Set<String> CLASS_PREFIXES_TO_REMOVE = ImmutableSet.of(DebugCompletableFuture.class.getName(),
        CompletableFuture.class.getName(), ThreadPoolExecutor.class.getName());
    private static final Set<Class<? extends Throwable>> EXCEPTIONS_TO_UNWRAP = ImmutableSet.of(AsynchronousException.class,
        CompletionException.class, ExecutionException.class);
    private final CompletableFuture<T> delegate;
    private final AsynchronousException asyncStacktrace;

    /**
     * @param delegate the stage to delegate to
     * @throws NullPointerException if any of the arguments are null
     */
    private DebugCompletableFuture(CompletableFuture<T> delegate)
    {
        requireThat("delegate", delegate).isNotNull();
        this.delegate = delegate;
        this.asyncStacktrace = new AsynchronousException();
        delegate.whenComplete((value, exception) ->
        {
            if (exception == null)
            {
                super.complete(value);
                return;
            }
            exception = Exceptions.unwrap(exception, EXCEPTIONS_TO_UNWRAP);
            asyncStacktrace.initCause(exception);
            filterStacktrace(asyncStacktrace, element ->
            {
                String className = element.getClassName();
                for (String prefix : CLASS_PREFIXES_TO_REMOVE)
                    if (className.startsWith(prefix))
                        return true;
                return false;
            });
            Set<String> newMethods = getMethodsInStacktrace(asyncStacktrace);
            if (!newMethods.isEmpty())
            {
                Set<String> oldMethods = getMethodsInStacktrace(exception);
                newMethods.removeAll(oldMethods);
                if (!newMethods.isEmpty())
                {
                    // The async stacktrace introduces something new
                    super.completeExceptionally(asyncStacktrace);
                    return;
                }
            }
            super.completeExceptionally(exception);
        });
    }

    /**
     * @param exception an exception
     * @return the methods referenced by the stacktrace
     * @throws NullPointerException if {@code exception} is null
     */
    private Set<String> getMethodsInStacktrace(Throwable exception)
    {
        requireThat("exception", exception).isNotNull();
        Set<String> result = new HashSet<>();
        for (StackTraceElement element : exception.getStackTrace())
            result.add(element.getClassName() + "." + element.getMethodName());
        for (Throwable suppressed : exception.getSuppressed())
            result.addAll(getMethodsInStacktrace(suppressed));
        return result;
    }

    /**
     * @param <T2>     the type returned by the delegate
     * @param delegate the stage to delegate to
     * @return if {@code RUN_MODE == DEBUG} returns an instance that wraps {@code delegate}; otherwise, returns {@code delegate}
     * unchanged
     * @throws NullPointerException if any of the arguments are null
     */
    public static <T2> CompletableFuture<T2> wrap(CompletableFuture<T2> delegate)
    {
        if (RUN_MODE != RunMode.DEBUG)
            return delegate;
        return new DebugCompletableFuture<>(delegate);
    }

    /**
     * Removes stack trace elements that match a filter. The exception and its descendants are processed recursively.
     * <p>
     * This method can be used to remove lines that hold little value for the end user (such as the implementation of utility functions).
     *
     * @param exception     the exception to process
     * @param elementFilter returns true if the current stack trace element should be removed
     */
    private void filterStacktrace(Throwable exception, Predicate<StackTraceElement> elementFilter)
    {
        Throwable cause = exception.getCause();
        if (cause != null)
            filterStacktrace(cause, elementFilter);
        for (Throwable suppressed : exception.getSuppressed())
            filterStacktrace(suppressed, elementFilter);
        StackTraceElement[] elements = exception.getStackTrace();
        List<StackTraceElement> keep = new ArrayList<>(elements.length);
        for (StackTraceElement element : elements)
        {
            if (!elementFilter.test(element))
                keep.add(element);
        }
        exception.setStackTrace(keep.toArray(new StackTraceElement[0]));
    }

    @Override
    public <U> CompletableFuture<U> thenApply(Function<? super T, ? extends U> fn)
    {
        return wrap(super.thenApply(fn));
    }

    @Override
    public <U> CompletableFuture<U> thenApplyAsync(Function<? super T, ? extends U> fn)
    {
        return wrap(super.thenApplyAsync(fn));
    }

    @Override
    public <U> CompletableFuture<U> thenApplyAsync(Function<? super T, ? extends U> fn, Executor executor)
    {
        return wrap(super.thenApplyAsync(fn, executor));
    }

    @Override
    public CompletableFuture<Void> thenAccept(Consumer<? super T> action)
    {
        return wrap(super.thenAccept(action));
    }

    @Override
    public CompletableFuture<Void> thenAcceptAsync(Consumer<? super T> action)
    {
        return wrap(super.thenAcceptAsync(action));
    }

    @Override
    public CompletableFuture<Void> thenAcceptAsync(Consumer<? super T> action, Executor executor)
    {
        return wrap(super.thenAcceptAsync(action, executor));
    }

    @Override
    public CompletableFuture<Void> thenRun(Runnable action)
    {
        return wrap(super.thenRun(action));
    }

    @Override
    public CompletableFuture<Void> thenRunAsync(Runnable action)
    {
        return wrap(super.thenRunAsync(action));
    }

    @Override
    public CompletableFuture<Void> thenRunAsync(Runnable action, Executor executor)
    {
        return wrap(super.thenRunAsync(action, executor));
    }

    @Override
    public <U, V> CompletableFuture<V> thenCombine(CompletionStage<? extends U> other,
                                                   BiFunction<? super T, ? super U, ? extends V> fn)
    {
        return wrap(super.thenCombine(other, fn));
    }

    @Override
    public <U, V> CompletableFuture<V> thenCombineAsync(CompletionStage<? extends U> other,
                                                        BiFunction<? super T, ? super U, ? extends V> fn)
    {
        return wrap(super.thenCombineAsync(other, fn));
    }

    @Override
    public <U, V> CompletableFuture<V> thenCombineAsync(CompletionStage<? extends U> other,
                                                        BiFunction<? super T, ? super U, ? extends V> fn,
                                                        Executor executor)
    {
        return wrap(super.thenCombineAsync(other, fn, executor));
    }

    @Override
    public <U> CompletableFuture<Void> thenAcceptBoth(CompletionStage<? extends U> other,
                                                      BiConsumer<? super T, ? super U> action)
    {
        return wrap(super.thenAcceptBoth(other, action));
    }

    @Override
    public <U> CompletableFuture<Void> thenAcceptBothAsync(CompletionStage<? extends U> other,
                                                           BiConsumer<? super T, ? super U> action)
    {
        return wrap(super.thenAcceptBothAsync(other, action));
    }

    @Override
    public <U> CompletableFuture<Void> thenAcceptBothAsync(CompletionStage<? extends U> other,
                                                           BiConsumer<? super T, ? super U> action,
                                                           Executor executor)
    {
        return wrap(super.thenAcceptBothAsync(other, action, executor));
    }

    @Override
    public CompletableFuture<Void> runAfterBoth(CompletionStage<?> other, Runnable action)
    {
        return wrap(super.runAfterBoth(other, action));
    }

    @Override
    public CompletableFuture<Void> runAfterBothAsync(CompletionStage<?> other, Runnable action)
    {
        return wrap(super.runAfterBothAsync(other, action));
    }

    @Override
    public CompletableFuture<Void> runAfterBothAsync(CompletionStage<?> other, Runnable action, Executor executor)
    {
        return wrap(super.runAfterBothAsync(other, action, executor));
    }

    @Override
    public <U> CompletableFuture<U> applyToEither(CompletionStage<? extends T> other, Function<? super T, U> fn)
    {
        return wrap(super.applyToEither(other, fn));
    }

    @Override
    public <U> CompletableFuture<U> applyToEitherAsync(CompletionStage<? extends T> other, Function<? super T, U> fn)
    {
        return wrap(super.applyToEitherAsync(other, fn));
    }

    @Override
    public <U> CompletableFuture<U> applyToEitherAsync(CompletionStage<? extends T> other, Function<? super T, U> fn,
                                                       Executor executor)
    {
        return wrap(super.applyToEitherAsync(other, fn, executor));
    }

    @Override
    public CompletableFuture<Void> acceptEither(CompletionStage<? extends T> other, Consumer<? super T> action)
    {
        return wrap(super.acceptEither(other, action));
    }

    @Override
    public CompletableFuture<Void> acceptEitherAsync(CompletionStage<? extends T> other, Consumer<? super T> action)
    {
        return wrap(super.acceptEitherAsync(other, action));
    }

    @Override
    public CompletableFuture<Void> acceptEitherAsync(CompletionStage<? extends T> other, Consumer<? super T> action,
                                                     Executor executor)
    {
        return wrap(super.acceptEitherAsync(other, action, executor));
    }

    @Override
    public CompletableFuture<Void> runAfterEither(CompletionStage<?> other, Runnable action)
    {
        return wrap(super.runAfterEither(other, action));
    }

    @Override
    public CompletableFuture<Void> runAfterEitherAsync(CompletionStage<?> other, Runnable action)
    {
        return wrap(super.runAfterEitherAsync(other, action));
    }

    @Override
    public CompletableFuture<Void> runAfterEitherAsync(CompletionStage<?> other, Runnable action, Executor executor)
    {
        return wrap(super.runAfterEitherAsync(other, action, executor));
    }

    @Override
    public <U> CompletableFuture<U> thenCompose(Function<? super T, ? extends CompletionStage<U>> fn)
    {
        return wrap(super.thenCompose(fn));
    }

    @Override
    public <U> CompletableFuture<U> thenComposeAsync(Function<? super T, ? extends CompletionStage<U>> fn)
    {
        return wrap(super.thenComposeAsync(fn));
    }

    @Override
    public <U> CompletableFuture<U> thenComposeAsync(Function<? super T, ? extends CompletionStage<U>> fn,
                                                     Executor executor)
    {
        return wrap(super.thenComposeAsync(fn, executor));
    }

    @Override
    public CompletableFuture<T> exceptionally(Function<Throwable, ? extends T> fn)
    {
        return wrap(super.exceptionally(fn));
    }

    @Override
    public CompletableFuture<T> whenComplete(BiConsumer<? super T, ? super Throwable> action)
    {
        return wrap(super.whenComplete(action));
    }

    @Override
    public CompletableFuture<T> whenCompleteAsync(BiConsumer<? super T, ? super Throwable> action)
    {
        return wrap(super.whenCompleteAsync(action));
    }

    @Override
    public CompletableFuture<T> whenCompleteAsync(BiConsumer<? super T, ? super Throwable> action,
                                                  Executor executor)
    {
        return wrap(super.whenCompleteAsync(action, executor));
    }

    @Override
    public <U> CompletableFuture<U> handle(BiFunction<? super T, Throwable, ? extends U> fn)
    {
        return wrap(super.handle(fn));
    }

    @Override
    public <U> CompletableFuture<U> handleAsync(BiFunction<? super T, Throwable, ? extends U> fn)
    {
        return wrap(super.handleAsync(fn));
    }

    @Override
    public <U> CompletableFuture<U> handleAsync(BiFunction<? super T, Throwable, ? extends U> fn,
                                                Executor executor)
    {
        return wrap(super.handleAsync(fn, executor));
    }

    @Override
    public boolean complete(T value)
    {
        return delegate.complete(value);
    }

    @Override
    public boolean completeExceptionally(Throwable ex)
    {
        return delegate.completeExceptionally(ex);
    }
}

RunMode.java

/**
 * Operational modes.
 */
public enum RunMode
{
    /**
     * Optimized for debugging problems (extra runtime checks, logging of the program state).
     */
    DEBUG,
    /**
     * Optimized for maximum performance.
     */
    RELEASE
}

AsynchronousException。 java

/**
 * Thrown when an asynchronous operation fails. The stacktrace indicates who triggered the operation.
 */
public final class AsynchronousException extends RuntimeException
{
    private static final long serialVersionUID = 0L;

    public AsynchronousException()
    {
    }
}

用法:

DebugCompletableFuture.wrap(CompletableFuture.supplyAsync(this::expensiveOperation));

上行:您将获得相对干净的异步堆栈跟踪。

Upside: you'll get relatively clean asynchronous stack traces.

下行:每次创建未来时构造一个新的 AsynchronousException 是非常昂贵的。具体来说,如果你生成很多未来,这会在堆上产生大量垃圾,GC开销变得明显。

Downside: Constructing a new AsynchronousException every time a future is created is extremely expensive. Specifically, if you're generating a lot of futures, this generates a lot of garbage on the heap and the GC overhead becomes noticeable.

我仍然希望有人会提出了一种表现更好的方法。

I am still hopeful that someone will come up with a better-performing approach.

这篇关于如何创建异步堆栈跟踪?的文章就介绍到这了,希望我们推荐的答案对大家有所帮助,也希望大家多多支持IT屋!

查看全文
登录 关闭
扫码关注1秒登录
发送“验证码”获取 | 15天全站免登陆