ScheduledExecutorService异常处理
我使用 ScheduledExecutorService 定期执行一个方法。
p-code:
ScheduledExecutorService scheduler = Executors.newSingleThreadScheduledExecutor();
ScheduledFuture<?> handle =
scheduler.scheduleWithFixedDelay(new Runnable() {
public void run() {
//Do business logic, may Exception occurs
}
}, 1, 10, TimeUnit.SECONDS);
我的问题:
如果 run() 抛出异常,如何继续调度程序?
我应该尝试捕获方法 run()
中的所有异常吗?或者任何内置的回调方法来处理异常?谢谢!
如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱,如果其他用户或者作者回复了您的评论,将不能在第一时间通知您!
发布评论
评论(9)
tl;dr
任何转义
run
方法的异常都会停止所有进一步的工作,恕不另行通知。始终在
run
方法中使用try-catch
。如果您希望继续安排的活动,请尝试恢复。问题问题
涉及 的关键技巧
ScheduledExecutorService
:任何到达执行器的抛出异常或错误都会导致执行器停止。不再对 Runnable 进行调用,不再工作已完成。这次停工悄然发生,您不会被通知。这种顽皮的语言博客文章有趣地讲述了了解这种行为的艰难过程。解决方案
yegor256的回答和arun_suresh 的回答 似乎基本上都是正确的。这些答案有两个问题:
错误和异常?
在Java中,我们通常只捕获异常,而不是错误。但在 ScheduledExecutorService 的这种特殊情况下,未能捕获其中任何一个都将意味着工作停止。所以你可能想两者都抓住。我对此不是 100% 确定,不完全了解捕获所有错误的含义。如果需要请纠正我。
捕获错误和异常的原因之一可能涉及在任务中使用库。请参阅 jannis 的评论。
捕获异常和错误的一种方法是捕获它们的超类, Throwable 为例。
...而不是...
最简单的方法:只需添加一个
Try-Catch
但这两个答案都有点复杂。只是为了记录,我将展示最简单的解决方案:
Lambda 语法
使用 lambda(在 Java 8 及更高版本中)。
老式语法
在 lambda 之前的老式方式。
在每个 Runnable/Callable 中,
无论
ScheduledExecutorService
,对我来说,在任何中始终使用通用try-catch(Exception† e)
似乎是明智的运行
Runnable
< 的方法/a>.对于任何调用也是如此
方法rel="noreferrer">
可调用
。完整的示例代码
在实际工作中,我可能会单独定义 Runnable 而不是嵌套。但这是一个简洁的一体化示例。
运行时。
另一个例子
这是另一个例子。在这里,我们的任务打算运行大约二十次,每五秒运行一次,持续一分钟。但在第五次运行时,我们抛出异常。
运行时。
注意:
因此,当您的任务抛出异常时,您可能会得到最糟糕的结果:无声地停止工作,没有任何解释。
解决方案,如上所述:始终在
run
方法中使用try-catch
。† 或者也许
Throwable
而不是异常
< /a> 捕捉Error
对象也是如此。tl;dr
Any exception escaping your
run
method halts all further work, without notice.Always use a
try-catch
within yourrun
method. Try to recover if you want scheduled activity to continue.The Problem
The question refers to the critical trick with a
ScheduledExecutorService
: Any thrown exception or error reaching the executor causes the executor to halt. No more invocations on the Runnable, no more work done. This work stoppage happens silently, you'll not be informed. This naughty-language blog posting entertainingly narrates the hard way to learn about this behavior.The Solution
The answer by yegor256 and the answer by arun_suresh both seem to be basically correct. Two issues with those answers:
Errors and Exceptions ?
In Java we normally catch only exceptions, not errors. But in this special case of ScheduledExecutorService, failing to catch either will mean a work stoppage. So you may want to catch both. I'm not 100% sure about this, not knowing fully the implications of catching all errors. Please correct me if needed.
One reason to catch errors as well as exceptions might involve the use of libraries within your task. See the comment by jannis.
One way to catch both exceptions and errors is to catch their superclass, Throwable for an example.
…rather than…
Simplest Approach: Just Add a
Try-Catch
But both answers are a bit complicated. Just for the record, I'll show the simplest solution:
Lambda Syntax
With a lambda (in Java 8 and later).
Old-Fashioned Syntax
The old-fashioned way, before lambdas.
In Every Runnable/Callable
Regardless of a
ScheduledExecutorService
, it seems sensible to me to always use a generaltry-catch( Exception† e )
in anyrun
method of aRunnable
. Ditto for anycall
method of aCallable
.Complete example code
In real work, I would likely define the
Runnable
separately rather than nested. But this makes for neat all-in-one example.When run.
Another example
Here is another example. Here our task is meant to run about twenty times, once every five seconds for a minute. But on the fifth run, we throw an exception.
When run.
Notice:
So when your task throws an exception, you get the worst outcome possible: Silent work stoppage with no explanation.
The solution, as mentioned above: Always use a
try-catch
within yourrun
method.† Or perhaps
Throwable
instead ofException
to catchError
objects too.您应该使用
scheduler.scheduleWithFixedDelay(...)
返回的ScheduledFuture
对象,如下所示:You should use the
ScheduledFuture
object returned by yourscheduler.scheduleWithFixedDelay(...)
like so :老问题,但接受的答案没有给出解释,并提供了一个糟糕的例子,而最受支持的答案在某些方面是正确的,但最终鼓励您在每个 Runnable.run( ) 方法。
我不同意,因为:
我认为异常传播应该由 ExecutorService 框架执行,实际上它提供了该功能。
此外,试图通过短路 ExecutorService 的工作方式来变得太聪明也不是一个好主意:框架可能会发展,而您希望以标准方式使用它。
最后,让
ExecutorService
框架完成其工作并不意味着一定要停止后续的调用任务。如果计划任务遇到问题,调用者有责任根据问题原因重新计划或不计划任务。
每一层都有其职责。保留这些使代码既清晰又可维护。
ScheduledFuture.get() :捕获任务中发生的异常和错误的正确 API
ScheduledExecutorService.scheduleWithFixedDelay()/scheduleAtFixRate()
规范中的状态:这意味着
ScheduledFuture.get()
不会在每次计划调用时返回,而是在最后一次调用任务时返回,即任务取消:由ScheduledFuture.cancel( 引起) )
或任务中抛出的异常。因此,处理
ScheduledFuture
返回以使用ScheduledFuture.get()
捕获异常看起来是正确的:默认行为示例:如果任务执行之一遇到问题,则停止调度
它执行的任务在第三次执行时抛出异常并终止调度。
在某些情况下,我们希望如此。
输出 :
如果其中一个任务执行遇到问题,则可以继续进行调度的示例
它执行的任务在前两次执行时抛出异常,并在第三次执行时抛出错误。
我们可以看到任务的客户端可以选择停止或不停止调度:这里我在出现异常的情况下继续,在出现错误的情况下停止。
输出 :
Old question but the accepted answer doesn't give explanations and provides a poor example and the most upvoted answer is right on some points but finally encourages you to add
catch
exceptions in everyRunnable.run()
method.I disagree because :
I think that the exception propagation should be performed by the
ExecutorService
framework and actually it offers that feature.Besides, trying to be too clever by trying to short-circuiting the
ExecutorService
way of working is not a good idea either : the framework may evolve and you want to use it in a standard way.At last, letting the
ExecutorService
framework to make its job doesn't mean necessarily halting the subsequent invocations task.If a scheduled task encounters an issue, that is the caller responsibility to re-schedule or not the task according to the issue cause.
Each layer has its its responsibilities. Keeping these make code both clear and maintainable.
ScheduledFuture.get() : the right API to catch exceptions and errors occurred in the task
ScheduledExecutorService.scheduleWithFixedDelay()/scheduleAtFixRate()
state in their specification :It means that
ScheduledFuture.get()
doesn't return at each scheduled invocation but that it returns for the last invocation of the task, that is a task cancelation : caused byScheduledFuture.cancel()
or a exception thrown in the task.So handling the
ScheduledFuture
return to capture the exception withScheduledFuture.get()
looks right :Example with the default behavior : halting the scheduling if one of the task execution encounters an issue
It executes a task that for the third executions thrown an exception and terminates the scheduling.
In some scenarios, we want that.
Output :
Example with the possibility to go on the scheduling if one of the task execution encounters an issue
It executes a task that throws an exception at the two first executions and throws an error at the third one.
We can see that the client of the tasks can choose to halt or not the scheduling : here I go on in cases of exception and I stop in case of error.
Output :
另一种解决方案是在 Runnable 中吞掉异常。您可以使用方便的
VerboseRunnable
来自 jcabi-log 的类,例如:Another solution would be to swallow an exception in the
Runnable
. You can use a convenientVerboseRunnable
class from jcabi-log, for example:我知道这是一个老问题,但是如果有人将延迟的 CompletableFuture 与 ScheduledExecutorService 一起使用,那么应该以这种方式处理这个问题:
并在 CompletableFuture 中处理异常>:
I know that this is old question, but if somebody is using delayed
CompletableFuture
withScheduledExecutorService
then should handle this in that way:and handling exception in
CompletableFuture
:捕获异常并使计划任务保持活动状态的巧妙方法。
首先,定义一个函数式接口。
然后,像这样提交工作。
An elegent way to catch the exception and keep scheduled tasks alive.
First, define a functional interface.
Then, commit the job like this.
受 @MBec 解决方案的启发,我为 ScheduledExecutorService 编写了一个很好的通用包装器:
:)
Inspired by @MBec solution, I wrote a nice generic wrapper for the ScheduledExecutorService that:
:)
传递给 (ScheduledExecutorService) 的线程的 run() 中的任何异常都不会被抛出,如果我们使用 future.get() 来获取状态,那么主线程将无限等待
Any exception in the run() of a thread which is passed to (ScheduledExecutorService) is never thrown out and if we use future.get() to get status, then the main thread waits infinitely
就我个人而言,我不同意这里的所有答案。它们的主要问题是它们以奇怪的方式提供相同的解决方案。相反,您应该做的是创建自己的线程工厂,在正在创建的线程上安装未捕获的异常处理程序。例如,这是安装到任何可以自行创建线程的执行程序中的 DefaultThreadFactory。遗憾的是,从 Java 11 开始它仍然是一个私有类,因为我想扩展它而不是将它复制到我的代码库中。下面是它在 Executors.java 文件中的显示方式的片段。
正如您所看到的,该接口本身是一个处理创建新线程的单一方法。除了找出创建线程工厂的线程组之外,它没有什么魔力。有趣的是,线程是作为非守护进程创建的。
创建线程时,您可以调用 setThreadUncaughtExceptionHandler ,它接受一个处理程序,您应该在其中处理该线程中发生的任何未捕获的异常。默认情况下,它将从您的线程组继承,该线程组具有以下内容
默认情况下,它将尝试将处理委托给父线程组(如果存在),然后才测试平台默认未捕获的异常处理程序。通常它没有显式安装。如果您想对不了解这一点的不良代码库造成一些真正的损害,您可以通过 Thread#setDefaultUncaughtExceptionHandler 安装一个。不用担心,如果运行时具有安全管理器,您将无法执行此操作。
如果您要安装自己的处理程序,则将调用该处理程序而不是第一组。
现在,解决这个问题,回答你的问题:如何处理执行器中的异常。默认情况下,如果代码无法处理自己的错误,则线程被视为死亡。我认为你应该坚持这一点。未捕获的异常处理程序不会保存您的线程。相反,它将帮助您诊断发生了什么。为了进入 ScheduledExecutor 实现(允许定期执行可运行对象),适用相同的规则:如果一次执行失败,则该线程以及应该运行的可运行对象将被终止。
简而言之,处理你自己的错误。我们检查异常是有原因的。
但是未经检查的异常呢?
有趣的是,因为我会犯与其他发帖者相同的错误:在
Throwable
上使用 try/catch,但断言它不是ThreadDeath 错误。如果你确实得到了一个,你必须重新抛出它以确保线程确实死亡。
Personally, I disagree with all the answers here. The main issue with all of them is they provide the same solution in weird flavors. Instead, what you should be doing is creating your own thread factory that installs an uncaught exception handler on the thread that is being created. For example, this is the DefaultThreadFactory that is installed into any executor that would create threads on its own. Shamefully, it's still a private class as of Java 11, since I would like to extend it instead of copying it into my codebase. Below is a snippet how it appears in
Executors.java
file.As you can see, the interface itself is a single method that handles creating new threads. There isn't much magic to it besides figuring out the thread group where is the thread factory created in. The interesting bit is that threads are created as non-daemon.
When the thread is created, you can call
setThreadUncaughtExceptionHandler
which accepts a handler where you should be handling any uncaught exceptions that had happened in that thread. By default, it will be inherited from your thread group, which has the followingBy default, it will attempt to delegate handling to parent thread group if it exists, and only then test for platform default uncaught exception handler. Usually it is not explicitly installed. If you want to do some real damage to poor codebases that are not aware of this, you can install one via
Thread#setDefaultUncaughtExceptionHandler
. Don't worry, you won't get to do that if the runtime has Security manager in place.If you were to install your own handler, that handler will be called instead of the group one.
Now with that out of the way, to your question: How do you handle exceptions in Executors. By default, a thread is considered dead if code is unable to handle its own errors. And I think you should adhere to that. Uncaught exception handler won't save your thread. Instead it will help you diagnose what happened. To segway into ScheduledExecutor implementations, which permit periodic execution of a runnable, the same rules apply: if one execution fails, the thread is killed, along with the runnable that was supposed to get run.
In short, handle your own errors. We have checked exceptions for a reason.
But what about unchecked exceptions?
Funny, since I will commit the same sin as other posters do: use try/catch on
Throwable
, but assert that it's not aThreadDeath
error. If you do get one, you must rethrow it to ensure the thread actually does die.