[
https://issues.apache.org/jira/browse/HADOOP-12748?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sidharta Seethana updated HADOOP-12748:
---------------------------------------
Description:
When tasks running in a {{ThreadPoolExecutor}} throw an error/exception that is
otherwise uncaught, it might not be propagated to the thread's uncaught
exception handler because enclosing tasks (e.g {{FutureTask}}) might capture
these exceptions (and do nothing). For more details, see :
https://docs.oracle.com/javase/8/docs/api/java/util/concurrent/ThreadPoolExecutor.html#afterExecute-java.lang.Runnable-java.lang.Throwable-
To ensure such errors are logged, one mechanism would be to override
{{ThreadPoolExecutor}} with an {{afterExecute}} method that would log these
exceptions/errors. Logging these exceptions/errors would be useful in debugging
issues that are otherwise difficult to trace (e.g YARN-4643) because there is
nothing in the logs indicating an uncaught exception/error
Some test code that demonstrates this issue is attached.
was:
{{ThreadPoolExecutor}} catches (otherwise uncaught) exceptions/errors thrown by
the tasks running in the corresponding thread pool. These are passed to an
{{afterExecute}} method which seems to do nothing by default unless overridden.
Even though we register {{UncaughtExceptionHandler}} s in various places (e.g
{{YarnUncaughtExceptionHandler}}), these handlers are not invoked because the
uncaught exceptions/errors are not propagated all the way
To fix this, one mechanism would be to override {{ThreadPoolExecutor}} with an
{{afterExecute}} method that would log these exceptions/errors. Logging these
exceptions/errors would be useful in debugging issues that are otherwise
difficult to trace (e.g YARN-4643) because there is nothing in the logs
indicating an uncaught exception/error
Edit: uploaded some test code that demonstrates this issue.
> Uncaught exceptions are not caught/logged when ExecutorService is used
> ----------------------------------------------------------------------
>
> Key: HADOOP-12748
> URL: https://issues.apache.org/jira/browse/HADOOP-12748
> Project: Hadoop Common
> Issue Type: Bug
> Reporter: Sidharta Seethana
> Assignee: Sidharta Seethana
> Attachments: TestUncaughExceptionHandler.java
>
>
> When tasks running in a {{ThreadPoolExecutor}} throw an error/exception that
> is otherwise uncaught, it might not be propagated to the thread's uncaught
> exception handler because enclosing tasks (e.g {{FutureTask}}) might capture
> these exceptions (and do nothing). For more details, see :
> https://docs.oracle.com/javase/8/docs/api/java/util/concurrent/ThreadPoolExecutor.html#afterExecute-java.lang.Runnable-java.lang.Throwable-
> To ensure such errors are logged, one mechanism would be to override
> {{ThreadPoolExecutor}} with an {{afterExecute}} method that would log these
> exceptions/errors. Logging these exceptions/errors would be useful in
> debugging issues that are otherwise difficult to trace (e.g YARN-4643)
> because there is nothing in the logs indicating an uncaught exception/error
> Some test code that demonstrates this issue is attached.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
