commons-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Siegfried Goeschl (JIRA)" <j...@apache.org>
Subject [jira] Commented: (EXEC-34) Race condition prevent watchdog working using ExecuteStreamHandler
Date Sat, 14 Aug 2010 19:35:18 GMT

    [ https://issues.apache.org/jira/browse/EXEC-34?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12898631#action_12898631
] 

Siegfried Goeschl commented on EXEC-34:
---------------------------------------

Well, actually the sample is buggy since the following line is missing

exec.setWatchdog(watchdog)

which injects the newly created 'Process' instance into the watchdog and without *Process'
instance 'watchdog.destroy()' has no effect

> Race condition prevent watchdog working using ExecuteStreamHandler
> ------------------------------------------------------------------
>
>                 Key: EXEC-34
>                 URL: https://issues.apache.org/jira/browse/EXEC-34
>             Project: Commons Exec
>          Issue Type: Bug
>         Environment: Windows Vista 64bit, dual core CPU
>            Reporter: Marco Ferrante
>            Assignee: Siegfried Goeschl
>            Priority: Minor
>
> Consider this test case (in _DefaultExecutorTest_ class):
> {noformat}
>     /**
>      * Start a async process using a stream handler and terminate it manually
>      * before the watchdog timeout occurs
>      */
>     public void testExecuteAsyncWithStreamHandlerAndUserTermination() throws Exception
{
>         CommandLine cl = new CommandLine(foreverTestScript);
>         ExecuteWatchdog watchdog = new ExecuteWatchdog(Integer.MAX_VALUE);
>         PumpStreamHandler streamHanlder = new PumpStreamHandler(System.out, System.err);
>         exec.setStreamHandler(streamHanlder);
>         MockExecuteResultHandler handler = new MockExecuteResultHandler();
>         exec.execute(cl, handler);
>         // DON'T wait for script to run
>         //Thread.sleep(2000);
>         // teminate it
>         watchdog.destroyProcess();
>         assertTrue("Watchdog should have killed the process",watchdog.killedProcess());
>     }
> {noformat}
> It fails (at least in my environment) because when _watchdog.destroyProcess()_ is invoked
the external process is not bound to the watchdog yet.
> Although there are possible several workarounds, but all of them seem to me very intrusive
in the code. So, I prefer some discussion before preparing and submitting a patch.
> IMHO, the watchdog should handle a reference to the thread running the process, not to
the process itself. In this way, interrupting signals can be transport using default _interrupt()_
method of class _Thread_.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message