[Ace-users] [ace-users] Race condition between handle_input() & handle_timeout() in ACE_Svc_Handler ?

[Douglas C. Schmidt]

Hi Shawn,

Thanks for using the PRF.

>ACE VERSION: 5.4

ACE 5.4 is ancient.  Please upgrade to ACE+TAO+CIAO x.6.1 (i.e., ACE
5.6.1, TAO 1.6.1, and CIAO 0.6.1), which you can download from

http://download.dre.vanderbilt.edu

under the heading: "Latest Beta Kit."

The DOC groups at Washington University, UC Irvine, and Vanderbilt
University only provide "best effort" support for non-sponsors for the
latest release, as described in

http://www.dre.vanderbilt.edu/~schmidt/DOC_ROOT/ACE/docs/ACE-bug-process.html

Thus, if you need more "predictable" help for earlier versions of
ACE+TAO, I recommend that you check out

http://www.dre.vanderbilt.edu/support.html

for a list of companies that will provide you with ACE+TAO commercial
support.

>HOST MACHINE and OPERATING SYSTEM:
>     Windows XP sp2
>
>COMPILER NAME AND VERSION (AND PATCHLEVEL):
>     MSVC6 SP6
>
>CONTENTS OF $ACE_ROOT/ace/config.h:
>    config-win32.h
>
>AREA/CLASS/EXAMPLE AFFECTED:
>    TP_Reactor, ACE_Svc_Handler
>
>DOES THE PROBLEM AFFECT:
>        EXECUTION
>
>SYNOPSIS:
>        Application crashes randomly if a handler is handling the
>request and at the same time it's expired by timer.
>
>DESCRIPTION:
>
>       Hi all ACE experts,
>
>I'm newbie to ACE. Now I'm struggle with a nasty crash in my
>application, and it's seems caused by race condition between the
>handle_timout() and handle_input()
>
>This is a service application, using ACE5.4, running under windows XP
>I used TPReactor, and extended ACE_Svc_Handler to handle incoming tcp
>requests.
>
>In order to recycle/close idle connections, we also registered a timer
>and in our Svc_Handler::handle_timeout(), if detected that handler is
>idle, then we'll return -1 to release the handle.
>
>But I found an interesting problem by this *timeout* mechanism, that
>is, if a request come in at the *exact* time when *timeout* is fired,
>ACE might crash.
>
>To reproduce this problem, I changed the server idle timeout interval
>to 5s, and write a client app to send request to server every 5s, soon
>the service application crashed.
>By analysising the log and call stack trace and the thread states, I
>found that there seems to be a race condition.
>
>Request -> come in -> Handler ctor called and timer is registered ->
>back to the ACE -> OS Switch to another thread -> ACE trigger the
>timer and spawn a new thread to call the handle_timeout on this newly
>created handler -> handle_timeout return -1(detected the busy/active
>state, but got idle information) -> back to ACE -> handler destructor
>called -> OS switch back to the first thread -> crash...
>
>Any ideas ? or Could anyone give me some advice about how to close the
>handler gracefully & thread-safely when idle timeout is triggered ?

If you're using ACE_TP_Reactor you probably need to use an
ACE_Thread_Mutex to serialize the processing between the various
ACE_Svc_Handler methods.  There are some examples of how to do this
sort of thing in the C++NP books <www.cs.wustl.edu/~schmidt/ACE/>.

Thanks,

Doug
-- 
Dr. Douglas C. Schmidt                       Professor and Associate Chair
Electrical Engineering and Computer Science  TEL: (615) 343-8197
Vanderbilt University                        WEB: www.dre.vanderbilt.edu/~schmidt
Nashville, TN 37203                          NET: d.schmidt at vanderbilt.edu