Re: [Boost-testing] [interprocess] upgradeable_mutex_test deadlocking on sun-5.8

From: K. Noel Belcourt (kbelco_at_[hidden])
Date: 2007-09-19 22:37:38

On Sep 19, 2007, at 7:34 PM, Beman Dawes wrote:

> K. Noel Belcourt wrote:
>>
>> On Sep 15, 2007, at 11:52 AM, Rene Rivera wrote:
>>
>>> K. Noel Belcourt wrote:
>>>> On Sep 14, 2007, at 9:37 AM, K. Noel Belcourt wrote:
>>>>> On Sep 14, 2007, at 9:30 AM, Rene Rivera wrote:
>>>
>>>>>> <http://svn.boost.org/trac/boost/ticket/1266>.
>>>
>>>>> This sounds like fun.  I'll take a look.
>>>
>>>> I've got this in and working.
>>>
>>> Awesome!
>>
>> The attached patch has been compiled with Sun, Ibm (vacpp), Gcc,  
>> Intel,
>> Pathscale, and Pgi unix toolsets.  As you'll notice in the patch,  
>> I use
>> both setrlimit, to set a maximum cpu process usage limit, and have  
>> logic
>> to compute the elapsed time in seconds to catch processes that  
>> hang and
>> fail to consume any cpu.  The setrlimit call is not strictly  
>> necessary
>> and could be removed, but I've left it in for now, and it may be  
>> useful
>> to constrain other resource usage in the future).
>>
>> I've set up all my nightly Boost regression tests to pass -l100,  
>> just to
>> make sure something gets killed and reaped.  This means the Sandia
>> results will be suspect for the next day or so as I may be killing  
>> tests
>> that would, if given enough time, complete and pass just fine.  If my
>> tests run okay tonight (no hangs or other weirdness), and there's no
>> other objections, I'll commit this patch to the trunk sometime  
>> tomorrow.
>
> What does the reporting look like when a test times out?  We need  
> to make
> sure timeouts get reported as fails, and the detail report gives the
> reason as "100 second time limit exceeded" or some similar message  
> that
> makes it clear what happened.

Good idea.  I'm just sending a SIGKILL to any expired process.  Let  
me see if I can add some verbiage to the output text to make it clear  
the action timed out.

-- Noel