[Pvfs2-developers] valgrind?

[Sam Lang]

Hi Walt,

Just to follow up, I valgrind the server occasionally.  It works  
fairly well for finding memory errors (it found that early free with  
BMI_memfree we had last week).  I usually get a couple errors when  
the server starts that can be ignored, IIRC they're epoll related.   
Its especially nice for mem leak checking...RobL and I have talked  
about adding a nightly set of server tests where the server runs in  
valgrind to check for mem leaks and other errors valgrind may find,  
but talk is as far as we've gotten.

As a side note, I've also tried to use valgrind to find some race  
conditions (with the helgrind tool) in the client api that appear  
when using multiple threads.   The issue is that the latest version  
of valgrind doesn't support the helgrind tool, so you have to go back  
and use 2.2.0, which turns out to work fairly well, although I was  
getting a lot of false positives.

-sam

On Aug 29, 2006, at 9:55 AM, Walter B. Ligon III wrote:

> I've got RHEL3 and I'm running valgrind 3.1.1 - seems to be working  
> OK now, and its really not too bad.  I think I'm homing in on it,  
> though any hints for making it easier are welcome!
>
> Right now it looks like I'm calling job_null with a bad smcb -  
> things are a little weird ...
>
> Usually, when we start a new SM we immediately begin executing it  
> until it blocks, then we return to what we were doing.  The test  
> I'm running is weird because the little SM I'm running concurrently  
> in the middle of a get_config doesn't DO anything, so it never  
> blocks, so it actually runs to completion before returning to the  
> caller.
>
> Now, I considered altering this so that the new SMs are NOT started  
> immediately, but scheduled to run, but after talking to Pete  
> decided not to do that - after all, REAL SMs will block, and this  
> might reduce latency.
>
> So, what I'm getting it is, I might be calling job_null on a SM  
> that later completes and gets freed.  I'm not sure, I'm trying to  
> figure that out and decide on a way to cleanly deal with that.
>
> (you may recall I said back in July that termination was the real  
> bug-a-boo to this thing)
>
> Walt
>
> Rob Ross wrote:
>> hey walt,
>> i've had mixed results. what os and distro are you trying?
>> rob
>> Walter B. Ligon III wrote:
>>> has anyone run the server with valgrind?
>>> I have a nasty memory bug and I'm trying to use it to find the  
>>> problem, but the server won't run with it - tells me some of the  
>>> db functions don't work????
>>>
>>> Any experience?
>>>
>>> Walt
>
> -- 
> Dr. Walter B. Ligon III
> Associate Professor
> ECE Department
> Clemson University
> _______________________________________________
> Pvfs2-developers mailing list
> Pvfs2-developers at beowulf-underground.org
> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
>