[GE users] Memory leak in 6.1u2 ?

Andreas.Haas at Sun.COM Andreas.Haas at Sun.COM
Thu Nov 15 15:17:12 GMT 2007


On Thu, 15 Nov 2007, Richard Ems wrote:

> Andreas.Haas at Sun.COM wrote:
>> Hi Richard,
>>
>> On Thu, 15 Nov 2007, Richard Ems wrote:
>>
>>> Andreas.Haas at Sun.COM wrote:
>>>
>>>> First step is to patch the source code and built the binaries.
>>>> You using lx24-amd64 right now?
>>>
>>> Yes!
>>
>> you find your schedd binary under
>>
>>    http://gridengine.sunsource.net/files/documents/7/160/sge_schedd.gz
>>
>> the scheduler is identical with 6.1u2 for lx24-amd64 except for the
>> patch below. What the patch does for you is (a) enable allocation
>> recording at scheduler start-up time (b) disable it at shut-down time.
>> Make sure MALLOC_TRACE is set in schedd's environment as described in
>> mtrace(3), since otherwise no recording will happen.
>>
>> My recommendation is to shut-down schedd once you observe memory
>> leaking. That way mtrace(3) will report almost only real leaks when it
>> does the memory balance.
>
> Ok, thanks.
> I tried it, but the error is not triggering anymore. Since I tried a
> couple of times enabling just single queues, the jobs got started and
> with less jobs waiting the error seems not to happen.
>
> The mtrace log file was getting very big, I stopped SGE and it was at
> over 700MB, after just some minutes. Is this normal? What do you want me
> to do when the error triggers again and I get a 700 MB file? Send i to
> you? 8)

It is true that the file becomes large, but it is normal since MALLOC_TRACE 
simply records any single allocation/deallocation. Sending me the file shouldn't 
be necessary since Linux comes with /usr/bin/mtrace. This utility does the 
memory balance so that all information about memory that was malloc()'d and 
free()'d is sieved out. So just run mtrace with with you $MALLOC_TRACE as 
argument and send me it's output.

Regards,
Andreas

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list