How large should a mogilefsd process become?
Andreas J. Koenig
andreas.koenig.gmwojprw at franz.ak.mind.de
Tue Aug 22 08:12:00 UTC 2006
>>>>> On Fri, 11 Aug 2006 12:36:45 -0700, Mark Smith <junior at danga.com> said:
> Well, the replication process is getting a review/rework currently, so
> we'll keep an eye out for anything that would cause it to bloat.
you can relax, I have good news: there was a memory leak in DBI that
was fixed in 1.52. It was apparently the major acting partner for the
memory leak in mogilefs. I installed DBI 1.52 yesterday and over night
I saw just a tiny amount of growth which was nothing compared to the
growth I observed before the upgrade. I'll keep an eye on that.
So problem solved, at least mostly.
Now one might write
use DBI 1.52
or one might consider using prepare_cached() instead of prepare() in
so far the leak was caused by creating statement handles and
prepare_cached might improve performance as a side effect.
The concept of komtanoo.pinpimai to add some sort of
MaxRequestPerChild is probably the most robust solution against all
sorts of memory leaks which might appear in other perl/OS/module
combinations in the future.
Thanks to all contributors in this thread,
More information about the mogilefs