[RFC] Increase default ulimit nofile

Jon Masters jcm at redhat.com
Thu May 21 16:15:27 UTC 2009


On Thu, 2009-05-21 at 10:22 -0500, Chris Adams wrote:
> Once upon a time, Jon Masters <jcm at redhat.com> said:
> > On Thu, 2009-05-21 at 15:14 +0200, Lennart Poettering wrote:
> > > I am pretty sure this is an fd leak somewhere in totem.
> > 
> > Not sure yet - it's possible. However, let's bear in mind regardless of
> > this that 1024 is /way/ the heck too low when we allow a user to spawn
> > 1024 /processes/. And they only get one file each?!!! We should have,
> > for example, 8 x tasks, which is where 8192 came from.
> 
> That is a per-process limit, not per-user.

Late night, you're right. Strangely, I remembered where in fs.h it's
defined but somehow forgot it was per task, and not global. Anyway,
after having had some coffee I am willing to concede the only remaining
bug here is in that ugly totem code.

> Also, if something is leaking up to 1024, and you raise the limit, who's
> to say it won't leak right up to 8192 next?

Yeah. And I guess 1M nr_open gels well with 1024 per task. Makes sense
when you think about it. In any case, the totem code also needs fixing
to actually handle the error properly. I'll send them a bunch of debug
info, and let them poke at it.

Jon.





More information about the fedora-devel-list mailing list