Re: non-automated test failure report! :)

[christos%zoulas.com@localhost (Christos Zoulas)]

On Nov 16,  9:40am, njoly%pasteur.fr@localhost (Nicolas Joly) wrote:
-- Subject: Re: non-automated test failure report!  :)

| On Tue, Nov 15, 2011 at 06:20:10PM -0500, Christos Zoulas wrote:
| > On Nov 15, 11:36pm, njoly%pasteur.fr@localhost (Nicolas Joly) wrote:
| > -- Subject: Re: non-automated test failure report!  :)
| [...]
| > | > Either limits don't work... which they seem not to...
| > | 
| > | I had the need to limit user processes memory to not let them crash
| > | the machines. Unfortunately neither RLIMIT_DATA and RLIMIT_RSS worked.
| > | Checking the kernel sources showed that we can set/get values, but are
| > | unused otherwise ... except RLIMIT_DATA in sys_obreak().
| > 
| > You need both RLIMIT_AS and RLIMIT_DATA for malloc since some of them use
| > both sbrk() and mmap. What I was complaining was that RLIMIT_RSS seems to
| > be non-functional and I just grepped the source and it is :-)
| 
| Right. But limiting address space will prevent mmap on huge files or
| big number of medium sized ones (we do have such cases @work).
| 
| I wonder if anon mmap sizes should be taken into account for process
| data size too ... they are mostly malloc in disguise.

I agree, I think that anon vs mapped space should be counted differently.

christos