[packman] Build host memory limits

[Hendrik Vogelsang]

On 25.07.2013 11:21, Guido Berhoerster wrote:
> * Stefan Botter <jsj at jsj.dyndns.org> [2013-07-25 10:58]:
>> Henne, my man ;-)
>>
>> On Thursday, July 25, 2013 10:24:31 AM Hendrik Vogelsang wrote:
>>> On 24.07.2013 12:23, Stefan Botter wrote:
>> :
>>>> I am not really sure, how the contraints management in OBS is
>>>> working (regarding the workers). As usual the documentation is
>>>> sparse at best. I do not know how the dispatcher gets to know of
>>>> the capabilities of workers, and if this is a feature in newer
>>>> versions of the workers only.
>>>
>>> The hardware get's analyzed every time a build is run (see
>>> $workerenvstate in bs_worker) and with the _constraints file you can
>>> match against that.
>>>
>>>> My workers "should" provide the necessary resources, at least
>>>> swkj04 and swkj07, the workers run in chroot mode and have thus,
>>>> at least potentially, access to the whole system resources, as
>>>> long as the other processed do not fill up their (RAM)disks to the
>>>> max.
>>>
>>> In the same function you'll find the explanation. memory is undefined
>>> in non vm mode since no guarantees exist :-)
>>
>> So having the _contraints in an all-chroot env is worthless, or better 
>> "buildless" ;-)
> 
> That's not good, it means the build success of such packages
> essentially depends on the build host's current load. That's what
> I've been experiencing before the recent outage where I needed to
> manually trigger builds that take ~ 40min until they fail during
> linking numerous times until they succeeded.

Yeah we know but there really isn't any way to predict what resources
will be available in a chroot :-/

Henne

-- 
Henne Vogelsang
http://hennevogel.de
"To die. In the rain. Alone."
                   Ernest Hemingway