did anything change in F18+ building stack which would increase the requirements?

[Julian Sikorski]

W dniu 25.11.2012 20:44, Panu Matilainen pisze:
> On 11/25/2012 09:12 PM, Julian Sikorski wrote:
>> W dniu 2012-11-24 15:05, Julian Sikorski pisze:
>>> W dniu 24.11.2012 14:37, Panu Matilainen pisze:
>>>> On 11/23/2012 09:19 PM, Julian Sikorski wrote:
>>>>> W dniu 22.11.2012 07:55, Panu Matilainen pisze:
>>>>>> On 11/22/2012 12:03 AM, Julian Sikorski wrote:
>>>>>>> W dniu 10.11.2012 10:51, Julian Sikorski pisze:
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> I was wondering if it was possible to monitor the maximum and/or
>>>>>>>> average
>>>>>>>> memory usage of a mock build process. I am trying to investigate
>>>>>>>> why a
>>>>>>>> package takes less 2 hours to build on F16/F17, 24 hours on F18
>>>>>>>> and 7
>>>>>>>> hours on rawhide:
>>>>>>>> https://bugzilla.rpmfusion.org/show_bug.cgi?id=2554
>>>>>>>>
>>>>>>>> Regards,
>>>>>>>> Julian
>>>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>> I would like to follow up on this issue. I did not notice the
>>>>>>> differences when mock building locally, but on the plague RPM Fusion
>>>>>>> builder it took 2 hours on F16/F17 and 30 (sic!) on F18 to build
>>>>>>> mame.
>>>>>>> Suspecting memory usage issue I already added
>>>>>>>
>>>>>>> %global optflags %{optflags} -Wl,--no-keep-memory
>>>>>>> -Wl,--reduce-memory-overheads
>>>>>>>
>>>>>>> to the spec file. Thank you for the input in advance.
>>>>>>
>>>>>> Here's one difference between F18 and earlier that does affect memory
>>>>>> usage:
>>>>>> https://fedoraproject.org/wiki/Features/DwarfCompressor
>>>>>> See https://bugzilla.redhat.com/show_bug.cgi?id=833311 and
>>>>>> /etc/rpm/macros.dwz for further info on memory consumption and
>>>>>> tunables.
>>>>>>
>>>>>> But then this is the same in rawhide and f18 so it doesn't explain
>>>>>> why
>>>>>> the build takes *that* much longer on f18.
>>>>>>
>>>>>>       - Panu -
>>>>> Is there a macro to disable this from within the spec so that I can
>>>>> check whether dwz is the culprit?
>>>>
>>>> This should disable dwz entirely:
>>>>
>>>> %global _find_debuginfo_dwz_opts %{nil}
>>>>
>>>>      - Panu -
>>>>
>>>>      - Panu -
>>> Thank you. I have submitted a build with dwz disabled, let's see how
>>> long it takes.
>>>
>>> Julian
>>>
>> That was it, build with dwz disabled took 2 hours as f17 and f16 ones.
>> Do you think it would make sense to look into changing the default
>> setting, or should we (at RPM Fusion) rather look into giving the
>> builder more memory?
> 
> You should talk to the feature owner (cc'ed).
> 
> Given how dramatic the difference can be, perhaps dwz should be looking
> at eg percentage of actually available memory instead of using absolute
> values which are always going to be wrong from somebody.
> 
> Anyway, in the worst case you can just override the limits to something
> more appropriate for the rpmfusion builder in local config.
> 
>     - Panu -
> 
OK, we'll see what we can do. I checked locally and dwz tops at 2.9 GB
for the x86_64 build. This is not overly dramatic, but i686 and x86_64
builds running concurrently are causing the builder to swap itself to
death atm.

Regards,
Julian