On (03/09/14 16:06), Nikolai Kondrashov wrote:
On 09/03/2014 03:19 PM, Lukas Slebodnik wrote:
>On (03/09/14 07:55), Nikolai Kondrashov wrote:
>>Hi everyone,
>>
>>As CI mock builds are taking a god-awful long time sometimes and Clang
doesn't
>mock build does not take "god-awful long".
It does sometimes. Take a look at this:
http://sssd-ci.duckdns.org/logs/job/0/19/rhel7/ci.html
http://sssd-ci.duckdns.org/logs/job/0/21/rhel7/ci.html
I am not sure what is wrong.
I have 4 CPUs on my machine and rpms are already cached in /var/cache/mock/
I don't have SSD.
[build@host mock]$ time mock --root epel-6-x86_64-ding-libs --resultdir
. --rebuild sssd-1.12.1-0.el6.src.rpm
//snip
real 1m50.537s
user 1m54.073s
sys 0m15.984s
and with empty cache
[build@host mock]$ time mock --root epel-6-x86_64-ding-libs --resultdir
. --rebuild sssd-1.12.1-0.el6.src.rpm
//snip
real 6m55.672s
user 4m22.671s
sys 0m27.573s
Depends on the VM host load. Which probably spikes if we start several
mock
builds at the same time. That's with uncoditional caching, though.
>The londest part is downloading rpms to cache on firs time or after
>modification of mock config.
Yes, caching can take a while, which my other CI patch tries to avoid.
>I have few tips for you how you can speed up this part.
>At the moment, you are using mock on for building in fedora-20 chroot
>and rhel6 chroot. Even on rhel6, fedora21, fedora-rawhide.
>So you will reduce two mock build into one.
You don't exactly provide a tip here, but I'll assume that you meant some
hosts don't need to run mock builds. However, the whole CI run will be as long
as the longest host run, which will be one of those which include mock builds,
so this doesn't exactly speed up getting results.
I meant to call mock just one on each distro
epel6 mock --root epel-6-$ARCH ...
epel7 mock --root epel-7-$ARCH ...
fedora20 mock --root fedora-20-$ARCH ...
fedora20 mock --root fedora-21-$ARCH ...
fedora-rawhide mock --root fedora-rawhide-$ARCH ...
LS