Debian Gump (was: Are regression/unit test suites being run on upstream sources?)

Leo Simons mail at leosimons.com
Sun Aug 7 08:14:14 UTC 2005


On 06-08-2005 16:16, "Mark Wielaard" <mark at klomp.org> wrote:
> On Sat, 2005-08-06 at 14:57 +0200, Arnaud Vandyck wrote:
>>> If there is anybody that wants to create a Gump setup
>>> on a Debian GNU/Linux (x86) using kaffe or gcj that would be a huge help
>>> since testing and comparing results on such a setup is a lot easier for
>>> most (upstream) hackers (since a lot do actually run Debian).
>> 
>> That would be great. Mark, do you know what is the load of such a
>> configuration? Network, CPU, Mem etc so we can ask the Debian project to
>> host such a configuration. And maybe on several different architectures.
>> 
>> I'm willing to help in this direction.

Cool!

> Thanks. Unfortunately I still haven't setup Gump myself.
> I have CCed Leo Simons which has been helping a lot with the kaffe setup
> and gave a presentation on Gump at Fosdem. Hopefully he has some raw
> estimates on resource numbers.

Most important here is that you get a dedicated machine, or at least a
machine that's not doing anything else, or a good partitioning system (we
run inside a solaris zone at apache for example). Gump tries pretty hard to
eat up all CPU resources. There's also security issues to ponder, but I'm
sure the debian buildfarm maintainers know more about those than me :-)

The next bottleneck is then basically what resources you need to build your
projects plus a few percent (for memory, cpu, disk space). And if you want
them to build fast (which is nice as the integration becomes more
"continuous"), that means more resources. But gump will probably run on an
old 386 if it has to (I have no 386 so I can't test ;)).

> General setup instructions are at:
> http://gump.apache.org/gettingstarted.html

I'm afraid those are likely to be a little dated. However, we maintain the
installation instructions for our main gump install at

  http://wiki.apache.org/gump/VmgumpConfig

Which runs debian sid, so they should be rather easy to mirror.

> We also have to come up with a free workspace for gump. The default has
> some non-free binary dependencies. One interesting idea would be to have
> gump depend on the actual unstable/experimental Debian packages
> themselves (but building them from source of course). Gump normally uses
> current CVS sources. But for Debian it might be more important to get a
> current state of Debian testing/unstable/experimental overview using the
> Gump. Since that is what your users are going the got in the end.

Interesting indeed. It might in fact be desireable for gump to fire off the
debian package management tools (I think they're called things like makepkg
and lots of stuff with 'deb' in the name? -- I just use checkinstall :-)).
Note that you're not really doing "continuous integration" as per Gump's
definition anymore then, but its definitely possible to set up.


Cheers!


Leo

PS: it'd be great if the CC to me could be replaced with a CC to
general at gump.apache.org, as there's no doubt other gump people that will be
willing to help, and I might not have a lot of time for that next week.





More information about the pkg-java-maintainers mailing list