dwmw2 asked me if I could do my full rawhide rebuilds on PPC as well as x86_64 and i386. I don't have ppc hardware, and he doesn't have ppc hardware with enough disk space or a local rawhide mirror. :-)
As it's just some scripts that call mock, could I run those on the ppc Extras builders in PHX? They use --uniqueext so they can't conflict with a normal Plague build job. Would need some disk space for the results directory (~20GB), and be able to export the results via a web server.
I'd still do the i386 and x86_64 builds in my lab here as I have been.
Thoughts?
Thanks, Matt
I've been out of it for a few months, but it looks like I should have some more time to devote to FI. I'm digging through my e-mail to find out what's been going on, looks like I've missed a lot. ;-)
Ryan
-- Ryan Ordway Unix Systems Administrator OSU Libraries E-mail: ryan.ordway@oregonstate.edu 121 The Valley Library Corvallis, OR 97331 Office: VLIB #4657
On 1/30/07, Ordway, Ryan Ryan.Ordway@oregonstate.edu wrote:
I've been out of it for a few months, but it looks like I should have some more time to devote to FI. I'm digging through my e-mail to find out what's been going on, looks like I've missed a lot. ;-)
Welcome back Ryah!
-Mike
On Jan 30, 2007, at 15:05, Matt Domsch wrote:
What kind of maintenance does the build system require? I have a currently unused Xserve G4 that I could install Linux ppc on.
- ask
On Tue, Jan 30, 2007 at 05:44:26PM -0800, Ask Bj?rn Hansen wrote:
On Jan 30, 2007, at 15:05, Matt Domsch wrote:
What kind of maintenance does the build system require? I have a currently unused Xserve G4 that I could install Linux ppc on.
It took just a little tweaking after not touching it since FC6's release. Mostly code cleanups to deal with the fact that my mock config files are named 'fedora-devel-i386-core.cfg' yet they cause directories named 'fedora-development-i386-core' to get created in various places. :-) The whole system is just a set of shell scripts that essentially do: for each SRPM in rawhide: setarch i386 mock ... mock .... if successful, generate some logs generate stats rsync to a web server send email
I tried to abstract out things like "directory where SRPMS live" and "what arches should I build for" into shell variables, but I very much expect to have the whole rawhide tree NFS-mounted. Would take some work to use remote repos instead. Something like 20 invocations of ls or find on various parts at various times.
It took ~60 hours to build core+extras, i386 and x86_64, on 4 dual-socket dual-core servers with 2-4GB RAM each and 2.8-3.6GHz Xeons. Would presumably take nearly a week on a single ppc at first, and a few hours the daily runs which are just incremental.
Matt Domsch (Matt_Domsch@dell.com) said:
dwmw2 asked me if I could do my full rawhide rebuilds on PPC as well as x86_64 and i386. I don't have ppc hardware, and he doesn't have ppc hardware with enough disk space or a local rawhide mirror. :-)
As it's just some scripts that call mock, could I run those on the ppc Extras builders in PHX? They use --uniqueext so they can't conflict with a normal Plague build job. Would need some disk space for the results directory (~20GB), and be able to export the results via a web server.
My concern would be overloading the Extras builders, but I suppose we wouldn't know how bad that would be unless we tried.
Bill
Once upon a time Tuesday 30 January 2007 5:05 pm, Matt Domsch wrote:
dwmw2 asked me if I could do my full rawhide rebuilds on PPC as well as x86_64 and i386. I don't have ppc hardware, and he doesn't have ppc hardware with enough disk space or a local rawhide mirror. :-)
As it's just some scripts that call mock, could I run those on the ppc Extras builders in PHX? They use --uniqueext so they can't conflict with a normal Plague build job. Would need some disk space for the results directory (~20GB), and be able to export the results via a web server.
I'd still do the i386 and x86_64 builds in my lab here as I have been.
Thoughts?
personally i would pefer to take one builder offline and let you use it. probably either ppc2 or 3 they have a little more ram than ppc1 all three of them are 4 cpu boxes.
plague uses --uniqueext with a job id we would need to set up the proxy to serve up the data.
Dennis
On 1/30/07, Dennis Gilmore dennis@ausil.us wrote:
personally i would pefer to take one builder offline and let you use it. probably either ppc2 or 3 they have a little more ram than ppc1 all three of them are 4 cpu boxes.
plague uses --uniqueext with a job id we would need to set up the proxy to serve up the data.
How many plague instances do we allow to be built at once on the ppc boxes? Would it make sense to increase it?
-Mike
Once upon a time Tuesday 30 January 2007 8:53 pm, Mike McGrath wrote:
On 1/30/07, Dennis Gilmore dennis@ausil.us wrote:
personally i would pefer to take one builder offline and let you use it. probably either ppc2 or 3 they have a little more ram than ppc1 all three of them are 4 cpu boxes.
plague uses --uniqueext with a job id we would need to set up the proxy to serve up the data.
How many plague instances do we allow to be built at once on the ppc boxes? Would it make sense to increase it?
one plague instance per system. each one is configured to build one package at a time per cpu so the ppc builders will do 4 at a time and the x86_64/i386 builders will do 2 at a time i dont think it makes any sense to do more than that.
Dennis
infrastructure@lists.fedoraproject.org
-
Ask Bjørn Hansen -
Bill Nottingham -
Dennis Gilmore -
Matt Domsch -
Mike McGrath -
Ordway, Ryan