On Thu, Sep 1, 2011 at 5:11 PM, <Matt_Domsch(a)dell.com> wrote:
Quick show of hands: who cares about seeing OpenStack be in EPEL?
raises both his personal and $dayjob hands; at least for swift. (And
yes, I know, that's fully within my power to solve)
I’ve been building on the excellent work that Mirantis, then Mark McLoughlin
has been doing to get OpenStack components built in rawhide, and maybe
(cross your fingers) F16. The same packages, with small tweaks, also work
on RHEL.
Status here:
https://fedoraproject.org/wiki/OpenStack#OpenStack_in_EPEL
Now, before I go any further: who is with me? Is there pent-up user
demand for OpenStack on RHEL, or service providers who are loyal to the RHEL
family who would benefit from having the packages in EPEL? Or do we expect
most OpenStack users to simply use Ubuntu/Canonical or a Citrix distribution
thereof? I don’t want to push packages that are likely to go unused.
So I have had infrastructure over the years compare public binary
downloads of f/loss software compared with parsing logs on some of our
mirrors, and my rough, anecdotal, numbers are essentially that:
There are 3x the number of yum install foo in Fedora for a given
period that there are downloads from that projects public download
repository (for all distros/platforms). That strikes me as a
compelling argument.
My concerns are: OpenStack is currently moving at an incredible pace,
6 months or 12 months from now swift and nova are going to be markedly
different, which EPEL seems ill suited to deal with, and I'd hate to
still be seeing swift 1.4.0 in EPEL in 12 months. I am also concerned
about dependency issues - I'd hate to see us have to stay with an
ancient version because they began depending upon an upstream package
substantially newer that what is in RHEL (or EPEL).