On Wed, 2008-11-05 at 09:49 -0600, Jason L Tibbitts III wrote:
>>>>> "MB" == Mike Bonnet
<mikeb(a)redhat.com> writes:
MB> Koji builders have never downloaded packages via XMLRPC. All
MB> downloading is done by mock/yum, via http (previously nfs).
Well, mock can cache all sorts of things these days. If there are
multiple builders at one location then having a single squid cache for
them all might be nice, but mock's caching would still help to avoid
having to hit the network.
Actually mock's caching doesn't really help us. It's all done
per-buildroot, and since every build is run in a different buildroot,
the caches would never be reused. For this reason Koji disables caching
in the mock configs it writes out.
A global (per-machine) rpm cache might be useful for reducing network
bandwidth. However, because mock/yum would have to lock this global
cache while interacting with it, it would become a bottleneck when
running concurrent builds. The best approach currently is probably a
local squid cache.