On Wed, 21 Nov 2007 08:27:56 -0900 "Jeff Spaleta" <jspaleta@xxxxxxxxx> wrote: > Both Jesse's idea of hardlinked archive and the srpm generation on > demand will help here. But the hardlinked archive isn't going to scale > out. > > 2) Koji is space constrained. We are going to need to flush things > from koji quite frequently. I'm not sure we can rely on koji to be the > point at which we archive anything even srpms. Nor do I think we have > the resources to scale jesse's big pile of archived srpms. His idea > may scale just fine right now, but what happens a year from now when > there are 12 different active spin SIGS, each producing monthly > re-spins? The space needed to house a hardlinked dump of all possible > 'released' srpms is still an unbounded constraint. The cvs and > lookaside space on the other hand are consuming much smaller amounts > of space currently and we can most certainly set a limit on the size > limit for the re-generated srpms cache. > > 3)re-generation of srpms on the fly lets us provide more than the > absolute minimum in required service. We should be aiming to provide > source distribution for everything we've built through koji and has > been in a publicly facing repository.. including rawhide. The big > ball of hardlinked srpms certainly can't give us source distribution > coverage that also includes everything seen in the public rawhide > tree. I think the two efforts can go in tandem, however I would rely upon the hardlinked tree as v2 3a distribution means, and not rely upon on the fly generation to fulfill v2 3b/c. Please avoid 3b/c at all costs. -- Jesse Keating Fedora -- All my bits are free, are yours?
Attachment:
signature.asc
Description: PGP signature
_______________________________________________ fedora-advisory-board mailing list fedora-advisory-board@xxxxxxxxxx http://www.redhat.com/mailman/listinfo/fedora-advisory-board
