On Sun, Aug 29, 2004 at 12:13:10PM -0700, Marc Wilson wrote: > On Sun, Aug 29, 2004 at 12:02:20AM -0700, Steve Langasek wrote: > > On Sat, Aug 28, 2004 at 07:08:05PM -0700, Marc Wilson wrote: > > > On Fri, Aug 27, 2004 at 11:41:26AM +0800, Dan Jacobson wrote: > > > > Isn't it nuts that the Packages files get to the mirrors before the > > > > .debs do? > > > Not particularly. I can see you think it's inconvenient for YOU. > > You don't think our mirrors being broken for 3 or more hours every day > > is a problem? How in the world could this be an ok situation? > I think that it would be a horrific problem if it affected a stable > release. Does it? No. Does it affect a stable release at all? No. Sure it does; this bug affects stable every time there's a point release. Though my gut feeling is that the majority of downloads from the Debian mirrors pull from testing/unstable anyway -- not by volume of users, but because testing/unstable users have systems needing much more frequent updates. Problems should not be dismissed just because they only affect testing and unstable. > Does it affect a significant enough percentage of the packages in testing > or unstable for it to be a serious problem? No. Sure it is; the percentage of users affected by the problem is not directly proportional to the percentage of packages updated on a given day. > What about the reverse situation? If you update the Packages file last, > then the *user* has a Packages file that may not match the contents of the > mirror, rather than the mirror having one that doesn't match. And now he > STILL can't install anything, and the mirror is claiming that the > out-of-date Packages file he has is the most current one. You clearly didn't read the parts of this thread where people discussed correct, two-pass methods of mirroring the archive that ensure the files available on all mirrors are self-consistent at all times.[1] > Use of testing or unstable implies a certain level of intelligence. Running a mirror does, too. Please stop defending broken mirror implementations. The only thing this does is ensure that users will switch to a different mirror. -- Steve Langasek postmodern programmer [1] Technically, to have this true at *all* times requires three passes, but two passes gets close enough that the period of unusability is negligible.
Attachment:
signature.asc
Description: Digital signature