On Tue, Aug 27, 2019 at 11:12:58AM +0100, Ian Jackson wrote: > I worried about this kind of skew problem quite early on, when I was > doing the detailed design. I discovered that the archive tended to > return *two* versions sometimes, and that in all my tests it was > possible to retrieve at least one of them from the mirror I was using. > > My design intent was that you ought to get *something*, even if it was > out of date. > > Unfortunately it is hard to debug these transient states but I think > what happened here is that the archive db ceased to mention your > previous version, but your mirror didn't have the new version yet.
That makes sense. > I'm not sure how to fix this. I guess it must be fairly rare since > this is the first such bug report I have received. Do you think you > did anything unusual that might have increased the odds of such an > occurrence ? I don't think so. It's certainly possible that I happened to hit a narrow window between the ftp-master API starting to report the new version and the pool on ftp.debian.org coming up to date; if it works anything at all like Launchpad then I suspect that there's a gap there, since ftp.debian.org is itself a mirror of the internal publishing master. (If this were Launchpad, it would be possible to fall back to fetching files directly from the web service rather than from a mirror, but AFAIK there's no equivalent of that for the Debian archive.) -- Colin Watson [[email protected]]
