Re: Too few up-to-date CD image mirrors
On Fri, 29 Nov 2002, Joey Hess wrote:
> Mattias Wadenstein wrote:
> > Yes, this would be very good. I have been thinking of writing such a
> > script myself, but never found the time for it. I would be very grateful
> > if you could write such a script.
> I have one sorta kinda working. I'll pound on it a bit more before
> releasing it. The overall inconsistency makes it hard, since right now
> mirrors use a variety of directory structures and names of files (and
> even do things like symlink the non-us cd image #1 to image #1).
Well, that thing has been more or less recommended on this list as I
understand it for sites without the restrictions that apply to the non-us
images. I just carry both versions though, less hassle that way.
> So my script will not find every mirror that has isos, but only those
> that have them in the semi-standard structure of:
Yes, that should be good enough.
> Just to give an idea (jigdo checking is probably still not working):
Well, looks good. Perhaps extrating the number of the latest version
instead of just the binary check of "has 3.0_r0"? And perhaps a check of
incomplete sets (just i386). Or perhaps those are the RFEs for version
> > > I will not try to check the http mirrors though. Indeed, I think the
> > > http mirrors are a bad idea, since (as is seen in my mistake reading one
> > > of them in the gentoo thread), they introduce a bunch of third party web
> > > sites of varying quality that users must navigate to find isos. I hope
> > > the http stuff is not a necessary evil.
> > For us http is just a better ftp, without the "logging in as anonymous"
> > part. The standard apache file listing is good enough.
> I've never seen a web browser that couldn't do an anonymous ftp login
> without the user even noticing it went on.
But it takes time, has additional protocol overhead and puts more load on
the server. Ok, for isos it probably doesn't matter much, but there is a
noticable difference in browsing, apt-geting or just downloading. I'm just
talking about less protocol overhead.
> And not all of the http mirrors use standard apache dir listing; some
> use crazy and broken dir listing that ends up looking like this in a web
> i386 -> jigdo
> And that confused me, anyway, to no end (isos are actually hidden under
> the i386 link; jigdo is not a subdirectory of it).
Sounds like something that should be reported to the mirror admin.
> Anyway, parsing http directory listings is too hard, so my script will
> only do ftp.
> Yes, the current situation is crazy, and I just cannot understand why
> cdimage.d.o has not been redirected to one of the mirrors that already has
> everything, long since.
I wish I had the confidence in our server to nominate it, but I don't, it
has a hard enough time as it is with the current load.