Jeroen van Meeuwen
kanarip at kanarip.com
Mon Mar 17 13:31:02 UTC 2008
Bram Neijt wrote:
> So would I, but for what I've seen the SHA1 information is not part of
> the mirrormanager database.
> For any "good" implementation of metalinks in the mirrormanager, I
> would at least want trustworthy SHA1 information in the database
> otherwise the only advantage of using metalinks would be load
> balancing on your mirrors (which is already mostly fixed with MM).
> So for the end user, there is not much to gain if metalinks is
> implemented in the mirrormanager. The only solution for the SHA1
> information I could think of was to implement a script on the main
> mirror which hosts the SHA1 files, which would mean not running on the
> server with the MM database, which would mean not having direct access
> to that database.
> Parsing HTML is pretty stupid, I agree, but there is probably a much
> cleaner way of getting the mirror list (I havn't worked with turbo
> gears before, so I couldn't find a view of the list I would want).
> Something like fedora-test-data/mirror-hosts-core.txt would be much
> Rene is still working on the MM patch, but without the SHA1
> information, I'm not really sure it would be _my_ ideal solution.
How about using:
instead of the HTML?
Incorporating that in the .metalink, trusting Mirrormanager with
managing whether that file is actually the file that is supposed to be
there, and the metalink client doing a check-up on the chunks it has
downloads and see if they match? Creating a metalink would then consist of:
- getting the iso
- verifying the sha1sum (so you don't start off on the wrong foot)
- creating chunks and their sha1sum for use in a .metalink file
- while/before the .metalink is being downloaded by a client,
incorporate the mirrormanager results and trust mirrormanager to return
the 1) right mirrors, 2) the right files, 3) the files with correct content.
- optionally allow the .metalink to be created on-the-fly with all
parameters mirrormanager has (&country=NL,DE,BE, but not &redirect=1)
Does that sound like a good idea?
Just my brain dump ;p
Jeroen van Meeuwen
More information about the infrastructure