Re: apt, mirror redirection systems and error reporting.
Hi Peter,
On 22 April 2014 13:32, peter green <plugwash@p10link.net> wrote:
[...]
> When this happens apt's current behaviour sucks. It reports the error with
> the original url it tried to retrieve and no indication that a redirect ever
> happened at all. This makes troubleshooting user complaints a PITA
> (difficult to tell the difference between "user forgot to run apt-get
> update" and "a mirror is broken" and difficult to determine which miror is
> broken in the latter case). It would be very useful to have better error
> reporting.
I've looked into that and AFAIR it was rather complex to push the
redirection information down the call stack to improve the error
reporting.
> Mirrorbrain also lists a small selection of other mirrors (in addition to
> the mirror in the location: header) in the http response (don't remember the
> header name offhand but I can find out if anyone is interested). Apt could
> be extended to use such information to request files from alternative
> mirrors if the first one fails. Not sure how the developers would feel about
> that.
Yes, mirrorbrain added support for RFC6249 after http-redirector. My
plan was, and still is, to add support for it in apt - some fault
tolerance is needed in here.
As far as progress goes... I haven't got that far. My WIP is at:
https://github.com/rgeissert/apt/commits/rfc6249
With it you should be able to make apt rewrite URLs locally after it
has received a Link rel=duplicate with a depth > 0. Bigger changes are
needed to allow all that information to be used for failover.
Nowadays http-redirector doesn't send them since they are not used by
anyone, you can only see them through the demo. I can re-add a test
address where I had support for rfc6249 enabled, so that client
support can be tested.
Cheers,
--
Raphael Geissert - Debian Developer
www.debian.org - get.debian.net
Reply to: