Re: ftp.br MD5Sum mismatch

[Raphael Geissert <geissert@debian.org>]

Carlos Carvalho wrote:
> Raphael Geissert (geissert@debian.org) wrote on 19 December 2012 14:17:
>  >The redirector already disables mirrors that reach the second stage of
>  >the sync
> 
> Sigh... It should not. First, the interval where the mirror is
> incoherent is much smaller.

There are two things here:
* Not all mirrors have a small window of inconsistency during the second 
stage
* The redirector needs to know what a mirror is updating to. It can't just 
assume that it will end up updating to the latest copy of the archive.

> Second, mirrors are not obliged to
> distinguish the two phases so it's unfair to use it at the redirector.

For the time being, it is a requirement for their use by the redirector.

I plan to make some changes to the code in the future so that it is not 
required under certain cases.

> Furthermore, the redirector sees a different moment than the client,
> so the client may still have a mismatch. The role of the redirector is
> only to filter mirrors that are not accessible or not sufficiently
> recent.

That's why it attempts to detect the beginning of the second stage as soon 
as possible. Yes, this is still error-prone, but it's better than nothing 
when you have multiple mirrors that are of equal eligibility.
If mirrors were not temporarily disabled, as soon as one of them finishes 
the second stage, without the redirector noticing it, it would make the 
user's view inconsistent.

> There's no way to avoid the issue outside apt.

APT definitely needs to be more fault-tolerant. However, it is not the only 
client out there and as such the redirector compensates by attempting to 
provide a consistent view.

Cheers,
-- 
Raphael Geissert - Debian Developer
www.debian.org - get.debian.net