What's up with fastestmirror?

[Patrick O'Callaghan]

On Fri, 2011-12-16 at 22:04 -0200, Itamar Reis Peixoto wrote:
> On Fri, Dec 16, 2011 at 9:59 PM, Patrick O'Callaghan
> <pocallaghan at gmail.com> wrote:
> > I have a slow (1Mbps) connection at home, so I'm keen to reduce yum
> > update times. The fastestmirror plugin ought to be good for someone like
> > me, but I can't discover any rationale for how it's supposed to work.
> > This is not helped by the complete absence of documentation.
> >
> > For example, I explicitly exclude the .br TLD since FM thinks Brazil is
> > close to Venezuela where I live. However networkologically they aren't
> > that close, and downloads from the Brazil mirror(s) general go at 20Kbps
> > or less. So the weighting algorithm is demonstrably wrong. I've actually
> > brought this up before on the list, but I don't remember BZing it. My
> > mistake.
> >
> > Anyway, despite this explicit exclusion, yum insists on trying to
> > download from a Brazil mirror. In fact today it wanted to update json-c
> > and decided to use Brazil, however there was a connection problem so the
> > update failed. Several retries gave the same result.
> >
> > I then disabled fastestmirror completely and tried again. The update
> > completed instantly. So I guess I'll be leaving fastestmirror turned off
> > from now on.
> >
> > poc
> >
> 
> fastestmirror is broken
> 
> it should detect the fast speed downloads, but it tries to find the
> nearest mirror or something else.

I think you're right. The smoking gun is this, from the Python source:

def get_mirrorlist(self):
        """
        This function pings/polls all the mirrors in the list
        C{FastestMirror.mirrorlist} and returns the sorted list of mirrors
        according to the increasing response time of the mirrors.

If that's all it does, it is indeed completely useless. [Insert lecture
about the difference between latency and bandwidth here]. And that's
aside from it apparently ignoring config file setings.

poc