Re: Mirror monitor for meta data repos

[Jeff Spaleta <jspaleta gmail com>]

On 6/8/05, david <dfarning sbcglobal net> wrote:
> Concerning the time issue, I was thinking that I was not clear about the
> point that my probes operate in parallel.  Thus, dealing with the
> network latency issue.

Yeah that wasn't clear initially.. but after i saw your time numbers i
assumed it was parallel.
Even in parallel.. if mirrors are full and not accepting connections..
there will be timeouts and several re-connection attempts needed...
this takes finite time. Hopefully you can tell the difference between
a mirror that is full from mirrors that are just plain dead via the
messages the servers return back to your process. Full mirrors that
can't be probed shouldn't be marked in the same way as dead mirrors. 
I have no problem informing users that a mirror is full in the
webpage. But full mirrors shouldnt be removed from the dynamically
created mirrorlist nor should users be encouraged to contact a
mirror-admin if the mirror is full instead of dead and unresponsive.


-jef