[Yum-devel] Selecting mirrors on the "freshness" of their repomd, not just their speed

[Karanbir Singh]

Willem Riede wrote:
>>> In   
>>> https://www.redhat.com/archives/fedora-test-list/2005-November/msg00610.htmlSeth  
>>> told me that submitting code was the only way to help make the 
>>> subject  happen, so I created a variant of the fastestmirror plugin 
>>> that also checks  the age of repomd.xml on the site.
>>>
>>> A non-up-to-date mirror, however fast, is not going to get the most 
>>> recent   updates to me, so it is of less use than a slightly slower, 
>>> but up-to-date   mirror.
>>>
>>> A full version is attached, a diff (probably mangled by my mail 
>>> client)   follows in-line. It works for me - YMMV.
>>>
>>> Comments? Willem Riede.
>>
>>
>> isnt this the sort of role, better performed at the server end ? The 
>> machine  handing out the mirrorlist should _only_ be giving out the 
>> urls' to mirrors  that are 'in sync'
>>
>> a cron job on the server that runs every 15 minutes should be able to 
>> update  the files handed out at the mirrorlist=<url>. surely that is 
>> going to be  more effective at checking mirrror freshness.
> 
> 
> While in theory, you have a point, there are two practical problems with 
> what  you say:
> 
> 1. I don't see RedHat volunteering to have that process going on their 
> server;

it might be worth asking, i am sure the Redhat / Fedora infrastructure 
guys would be willing to atleast talk about a solution that might 
prevent users hitting out of date mirrors.

> 2. Bestmirror checks all repo-mirrors of your choice, not (just) the 
> ones that  are listed on some server, so it works for all your package 
> sources.

Which is a valid point, but if you look at it - most of the yum repo's 
are publishing mirrorlists now, including RPMForge, Fedora Core, Fedora 
Extras.... There should be no real reason for them to not also run a 
mirror-validating script that can, in turn, generate the mirrorlist file.

The issue that I worry about is that if a few thousand users are going 
to extract the repomd.xml from each mirror, for each repository on the 
users yum.conf - even on a moderately configured setup were looking at 
25 - 30 urlget's.... major traffic ( even at 1K file ) ... and lots of 
'waiting for yum to start' time.

-- 
Karanbir Singh : http://www.karan.org/ : 2522219 at icq