Inconsistent mirrors

Tushar Teredesai tushar at linuxfromscratch.org
Thu Mar 25 18:53:02 PST 2004


Gerard Beekmans wrote:

>On Thu, 2004-03-25 at 19:05, Tushar Teredesai wrote:
>  
>
>>Rather than employing the server-pull scenario, how about a client-push 
>>scenario? Each client would have their own private key and whenever the 
>>download the new data, the server would update the time the client 
>>updated. In the round-robin, only include mirrors that have updated in 
>>the last X hours.
>>    
>>
>I considered that, it would make things easier too. But I don't want to
>make that a mirror requirement. Ideally all a mirror should have to do
>is run the rsync script twice an hour. The less mirror maintenance, the
>better IMO. We'll keep and get more mirrors that way.
>
>Maybe I'm not understanding what you meant to suggest. Do you envision
>mirrors scp'ing a file or something to the server signaling they are
>up-to-date?
>
>Either way, even if a timestamp file is present, doesn't mean the entire
>mirror is there. Maybe the sync wasn't completed. Ideally the mirrors
>would compute some md5sum of the tree which we can download too, to make
>sure all files are present and accounted for? Every mirror box will have
>the md5sum program that can be run to run against every file and make
>the file available. It should take minimal processing power to generate
>the file on the mirror, and minimal CPU power to process this on our
>server.
>
>  
>
Each mirror could be given a seperate key to connect (the easiest would 
be providing a user name to connect to the rsync server). Based on the 
data sent, on the server side would put the mirror in the dns list. 
Sounds easier to maintain rather than pinging each server every now and 
then.

-- 
Tushar Teredesai
   http://linuxfromscratch.org/~tushar/
   http://www.geocities.com/tushar/




More information about the website mailing list