[OpenAFS] Small suggestion for RH spec file? (keeping CellServDB up-to-date on *nix clients)

[Paul Blackburn]

Ken Hornstein wrote:

>>Well, no. See that's the whole point of using a crontab process to 
>>synchronise a local CellServDB.
>>It's automagical robot-tastical.
>>It just works.
>>    
>>
>
>You're assuming that the source of that file synced by cron is up-to-date.
>Many times it is not.
>
>It's been my experience that when I tell someone, "Hey, you can get this
>file out of our AFS cell", about 50% of the time their CellServDB has
>our old database server IP addresses ... the ones from 10 years ago.
>In one case this even happened with someone who is now a regular OpenAFS
>contributor.  In many cases the CellServDB that these sites used were
>automatically synced up by cron ... but the source file was out of date,
>and the admins who controlled it were too busy/lethargic to update it.
>
>If I was the OpenAFS dictator, I don't think I would distribute a
>CellServDB file _at all_, and simply tell sites that they should
>publish AFSDB records if they want to make their cells globally
>available.  But that's just me.
>
>--Ken
>  
>
Ken,

Is it easier or harder to maintain individual client CellServDB files on 
every client
or one /afs/@cell/common/CellServDB file per site?

Even if some AFS cell admins are lazy and lethargic, having one "master" 
CellServDB per cell
is (IMHO) a better situation than relying on individual owners of 
individual client machines
to keep their own CellServDB up-to-date.

AFSDB records in DNS is interesting but really only suitable for sites 
who _want_
their cell to be publically accessible. Also, there are situations where 
AFSDB records
would be no use at all such as in DMZs where AFS is providing shared 
filespace
to webservers and the AFS cell is deliberately blocked from Internet access.
--
cheers
paul                                            http://acm.org/~mpb