[OpenAFS] Partial replication

Turbo Fredriksson turbo@bayour.com
03 Jun 2002 22:46:21 +0200


> > Unfortunately I don't have enough disk to replicate my WHOLE AFS space
> > to these machines. I was thinking that I could just replicate the
> > root.* volumes and the 'user database'...
> 
> So, these additional machines will be db and file servers?

I'm building these machines at home, and I have one machine that's 
to be my 'test platform', basically identical to my live server, so I can
pre-test various issues before going live...

To make sure everything (as much as possible anyway :) works 'right out of
the box' when the two replacements go live, I started to setup a AFS system
on 'tuzjfi' (the test machine) with 'rmgztk' and 'morwen' (which is to be
my two backup/replacement machines) close by...


I had some problems getting tuzjfi up and running (AFS vice, everything else
worked fine) like 'papadoc' (the main server). IP addresses naturally don't
match, user database et all don't match identically... This got me thinking.

Do I really have to replicate the volumes?

I know I _WOULD_ like that, but papadoc have roughly 200Gb of disk, and buying
twice that to be on the replacement is not an option! I just don't have that
kind of money! I'm doing this (all alone, as a 'private citizen' :) because
it's fun, not for profit :) Getting a tape (drive) was enough strain on my
(already bad :) economy...


I got the rough understanding of what the root.* volumes do in a previous mail
(other topic), and from what I understand, it's not that big a loss if I loose
these... They can 'easily' (?) be recreated. As long as I have the database
intact (users etc?), I'm "good to go"... ?

> They  don't necessarily  have to  be both.   Since you're  not using
> kaserver, you'd  probably want to  run the vlserver and  ptserver on
> them.  Then  running the  file server too,  you could  replicate any
> volumes you feel are critical.

Do I _have to_ (run the file/volume servers) on these machines?


The 'only' point (for the time being) in having these replacements was so
that EVERYTHING is kept intact, and can 'easily' (?) be 're-replicated'
to the main server if/when it crashes... Sure, the service will be out
while the repair of papadoc is in progress, but...


In the end, when my economy permits, I _WILL_ add hard disks to these
replacements and have them as 'secondaries' (?) for AFS...


Morwen and Rmgztk will be the MAIN (and only) Kerberos/LDAP server (those
services will be removed from papadoc). By having two, replicated, machines
will make it easy to replace/reinstall one of them, because the database(s)
will still live on the other one, and 're-replicate' to the newly installed
machine (if it have to go that far) is easy. With the help of round robbing
and SRV records, they will even be load balanced...

This is basically only 'live-backups' instead of 'tape/file-backups'.
Would theoretically mean faster 'up-and-running-after-crash' times...

-- 
FSF SEAL Team 6 ammonium terrorist Treasury Iran Ft. Meade president
radar Saddam Hussein critical KGB Peking security Cuba
[See http://www.aclu.org/echelonwatch/index.html for more about this]