[OpenAFS] 'replication'

[Rubino Geiß]

Hi,

this is exactly what we want. Doing something like "vos dump | vos
restore" cost far too much time, I guess.  My concern is to get our
systems productive in case a major failure make one of the main servers
unusable. dump / restore is usable in case one volume get accidentally
killed / deleted, but not for restoring >> 100 GB and hundreds of
volumes...

Hartmut, can you publish the changes you made to MR-AFS so we can
integrate it into OpenAFS? Or does someone did this??? I'm sure a lot of
people would like to get this working.


Bye, ruby

> > Anyway, does anyone know a setup or trick to make a RO volume a RW 
> > volume?
> > 
> > Some time ago there was a reference to some "admin hack" or 
> something 
> > like that on this list, but nothing more. I'd like to 
> regularly do vos 
> > releases on a set of volumes and in case of disaster switch 
> to the RO 
> > as RW volumes... is possible or do we have to "hack" 
> openAFS for that? 
> > -- I know backups do the trick (of some kind of disaster 
> recovery; and 
> > we do daily  backups), but restoring > 100GB will cause significant 
> > outage!
> 
> We have a similar backup strategy: the user's home volumes are 
> replicated in their own partition and into a partition on 
> another cheap 
> big server. "vos release" runs each night.
> 
> Whenever the original partition should break I will do a
> "vos convert <otherserver> <partition> <volume>"
> which converts the RO into a RW-volume.
> 
> This "vos convert" command, however, is an extension I wrote 
> for MR-AFS, 
> but suppose it could easily be implemented for OpenAFS with 
> NAMEI-interface as well. practically it only does things:
> 
> 1) it checks the vldb and prompts for a yes if the RW-volume 
> exists at 
> another place.
> 2) it checks only one volume is in the volume group on this partition.
> 3) it renames the large and small vnode-files in the special 
> directory 
> and creates a new volinfo special file with the correct 
> volume id and name.
> 4.) it removes the old volinfo file and volume header file 
> and create a 
> new volume header-file.
> 5.) it updates the vldb-entry to point to this RW.
> 
> This command is pretty fast. You could regain the volumes of a 200 GB 
> partition in half an hour. Much faster than to restore the 
> contents from 
> tapes.