[OpenAFS] migrating / updating afs

[Paul Blackburn]

Hello Forrest,

One of the benefits of running multiple servers in a cell
is that you can juggle your AFS data around thus "freeing up"
an old server so that you can do maintenance with no disruption
to cell users.

I have done this many times in cells where we had 3 dedicated database
and separate dedicated fileservers.

There is a small impact (users may see some delay in authenticating)
if you take down 1 of 3 dedicated fileservers for re-installing.

With dedicated fileservers, it's possible to move all the AFS volumes
from "old" fileserver to new "standby" fileserver, upgrade "old"
fileserver and continue. No outages.

AFS is sweet.
--
cheers
paul                              http://acm.org/~mpb

forrest whitcher wrote:

>I'm about to take an aged afs fileserver (rh linux 6 on kernel 2.2.20/openafs 
>1.0.3) offline and have a few options about how to replace it.
>
>I plan for this system to be restored after installing the NSA SE-Linux kernel
>which will almost certainly involve some pain & testing, and probably more 
>downtime than I'm willing / able to accept.
>
>Viable options include sharing the current data over NFS for a brief period or 
>feilding a temporary fileserver to continue afs operations while I debug/
>redeploy the primary.
>
>Objectives / options:
>
>	also update the database server openafs 1.0.3 -> 1.2.7
>	minimize downtime
>	preserve acls if possible 
>		(I could restore the acl structure manually in ~2 hours)
>	consider rebuilding the cell from scratch
>
>presently the database server also houses the root volumes and a backup:
>
>Total number of volumes on server mozart partition /vicepa: 2 
>root.afs                          536870912 RW         29 K On-line
>root.cell                         536870915 RW         12 K On-line
>
>Total volumes onLine 2 ; Total volumes offLine 0 ; Total busy 0
>
>Total number of volumes on server mozart partition /vicepb: 2 
>root.afs.readonly                 536870913 RO         29 K On-line
>root.cell.readonly                536870916 RO         10 K On-line
>
>
>
>The server that's about to be taken out of service has all the active 
>data volumes and the following RO copies of the root volumes:
>
>Total number of volumes on server fileserver partition /vicepa: 2 
>root.afs.readonly                 536870913 RO         29 K On-line
>root.cell.readonly                536870916 RO         10 K On-line
>
>
>
>My plan is roughly:
>
>make tar backups of /afs/<cell>/* as a last-resort
>
>update openafs version on the database server
>install a new fileserver and migrate the current data volumes to it
>take the current fs offline and commence rebuild.
>migrate data back to original server
>
>
>Questions
>
>Is it possible (advisable) to make a raw copy of a volume and drop
>it in place into either the same server with a new afs version or
>into a secondary server? (I only think of this because it was how
>I was forced to preserve the the root volume(s) data on the database
>server after a total loss of the OS and recovery from tape backup)
>
>What's the preferred backup / restore method: 
>
>	if I choose to down the whole cell for awhile to rebuild or 
>	if I keep the cell 'live' and migrate to a temporary fs
>
>Thanks for any feedback / suggestions!
>
>forrest
>_______________________________________________
>OpenAFS-info mailing list
>OpenAFS-info@openafs.org
>https://lists.openafs.org/mailman/listinfo/openafs-info
>  
>