Coda File System

Re: push volumes to non-scm server

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Mon, 21 Jul 2008 11:58:23 -0400
On Sat, Jul 19, 2008 at 01:25:43AM -0600, Yves Dorfsman wrote:
> If you setup a non-scm server after the fact (after your volumes have 
> been initialised, used, etc...), what do you need to make the non-scm 
> server copy the files from the scm server ?

There is no automatic/reliable way to change the replication factor of
existing volumes. You have to create new volumes for which you can
specify exactly which servers they should use a replicas and copy the
data from the old volumes.

The root volume is sort of tricky, because pointing the servers at a new
root volume (by changing /vice/db/ROOTVOLUME) isn't sufficient, I
believe it requires at least a restart of all clients.

In theory it is possible to manually create new volume replicas, and add
them to existing volumes. However there is currently no way for the
server to inform clients that they need to invalidate any cached
information about the volume, so clients will not use the new replica
until they are at least restarted, but possibly not until they are
reinitialized. So creating new volumes, copying the data and changing
the root volume tends to be more reliable.

> I went through the vice-setup, everything went well (eventually !) it  
> created a new empty /vicepa and a new vice/db/vicetab with the unique  
> partition, but /vicepa only contains the FTREEDB file, which makes me 
> assume that nothing is getting replicated right now...

Right, by default it won't just start replicating, because if you add
more servers there is no way to know if you want the same data
replicated on those as well, or whether you rather gain more storage
space etc. Replication is treated as an administrative decision.

Jan
Received on 2008-07-21 12:01:09