Coda File System

Problems mounting volumes

From: John Anderson <johnha_at_ccbill.com>
Date: Fri, 15 Apr 2005 09:10:19 -0700
Hello all,

Just some quick questions preceeded by some background information.

Overview:
    I'm working on a little proof of concept project that has to do with 
load balanced webservers.  Basically, we have several identical 
webservers in clusters, or server farms, or whatever you wish to call 
them.  This requires we round robin through these web servers every time 
we need to update the content, cgi's, CPAN modules, etc.  It also means 
that we have to have completely seperate systems for each of these 
boxes.   I decided to try something different and go diskless with the 
web servers.  Each server would PXE boot, load a stripped out linux root 
filesystem from an initrd into ram, and mount coda volumes containing 
groups of content, CPAN modules, code, etc.  I decided to use coda 
because of the obvious performance and reliability benifits of local 
caching and disconnected operation.  None of the web servers will ever 
need to write back to the coda server. 

Specifics:

Server & Client -  Commodity Hardware, Fedora Core 2, Kernel 2.6.11.5 
(GRSEC Patched, but GRSEC disabled in config.....leaves me something to 
break later :-) ), lwp 1.12, rvm 1.10 rpc2 1.26 coda 6.0.8.

My problems so Far:
I have followed the directions in the Administration Manual 
(http://www.coda.cs.cmu.edu/doc/html/manual/book1.html).  I've built the 
required libraries and binaries and installed them and I ran the 
vice-setup script choosing mostly default values.  The / volume that was 
created during this process is named after the hostname of the machine I 
was testing on (kickstart.hq.local).  I PXE booted a client machine, and 
from the shell ran venus-setup and venus-init.   I then cd'ed into /coda 
and saw another directory kickstart.hq.local, so I cd'ed into that one 
and created some files which showed up on other PXE clients.  I guessed 
it was working.  However I was under the impression that the / 
filesystem volume would mount on /coda/ in the client's filesystems 
instead of /coda/servername as it seems to work.  Is this correct usage 
or did I screw something up?

My next question has more to do with best practices.  I'll be mounting 
coda volumes that would usually be located under /usr/lib/per5, 
/var/www/, /etc/httpd, etc.  Can Coda mount volumes using these types of 
directories as mountpoints, or is it a better idea (maybe the only 
idea?) to mount the necessary volumes under /coda/servername, and 
symlink them out the appropriate areas?

Thanks.

-- 
- John A.
Systems Administrator
CCBill, LLC.
Received on 2005-04-15 12:10:41