Coda File System

Re: Coda in a small web server cluster

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Tue, 3 Oct 2000 11:02:15 -0400
On Mon, Oct 02, 2000 at 05:34:42PM -0700, Adam Hunt wrote:
> I am setting up a small high availibility web server cluster (2-3 servers
> initilally).  I would like to use Coda but I am confused on a few points. 
> Because I only plan on having two to three web servers to start it is hard to
> justify two more servers to act as dedicated Coda servers.  So I would like to
> make each of the three web servers Coda servers also.  I am a little lost when

Well, each server will already have to run a Coda client and since Coda
clients can run disconnected you could also consider having only a
single Coda server. With single `replication' there is no chance of
getting server-server conflicts which affect all clients, so only
reintegration conflicts after the disconnection, which affect only a
single client, could cause a problem.

And they should even be minimal when there is little to no write
traffic. Our own webserver ofcourse serves it's files out of Coda, and
the only writes are to update the mailinglist archives and search
indexes. We did have reintegration problems once in a while, so I've
actually been considering moving the update traffic to another machine
completely to avoid losing `availibility' on the webserver.

> it comes to the relationship between the web services (Apache, php, etc.) and
> the Coda cell.  Are the web servers going to be required to obtain security
> tokens to access the Coda volumes?

Not for read access when the System:Anyuser ACL allows read access.
However, all write access requires a token because a client will not
reintegrate after a (temporary) write-disconnection without one.

My scripts which are run by the cron daemon look somewhat like this:

    #!/bin/sh
    set -e
    echo "supercalifrageous" | clog -pipe webserver
    update-mail-archives
    generate-search-indexes

Jan
Received on 2000-10-03 11:05:31