On 28/10/09 11:51, Kaloyan Kovachev wrote:
On Tue, 27 Oct 2009 14:32:39 -0500, David Teigland wrote
On Tue, Oct 27, 2009 at 08:21:50PM +0100, Fabio M. Di Nitto wrote:
2) if we use httpd only to distribute cluster.conf, then I?d like to see
"httploader" (or please find a better a name) being a wrapper for wget
rather than a brand new piece of code. It will allow us to automatically
gain access to ftp, https, and it has all the bits required for
username/password handling (in case somebody wants to secure httpd
properly).
Yes, I like wget.
I think it is very easy for an admin to include "wget<conf_location> -o
/etc/cluster/cluster.conf" inside cman start or even add a hash check of the
downloaded (in /tmp) file before moving it to /etc/cluster so there is no need
for this to be a module for the initial boot. This will help in the case when
a node was offline.
For the update on a running node, multicast from/to the cluster members is
much better i think.
Is it possible on startup to read from the local file only the cluster name,
version, multicast address and the key file, then to 'request' the current
version number and new config if newer from the running nodes?
Having a different protocol for initial join and updates is a
non-starter, it's far too complicated and inconsistent.
And if sysadmin adds the wget to the startup sequence then that only
fixes things until the configuration needs updating ... then we're back
to manually distributing the file, so it really doesn't solve anything.
Multicasting a whole config file is also fraught with difficulties -
some of which were exemplified by the now-abandoned CCSD.
Chrissie
--
Linux-cluster mailing list
Linux-cluster@redhat.com
https://www.redhat.com/mailman/listinfo/linux-cluster
