There's a LISA paper about this.. It may have been last year, it may be
this year. I have trouble keeping them straight sometimes. ;).
They built a provisioning system at a university where people could
choose one of several different OS's to be installed. We have a
consistent image with customizations for particular clusters. Things
like scheduling/queueing variations, amount of ram, amount of disk
space, whether system needs desktop services (X, vnc, etc.) or not.
Kickstart variations are pretty common and there are many to choose from.
We use DHCP/PXE. We provision the hosts directly, no intermediate
environment.
You could have multiple servers coexisting just fine as long as you:
1) don't use fully dynamic addressing
2) production vs test should have exclusive access to the MAC address.
Keep it in a database. Rocks is one system where it makes it easy to
manage this function. Xcat is another. It's not that hard to make a
lockout so that only one DHCP server will provide the address at a time
even though multiple listening. You just need to spend some effort on
the infrastructure side to make it work.
we're in the process of converting over to Xcat since:
*) it's OS agnostic (rocks is tied fairly tightly to CentOS, but can be
made to work in other, with varying degrees of difficulty)
*) it can manage hierarchies of install servers (federated)
*) you can also automatically provision the IPMI/LOM on the host, as
well as switch ports and descriptions.
*) you can do diskless or diskfull if you want.
*) other reasons that I've forgotten
On 10/26/2012 12:40 PM, Eystein Måløy Stenberg wrote:
Hi gents,
I would like to ask for your experiences when it comes to bringing a
server from bare metal to production-ready. From talking to people, it
seems like there are many ways this is done.
I assume you are using network-booting to kick things off?
Are you just provisioning one operating system? If not, how do you
make the selection between multiple OSes? Is this fully automated
(e.g. using mac-addresses)? Is this important to you, or do you
provision the same OS >90% of the time?
Do you have a separate environment where you do your provisioning? How
do you move between the provisioning environment and the production
environment?
One option I've seen is physically replugging the server to separate
production from PXE environment. Is there a more automated way to do
this? One problem I can see for automation is that the PXE booting
relies on DHCP, and you don't want multiple DHCP servers (production
vs. PXE environment). What about relying on static IPs for production
servers, and only using DHCP for new servers?
Finally, what are your biggest problems with your setup (if any)? How
could you save more time and make it easier? Do you have thoughts or
plans for the future?
I really appreciate any experiences you would like to share on this
topic.
Thank you.
_______________________________________________
Tech mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
http://lopsa.org/
