Michael Tiernan wrote:
>
> >Can I ask a question of the {puppet/chef} crowd?
> >
> >Looking for short answers. How do you handle the building of the system
> >from the bare-metal "power on" state until you have a self bootable
> >system built and ready for configuration specifics?
> >
> >I am, admittedly behind the curve, and am experimenting with some hand
> >tuned PXE boot to kickstart configs as well as tools like cobbler to
> >instantiate raw systems and trying to see it from a higher altitude
> >viewpoint.
Edmund White wrote:
> I bootstrap Puppet from the kickstart %post scripts. PXE or Netbook
> RHEL/CentOS installs.
>
> In terms of managing server roles/types, I define that in the Puppet
> Dashboard so subsequent runs allow the configuration to meet the system’s
> purpose.
>
My shop runs Cobbler/VMware/RHEL/Puppet/Puppet Dashboard/Nagios.
We manage more than 50 servers this way with several defined
roles (Oracle database, OFM appserver, Oracle Agent, ...). We
have defined about 20 Puppet classes.
Cobbler generates an Anaconda kickstart file from defined
distros, profiles, templates, snippets, and roles (management
classes).
We define a Cobbler system for each server, choose a profile,
input a few things like IP and hostname. We create a blank VM
with suitable quantities of disk, RAM, and CPU. We power on the
VM and it PXE boots. Cobbler serves the DHCP, the boot image, and
the kickstart file. Ten minutes later the base system reboots (RHEL
"first boot") and Puppet begins installing apps and configuring
itself.
We prefer to run Puppet only on demand from a client, but this
is not typical. We maintain a lot of install and configuration
logic outboard to Puppet in plain ol' bash scripts (Puppet
installs and runs the scripts). The advantage is it enables
people with no Puppet skills to develop and maintain parts of
the system. I believe this strategy was a big win for us. An
example is our Oracle OFM (appserver) install.
Puppet is far from perfect. We had a year to choose a CM system
and narrowed the choices to Cfengine, Chef, and Puppet.
Cfengine seemed more comprehensive than we needed, although it
probably has the best approach. Chef had a much smaller
community using it than Puppet, so Puppet was what we went with.
(Ansible and Salt were not on our radar at the time.)
While choosing Puppet was not a killer, I wouldn't choose it
again without first trying other solutions. I hope never to
build more than 4 servers in a system without using an automated
deployment/configuration management system. Our breakeven for
manual vs automated provisioning calculated out to 5 servers.
--
Charles Polisher
_______________________________________________
Tech mailing list
Tech@lists.lopsa.org
https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
http://lopsa.org/
