On 20/08/12 10:00, Steve Traylen wrote:
> On Aug 14, 2012, at 2:23 PM, Gareth Roy wrote:
>
>> Hi,
>>
>> I've been tasked with building a test cluster here at Glasgow from some old equipment that we have floating around. The idea is to put together a system so we can test out software updates etc before rolling them into the production environment (such as the EMI2 WN discussed today in Ops) and also as a new grid admin it gives me more familiarity with the systems and middleware we have in place. I'm taking the opportunity to try and start from scratch so we can experiment with stuff that might make the production cluster easier to manage.
>>
>> To begin with I'm auditing methods of automated machine provisioning, in my past life looking after HPC systems I've used commercial tools such as Bright Cluster Manager as well as tools like Cobbler etc for getting from bare metal to a live linux machine. Presently we use a custom set of tools that build our nodes via kickstart but I wanted to ask what methods you guys have used and had success with.
>>
>> The follow-up question is then what you have had success in using for configuration management. In following conversation in Ops I know there is a strong leaning towards Puppet as well as sites that use cfengine.
>>
>> Any comments, advice, experience you have would be greatly appreciated.
QMUL uses a kickstart templated with mason, together with a customised
pxe boot that
>>
>
> We seem to be slipping further and further down the http://theforeman.org slide at the moment and I'm trying to switch off various CERN written services with it.
> It certainly covers the job that cobbler will have done for you in the past but also a puppet ENC in integrated.
> This has mostly been to provision the metal so far but it seems the integration of it over openstack is also working (well almost) .
>
I do get the perception that many people (including CERN) are moving to
Puppet.
There were three talks at hepsysman this spring
https://indico.cern.ch/conferenceDisplay.py?confId=183204
And Tim Bell's CHEP talk
http://indico.cern.ch/contributionDisplay.py?contribId=408&confId=149557
And a poster by Doug Benjamin
http://indico.cern.ch/contributionDisplay.py?contribId=16&sessionId=8&confId=149557
There's clearly some scope for sharing puppet configs and both Doug and
Tim seemed happy in principle, but I don't think either had the effort
to do it on their own.
Chris
|