Re: configuration management

[Tom Grace <lists-in@deathbycomputers.co.uk>]

On 07/05/11 00:56, shawn wilson wrote:
> i'd prefer the system have some sort of ajax, soap, xml, syslog, or
> snmp v3 output (in order of preference - can ya tell how i feel about
> snmp?). but, i suppose i can parse and spit out pretty much any output
> if need be.
What information are you looking to get from the output of a config run ?
IMHO, if you are using configuration management software the process 
should be kinda as follows:
Have a method of creating consistent base installs (for Debian, probably 
preseed).
Create clean machine based on that.
Run configuration management tool.
Profit.
> btw, if i could take lvm, zfs, and virtualbox snapshots with whatever
> software i use, that would be cool too :)
Again, my opinion, but there should be no need to take snapshots of the 
result of running any config management tool, as the canonical version 
of your boxes can be created using said tool

> comments? suggestions?
I'm mostly familiar with Puppet, and use it extensively. Something I've 
found is that your first couple of attempts will probably not be ideal. 
A large part of the config management puzzle is getting your 
abstractions in the right place.
For example, it is fairly easy to copy /etc into file resources in 
Puppet and push them to several machines. Ideally, there should be as 
little machine specific configuration as possible.

Kinda like this:

        httpd_config => {
            'MPM' => 'prefork',
            'StartServers' => '10',
            'MinSpareServers' => '15',
            'MaxSpareServers' => '25',
            'ServerLimit' => '256',
            'MaxClients' => '256',
            'MaxRequestsPerChild' => '4000',
            'Port' => '8000'
        }

and using variables to put those values into config files and overriding 
as appropriate rather than having multiple versions of files.


Hope that gives you a couple of pointers.