Open Stack

On 05/15/2013 06:54 PM, Ryan Lane wrote:
> I've just upgraded my production nova/glance/keystone infrastructure
> from essex to folsom and I'd like to provide some feedback. I know I'm a
> full release behind, but I don't feel like I'm out of the ordinary for
> this as a production user.
>
> First off, this upgrade was the easiest so far; great job! I only have a
> couple complaints to the community:
>
> 1. The documentation for upgrading was basically non-existent. There was
> a small amount of information in the release notes regarding upgrading,
> and they were helpful, but upgrade documentation they are not. It looks
> like this is still the case with the grizzly release.
>
> There's no documentation on which configuration options were added (and
> are required) or removed. There's none about new configuration files
> that are used. There's none about the changes to the policy.conf files
> (which is really evil, because that has security concerns).
>
> I'm fairly familiar with the code base and very familiar with the
> services, so I was able to create upgraded configuration files somewhat
> easily, but a less experienced user would likely have had a much harder
> time with this.
>
> 2. Having configuration in the paste files *really* sucks. The vast
> majority of that simply isn't configuration. Since those files are in
> /etc, packages will treat them as configuration files, and the default
> for packages is to not replace configuration files (especially when
> using puppet/salt/etc). That means the paste files will be completely
> wrong after an upgrade. Can the configuration be split away from the
> paste files?

+1  That was actually the major tripping point I found in doing grenade 
testing from folsom to grizzly, there are still a number of services, 
like keystone, that mix config with paste.

We're getting better, but for us to be really deployer friendly we need 
to really separate the two for all services.

	-Sean

-- 
Sean Dague
http://dague.net