[Labs-l] Wikimedia Labs Software Stack

Thu Oct 4 19:55:11 UTC 2012

On Thu, Oct 4, 2012 at 4:13 AM, Andreas Trawoeger
<atrawog at kartenwerkstatt.at> wrote:
> Hi everybody!
>

Hi!

> I'm currently running a project called opengeoserver.org that collects and
> redistributes freely available Open Geodata (at the moment I'm at 1.3 TB of
> data) . The project is currently receiving some funding from Wikimedia
> Germany and is a likely candidate to be migrated to the Wikimedia Labs
> sometime in the near future.
>

Do you mean that your test and development will move to Labs, or that
your production site will? The former is ok, the latter is not. Labs
is not meant for production services. If the production service is to
be moved, then we should be getting it going in a test and development
mode in Labs, then move it to production.

> But being a sysadmin that neither wants to replace his fine tuned XFS
> storage with Gluster nor his Fabric [0] deployment scripts with Puppet
> without a lot of testing. I would actually prefer to set up my own OpenStack
> instance on my existing systems, test different configurations options and
> then migrate the services to Wikimedia Labs.
>

That's a decent amount of data. We have about 40TB of gluster space
available in the pmtpa cluster and will have another 40+TB in eqiad.
It'll be slower than direct XFS access, but we don't really have a
means of providing that much space in a direct access mode.

We'll be using saltstack for deployment soon, which works in a very
fabric-like way. It's a little more flexible than fabric, in fact.
Puppet really shouldn't be used for deployment, just configuration and
dependency management.

> It's easy to figure out that Wikimedia Labs is using Ubuntu, OpenStack &
> Gluster, but I couldn't find any detailed information about what versions
> are currently used and how the current hardware & storage configuration
> actually looks like.
>

We have some slightly outdated docs here:

http://wikitech.wikimedia.org/view/OpenStack

- Ryan