All Wikimedia production services will move to the Kubernetes pipeline

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view

All Wikimedia production services will move to the Kubernetes pipeline

James Forrester-4
*TL;DR*: As part of the annual plan, we are moving Wikimedia Services to be
run in an orchestrated platform driven by a new build pipeline (yes, that
means containers). All new services for Wikimedia production must now start
using this platform. By the end of June 2019, we will have started
converting most remaining services. If you're responsible for an existing
or new service, please talk to us.

Some Wikimedia functionality is provided by services which operate
alongside MediaWiki, complimenting or supporting important features for
users. They currently mostly run directly on the "bare metal" shared
service clusters. The Deployment Pipeline "TEC3" annual plan[0] involves
modernising production platforms, and we have a goal to move all of these
to the new platform we've been building.

There are many benefits of changing to use this platform:

* Adding/removing capacity is easy as easy as a deployment; consequently,
overall our clusters will be more scalable, and used more efficiently;
* Rolling deployments are the default, reducing disruption and increasing
dependability for readers, editors, and other end-users of our services;
* Deployments are automatically versioned, rather than relying on a manual
process from deployers;
* There is increased testing before the deployment, by means of a staging
cluster; and
* Developers no longer need to create and use deploy repos manually, the
pipeline does this.

If you are an owner, developer, or deployer of a current or forthcoming
Wikimedia production service, we need to hear from you.

Some services have already moved to this new platform, and others are
moving to it.[1] We are building and improving the features as we discover
needs, so while we are confident now that services can run on this platform
run reliably, we still lack comprehensive documentation for this platform.
We are already in the process of creating it but we would like your input
to make sure that we are supporting all our stakeholders.

Please peruse the page about the deployment pipeline on wikitech[2] and
poke around. What documentation do you actually want? How can we make your
work simpler and more straight-forward? Please file a task in
Phabricator[3] or comment on the talk page.[4]

[0] -
[1] -
[2] -
[3] -
[4] -

Deployment Pipeline team
*James D. Forrester* (he/him <> or they/themself
Wikimedia Foundation <>
Wikitech-l mailing list
[hidden email]