Wikimedia Platform Engineering/MediaWiki Core Team/Quarterly review, March 2013/Summary
What's on the TODO list for the next 3 months, what's "As time allows" and what's on the backburner (see also: agenda)
Priority projects up to June
We presented three projects as priorities leading up to June:
- Volunteer contributed deployments
This is potentially a large project, but we would like to keep the scope small. We went in with the initial goal of asking Mobile to be the primary/litmus customer for this work, but it became clear that Labs needs are the better litmus.
We plan to start this work in earnest after the dust settles on the Scribunto deployment.
Volunteer contributed deployments
A few we plan to get to:
- Category collation improvements - code review, scripts
- VipsScaler (just a few days of work)
- Score (nearly all done)
No issues there. We discussed Blame Maps as well, and determined that we likely won't be ready for deployment, but will have done some more code review work on this. We need to scope out hardware requirements before going out further with this.
In post meeting wrapup conversations, we determined that the review queue process needs revisiting.
Greg will figure out logistics on this, working to make sure we cover both architectural readiness (e.g. Tim's review) and product scope (e.g. Howie's review).
Ram and Chad are working on this, with Tim's guidance. The team will investigate deploying Solr more widely relatively soon. As a prerequisite, we'd like to hire our Search Engineer before going down this road, but we may have enough pieces in place that we should just move forward. We already use it in prod for GeoData & translation memory. Wikidata engineers have done development work toward deployment for their work. We'll coordinate with Ops (particularly Asher & Peter) on this.
Other projects and priorities
Central code repo
The idea of a central wiki for code, such as Lua and Gadgets code, is very appealing. However, we decided not to do this work prior to deploying Lua, but plan to revisit this idea next quarter.
Lua will remain our primary performance deliverable for the 2012-13 fiscal year. We discussed HipHop readiness, and determined that the other thing we might try to do (assuming Ops and Platform have available time) is an upgrade to PHP 5.4. If this is still outstanding in a quartery, we'll revisit this in the next review.
We didn't cover this in detail in the meeting, but some post-review discussions led to the following:
- Tim will be working with one or more engineers in the Mobile group on Performance
- A couple team members expressed the desire to keep plugging away at code review rather than explicit LevelUp activities.
Questions we discussed
Should we aspire to a faster cycle than once every 2 weeks?
- Yes. Greg is leading this conversation.
How active of a role should we play in support of non-Wikimedia uses of MediaWiki?
- We care about the third-party community of enterprise users, etc., but we aim to keep the dollar investment bounded.
- Greg & Sumana to move forward on this to formalize the maintainership of the tarball & related activities.
Test infrastructure: What role should beta play in the deployment cycle? Do we make people deploy to beta before production, period?
- Sort of. Greg is conversing with lead devs on different teams re: how they deploy code today. We should also evaluate alternate possibilities.