Wikimedia Release Engineering Team/Goals/201516Q2/checkin-10-27


30 (ish) day checkin on Q2 goals[edit]


  • do mw core gate-processing time for this quarter, reassess for next quarter if needed
  • warning: carbon aggregate metrics per average after roughly a month. So old data most probably do not represent what you want.

Reduce number of deploy tools from 3 to 2[edit]

Migrate all Service team owned services and MW deploys to scap3[edit]
technically mediawiki deploys already use scap3 but none of the new features ;)
I really want to use it for Phabricator deployments but hit a few blockers last time I tried.
I (Tyler) have concerns about this goal
* We've done 1 deploy to beta cluster
** service team as a whole, unaware
* One deploy wasn't flawless
** Ugly Puppet wrangling
** I (Tyler) think it went well, but I knew where the sticking points would be
** Configuration concerns were raised (resulting doc sprint this week)
* "How far from AQS deployments are we?"

How to stay on track[edit]

Open question.

Initial thoughts:

  • Focus on Analytics Query Service (based on RESTBase) —definitely first real-world use-case
    • Limited permissions for deployers
    • Seemingly means they can't use ansible (likely)
    • TODO Puppet wrangling required
    • TODO Who else to talk to about it?
    • TODO How can we test it?

Retire Gerrit and Gitblit in favor of Phabricator[edit]

Decommission Gitblit[edit]
upstream is adding an "optional, unique repository name" which should solve our issues with callsigns
Phabricator hosted repositories are now available over https and ssh.
  • url redirect (gerrit->phab repo name) stuffness
  • stupid naming

Code review RFC: creation, publication, discussion, feedback etc[edit]
Phabricator can now merge differential patches with dry dock.
  • talk more in the weekly

Weekly checkin[edit]

Weekly checkin created on thursday. Pending Antoine familly schedule ... Will follow up on internal mailling list.

Reduce CI wait time[edit]

CI checkin minutes

All: Please read and comment on

Need a better time for US folks.

CI cluster responds to spike in queued builds by starting and registering additional jenkins slaves[edit]
  • 10 standing instances, will automatically spike to 20 as needed
  • nodepool doesn't send metrics to statsd :( :( :(

Migrate majority of CI jobs to Nodepool[edit]
  • < 10% right now
  • we need the caching implemented before we can do much more otherwise we risk being blocked/rate-limited by upstreams (github etc)
    • everyone, please review the caching proposal! :)

Release MediaWiki 1.26[edit]

A quality MW 1.26 successfully released[edit]
Board pruned of extraneous material, needs more triage.
Chad will send email to wikitech-l asking for help with the blockers today

Be firm in what we are going to accept in 1.26.