https://wikitech.wikimedia.org/wiki/Deployments/One_week
[tl;dr] My recommendation for us to switch to a one-week deploy is to
use Robla's suggestion (Option B on the wiki page[0]) starting on Monday
June 10th. This will coincide with the start of the MediaWiki 1.22wmf6
deployment on WMF servers.
Unless there are any major concerns with this proposal, I'll start
modifying the relevant parts of the wiki (mostly
https://wikitech.wikimedia.org/wiki/Deployments and
https://www.mediawiki.org/wiki/MediaWiki_1.22/Roadmap#Schedule_for_the_depl…
).
== Riding the Train ==
Now is the time to start getting more development teams to start riding
the MW Core deploy train. If a one-week cycle/14 day lifespan (with the
always possible important backports available) is something doable for
your team right now, please do let me know. If it isn't, please let me
know what would be "good enough". Would Option A have been?
== Reasoning for Option B ==
Although we strive to be a continuous integration and deployment house,
we aren't quite there yet. With that in mind, having the space between
deploying to
test/test2/mediawiki.org and the non-WP project sites is
beneficial.
Possible workarounds to this problem included having two deploy windows
on Monday (in the Option A version); one for test/test2/mediawiki in the
morning, with a few hours gap then one for all non-WP project sites. The
gap would be used for both automated and manual tests. While the
automated tests could be scheduled and completed in a reasonable time
frame, counting on the QA team to complete manual tests and report bugs
in a small window isn't a dependency we should have right now; manual
tests still catch a lot of issues that the automated ones do not.
Going with Option B does already reduce our wmfXX branch lifespan from
24 days to 14 days. "Lifespan" being the number of days that any branch
is live on any wiki (from the day it goes to test/test2/mediawiki to the
day its "slot" is replaced with a new wmfXX branch). Limiting our
testing time to get down to 10 days doesn't seem worth it yet.
== Future ==
To get us down to a quicker cycle (lifespan of 10 or fewer days) we'll
need to do a bit more work on the QA/testing side of things (which
involves everything from betalabs, to test/test2, to unit tests, to
selenium, et al). We're on the right path, thankfully.
Greg
--
| Greg Grossmeier GPG: B2FA 27B1 F7EB D327 6B8E |
| identi.ca: @greg A18D 1138 8E47 FAC8 1C7D |