This is also *not* Mihai.

All of Mozilla gathered in Orlando, Florida last week for one of our twice-yearly all-hands meetings. Affectionately called “Mozlando”, it was a chance for Mozilla contributors (paid and not) to come together to celebrate successes and plan for the future…in between riding roller coasters and drinking beer.

Even though I’ve been involved in the day-to-day process, have manage a bunch of people working on the relevant projects, and indeed have been writing these quasi-weekly updates, it was until Chris AtLee put together a slide deck retrospective of what we had accomplished in releng over the last 6 months that it really sunk in (hint: it’s a lot):

Slide deck from @chrisatlee’s #releng 6-month retrospective yesterday: https://t.co/ebVpC9PRqm #mozlando

— Chris Cooper (@ccooper) December 9, 2015

But enough about the “ancient” past, here’s what has been happening since Mozlando:

Modernize infrastructure: There was a succession of meetings at Mozlando to help bootstrap people on using TaskCluster (TC). These were well-attended, at least by people in my org, many of whom have a vested interest in using TC in the ongoing release promotion work.

Speaking of release promotion, the involved parties met in Orlando to map out the remaining work that stands between us and releasing a promoted CI as a beta, even if just in parallel to an actual release. We hope to have all the build artifacts generated via release promotion by the end of 2016 — l10n repacks are the long pole here — with the remaining accessory service tasks like signing and updates coming online early in 2016.

Improve CI pipeline: Mozilla announced a change of strategy in Orlando with regards to FirefoxOS.

FirefoxOS is alive and strong, but the push through carriers is over. We pivot to IoT and user experience. #mozlando

— Ari Jaaksi (@jaaksi) December 9, 2015

In theory, the switch from phones to connected devices should improve our CI throughput in the near-term, provided we are actually able to turn off any of the existing b2g build variants or related testing. This will depend on commitments we’ve already made to carriers, and what baseline b2g coverage Mozilla deems important.

During Mozlando, releng held a sprint to remove the jacuzzi code from our various repos. Jacuzzis were once an important way to prevent “bursty,” prolific jobs (namely l10n) from claiming all available capacity in our machine pools. With the recent move to AWS for Windows builds, this is really only an issue for our Mac build platform now, and even that *should* be fixed sometime soon if we’re able to repack Mac builds on Linux. In the interim, the added complexity of the jacuzzi code wasn’t deemed worth the extra maintenance hassle, so we ripped it out. You served your purpose, but good riddance.

Release: Sadly, we are never quite insulated from the ongoing needs of the release process during these all-hands events. Mozlando was no different. In fact, it’s become such a recurrent issue for release engineering, release management, and release QA that we’ve started discussing ways to be able to timeshift the release schedule either forward or backward in time. This would also help us deal with holidays like Thanksgiving and Christmas when many key players in the release process (and devs too) might normally be on vacation. No changes to announce yet, but stay tuned.

With the upcoming deprecation of SHA-1 support by Microsoft in 2016, we’ve been scrambling to make sure we have a support plan for Firefox users on older versions on Windows operating systems. We determined that we would need to offer multiple dot releases to our users: a first one to update the updater itself and the related maintenance service to recognize SHA-2, and then a second update where we begin signing Firefox itself with SHA-2. (https://bugzil.la/1079858)

Jordan was on the hook for the Firefox 43.0 final release that went out the door on Tuesday, December 15.

As with any final release, there is a related uplift cycle. These uplift cycles are also problematic, especially between the aurora and beta branches where there continues to be discrepancies between the nightly- and release-model build processes. The initial beta build (b1) for Firefox 44 was delayed for several days while we resolved a suite of issues around GTK3, general crashes, and FHR submissions on mobile. Much of this work also happened at Mozlando.

Operational: We continue the dance of disabling older r5 Mac minis running 10.10.2 to replace them with new, shiny r7 Mac minis running 10.10.5. As our r7 mini capacity increases, we also able/required to retire some of the *really* old r4 Mac minis running OS X 10.6, mostly because we need the room in the datacenter. The gating factor here has been making sure that tests works still work on the various release branches on the new r7 minis. Joel has been tackling this work, and this week was able to verify the tests on the mozilla-release branch. Only the esr38 branch is still running on the r5 minis. Big thanks to Kim and our stalwart buildduty contractors, Alin and Vlad, for slogging through the buildbot-configs with patches for this.

Speaking of our buildduty contractors, Alin and Vlad both received commit level 2 access to the Mozilla repos in Mozlando. This makes them much more autonomous, and is a result of many months of steady effort with patches and submissions. Good work, guys!

The Mozilla VR Team may soon want a Gecko branch for generating Windows builds with a dedicated update channel. The VR space at Mozilla is getting very exciting!

I can’t promise much content during the Christmas lull, but look for more releng updates in the new year.

http://coopcoopbware.tumblr.com/post/135471138340