Release-Mgmt: My First Beta from the ‘other’ side

Hello and welcome to my continued documentation off my learning curve in Release Management, something I’ve now been working at for 6 weeks.  Last time I was reeling from the new-to-me meeting/email/bugmail firehose.  Now I’ve got that more under control, having created many more filters and folders in Thunderbird as well as having a chance to do more work on the automated tracking emails script.

To continue to spread out the knowledge and tasks for release managers across more than just one Alex, last week I ran my first beta (Firefox 12 beta 4) and now I’m going to tell you what that involved because the bugzilla API is down.

Monday

Monday is about getting the queries down as much as possible.  This means making sure anything with Tracking?, approval-mozilla{beta,aurora,esr10}? is triaged and nudged further towards its final destination on the trains. We’re also needing to watch out at this point for riskier fixes, things that really need some bake time with actual users need to be landed for a beta4 since beta5 is more for low-risk regression back-outs and security fixes that need to land just in time for beta 6 (what is usually re-built as the appropriately branded final release).  At this point in the week there might be about 80-90 bugs we need to get sorted and at the end of the day only about 6 bugs were in the ‘really want this in beta 4’ list.

Tuesday

For this particular beta, we had been asked if it would be possible to give the ‘go’ to build earlier than usual because there was a fairly popular holiday weekend coming up (Easter) and our QA lead for this release was in Canada where Good Friday is a statutory holiday. The QA lead asked if we could go to build earlier so that a Thursday release of the beta would be possible. We confirmed with RelEng that this was do-able and agreed to do our best to get the ‘go’ out at an earlier time.  The plan was for everything to be landed by 2:30pm (Pacific) in order to have a changeset ready to fire off to Release Engineering by 5pm.

Going through the ‘burn list’ from Monday (6 bugs) mostly entailed tracking down people to land patches. There has to be a cutoff time for landings since it takes about 4 hours to get all the builds and tests for a push to the hg repo to report back completely. Note that improvements to build times are being worked on, case in point: faster Mac builds (newer hardware and using OS X 10.7)  take ~2.5hours off the normal Mac build times.

For one of the bugs we weren’t able to reach the dev and a volunteer committer found that the patch didn’t apply cleanly so that bug had to miss the train.  The others got in and I sent the ‘go’ to build Firefox 12 beta 4 at approximately 5:45pm (45 minutes later than desired).  All the results weren’t in yet for this changeset but I wanted to get a shot at making the request for earlier builds to QA so I took the risk that the builds/tests would be OK and we’d be already building when we confirmed that.  Had the builds/tests *not* turned out we’d have to scrap the chance at moving up the release window and one of our QA leads would have worked on a holiday to meet our ‘normal’ Friday release window.  So I took the leap (and this ended up being fine, though I wouldn’t do this again without good reason as it was a stressful call to make and it’s not a good practice to get into).

Right after the ‘go’ email was sent hg.mozilla.org went down and we lost 3 hours of build time.  This is not a normal result of giving the ‘go’ to build.  It was probably just because Hal was new to doing releases and I was new to running a beta so at least one thing had to come along and shake our confidence.

Wednesday

There’s not much (beta-running) to do on Wednesday except wait for Desktop & Mobile QA to do their thing.

Thursday

Mobile & Desktop QA send their results out – either signing off on the builds/updates or calling out issues. At this point QA signed off so at that point I could request the Release Engineer to push the updates to the beta channel (and upload a new beta apk to the Google Play Store). An hour or so later both QA leads signed off on the updates for the release that’s now live to our beta users.  After that, there’s just some product details to change for our websites to include beta 4.  We don’t do release notes per-beta which is good to know for when I run a beta 1.

Friday

Normally we’d be doing the push to beta channel on a Friday, so what would have been different was:

• getting everything landed to mozilla-beta could have gone until later in the day
• the ‘go’ to build email wouldn’t have resulted in immediate builds, they could start on Wednesday at the beginning of the Release Engineer’s day
• QA needs two full nights of testing, so we’d get sign offs from QA on Friday morning instead – hence the push to beta channel (and store) on Friday afternoon

Nothing too crazy for my first beta.  I think beta 4 is a good one to start on – it’s not the “OMG last call!!1!” beta before a release and it’s not beta 1 where any fallout from our merge of mozilla-aurora -> mozilla-beta shake loose. Having done a beta release now,  I have a much more complete mental map of how the 6 week release cycle plays out for Release Management:

Multiply by the above by 6, sprinkle extra bug & meeting communication cycles to weeks 1 and 6, throw in twelve channel meetings, approx 30 more iterations of various queries triage to keep our tracking lists up to date and to know what’s really needing attention vs. what’s taking care of itself.

A ton of email/irc/automated notifications all with the goal of keeping tracked bugs moving forward and you’ve got your 6 week result: A fabulous new Firefox release.

Thanks for reading, more about automatic emails & wiki updates soon.