607531 - Test buildbot-0.8.2 on staging before deploying

Status: RESOLVED FIXED

(Release Engineering :: General, defect)

Description

[Dustin J. Mitchell [:dustin] (he/him)]

Buildbot-0.8.2rc1 is out, and should be tested.

So far, I've set this up on sm02, and some slaves connected, and, well, it's doing stuff.

I found and fixed one bug (the rebuild failed when rebuilding a job originally run with 0.8.1).

I need advice on other things to try.

I'm also told that there are buildbot patches in place for 0.8.1. How should I try to apply that to 0.8.2?

Comment 1

[[e-mail] (:bhearsum)]

Your best bet to see differences is diffing the tip of http://hg.mozilla.org/build/buildbot against the BUILDBOT_0_8_1 tag and/or looking at the log. There should be very few things that aren't just imported upstream changesets, though.

Comment 2

[Dustin J. Mitchell [:dustin] (he/him)]

catlee suggests testing try builds, too.

Comment 3

[Dustin J. Mitchell [:dustin] (he/him)]

14:11 < catlee> dustin: oh, wrt to local patches to buildbot, I think the only thing we have a is a custom slave-side command
14:11 < catlee> I think everything else has been consolodated with upstream in one way or another
14:11 < catlee> http://hg.mozilla.org/build/buildbot/file/58225ba72678/slave/buildslave/commands/mozilla.py
14:12 < catlee> and http://hg.mozilla.org/build/buildbot/file/58225ba72678/slave/buildslave/bot.py#l19

Comment 4

[Dustin J. Mitchell [:dustin] (he/him)]

In a local repo, I have the results of

  hg co upstream
  (untar 0.8.2rc2)
  hg addremove
  hg commit
  hg co default
  hg merge upstream
  (lots of patches resolved) <-- the interesting part to review
  hg commit

The resulting diff is too big and impenetrable to upload here, so it sounds like it will, more or less, get stamped.

My plan is to get 0.8.2 set up on a staging and test box, and let the rest of releng go to town on it, trying to break it.  If nothing breaks, I will land these changesets and we can start rolling the new version out in downtimes.

As a note to self:

 * buildbot-configs' mozilla/master_common.py needs to have c['changeHorizon'] set to None, lest changes get purged too soon and mess up the 'set_who' step.

The only other bug I ran into (a traceback involving properties.runtime when rebuilding) is solved in 0.8.2rc2, so nothing more needs to be done.

Comment 5

[Dustin J. Mitchell [:dustin] (he/him)]

Attachment: 607531-buildbot-configs.diff

buildbot-0.8.2 is up and running on sm03.  I've reset the database since I let the changeHorizon pruner run and thereby invalidated the existing db.  But it did successfully upgrade the db and continue running builds.

buildbot changes are here
  https://bitbucket.org/djmitche/moz-buildbot/changesets
and the buildbot-config changes are attached and also here
  https://bitbucket.org/djmitche/moz-buildbot-configs/changesets

(I'm using bitbucket until I learn enough 'hg' to not hurt anyone)

At this point, I need folks more familiar with what a working buildmaster should look like to give this one a hard time and see what breaks.

Comment 6

[Dustin J. Mitchell [:dustin] (he/him)]

This bug is arguably done - it's tested.

Bug 610025 deals with getting the buildbot-configs patch landed early.  610045 is the final deployment.