1293787 - Windows debug builds as tier 1

Status: RESOLVED DUPLICATE of bug 1370297

(Firefox Build System :: Task Configuration, task)

Comment 1

[Greg Arndt [:garndt]]

So it looks like the one bug that this depended on is resolved.  Should we come up with a plan for moving to tier 1 status for windows?

Comment 2

[Pete Moore [:pmoore][:pete]]

(In reply to Greg Arndt [:garndt] from comment #1)
> So it looks like the one bug that this depended on is resolved.  Should we
> come up with a plan for moving to tier 1 status for windows?

Hey Chris, Sheriffs,

Do you know what other requirements we might need to meet before we can promote windows taskcluster firefox desktop builds from tier 2 to tier 1?

Pete

Comment 3

[Ryan VanderMeulen [:RyanVM]]

https://wiki.mozilla.org/Sheriffing/Job_Visibility_Policy

Comment 4

[Pete Moore [:pmoore][:pete]]

I think we meet all these guidelines. Rob, what do you think?

Comment 5

[Rob Thijssen [:grenade (EET/UTC+0300)]]

I've read the guidelines now. I believe the tc windows builds comply with them.

My only hesitation is that we do not yet run the full test suite against the build artefacts (we run all tests that are part of the build configuration, but test instances do not then collect the build artefacts and run the Win xp, 7, 8, 10 tests). I couldn't see any evidence of this being a requirement for Tier 1 though, so I'd say, there isn't a compelling reason not to promote unless anyone wants to discuss the test stuff further.

Comment 6

[Greg Arndt [:garndt]]

So I think the goal when going to tier 1 is that buildbot is shut off soon thereafter.  If tests in taskcluster are not ready for tier 1, then we need to maintain buildbot for awhile longer for it and thus, need to keep taskcluster windows stuff as tier > 1.

Comment 7

[Phil Ringnalda (:philor)]

Not exactly: there's nothing that prevents us from having two tier-1 builds on the same platform (how could there be and why would there be? these rules weren't carved on stone tablets, they're just people writing down what seemed like reasonable things so that they wouldn't have to keep answering the same questions multiple times).

There are only two differences between tier-2 and tier-1 that matter to you: by saying "I want this to be tier-1" you are committing to 24/7 support (to the same extent you committed to 24/7 support for your other tier-1 builds) for infra bustage, and in return if someone breaks your build but not the buildbot build with a push, then we will back them out immediately rather than just filing a bug and letting you negotiate backout-or-fix with them.

Comment 8

[Greg Arndt [:garndt]]

Good points, I somehow overlooked that. If everyone feels that they meet the visibility requirements and agrees that they should be tier 1, so be it.

Comment 9

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

The risk of two tier-1 builds is that they compete for space in the index.  This might be a problem for artifact builds, for example.  Given continued active attention to the platform as you're working on tests, I think keeping tier 2 for now might be the best plan -- it gives you time to respond, get a feel for infra problems, and so on without closing trees.  When the tests are close to done and you're very confident in the build quality [*] you can overlap for a few days (confidence meaning nobody should care which job they find in the index) and then demote buildbot to tier 2 for a while.

[*] glandium found some very difficult-to-spot issues with the linux build by doing an object-file-level comparison.  I don't know if such a thing is practical for windows builds?

Comment 10

[Chris Cooper [:coop] (he/him)]

(In reply to Phil Ringnalda (:philor) from comment #7)
> There are only two differences between tier-2 and tier-1 that matter to you:
> by saying "I want this to be tier-1" you are committing to 24/7 support (to
> the same extent you committed to 24/7 support for your other tier-1 builds)
> for infra bustage, and in return if someone breaks your build but not the
> buildbot build with a push, then we will back them out immediately rather
> than just filing a bug and letting you negotiate backout-or-fix with them.

This is the key point for me: if we promote these Windows debug builds to tier 1, the burden of platform support will shift from releng to TC. If we have faith in the new platform and the TC team is fine with this, let's do it.

Comment 11

[Phil Ringnalda (:philor)]

Should probably do a little more provisioning first, though: I've noticed that as often as not the first set sits through a claim-expired retry.

Comment 12

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

There's still quite a bit of work to do here - it's not a matter of just switching things on.

  http://migration.taskcluster.net/#/details/windows-debug-tier1?_k=t9si1w

Comment 13

[Chris AtLee [:catlee]]

This was done as part of bug 1370297