I noticed that when running Fenix performance tests (at least on try), a full PGO build of Geckoview is made.

e.g.
https://treeherder.mozilla.org/#/jobs?repo=try&tier=1%2C2%2C3&revision=13afa153c0e7c815de1b7b2f71145fb103c62812

However this build is not needed since the Fenix performance tests use a baked (nightly) build of Fenix.

To save cloud costs, I believe we can remove the build of Geckoview from Fenix jobs.

These were the tests triggered:

"test-android-hw-g5-7-0-arm7-api-16-shippable/opt-browsertime-tp6m-fenix-cold-bbc-e10s",
"test-android-hw-p2-8-0-arm7-api-16-shippable/opt-browsertime-tp6m-fenix-cold-bbc-e10s"

These Fenix tests also triggered a Geckoview PGO build:

"test-android-hw-g5-7-0-arm7-api-16-shippable/opt-raptor-speedometer-fenix-e10s",
"test-android-hw-p2-8-0-arm7-api-16-shippable/opt-raptor-speedometer-fenix-e10s"

:bhearsum, is this something you would be able to help us with? I'm not entirely sure what is causing these to get scheduled.

(In reply to Greg Mierzwinski [:sparky] from comment #1)

:bhearsum, is this something you would be able to help us with? I'm not entirely sure what is causing these to get scheduled.

I'm happy to help! To be clear: it sounds like you're saying that the Fenix performance tests do not need to depend at all on any Android builds? Eg: All of the "Bpgo" jobs in the try push linked don't need to run?

(In reply to bhearsum@mozilla.com (:bhearsum) from comment #2)

(In reply to Greg Mierzwinski [:sparky] from comment #1)

:bhearsum, is this something you would be able to help us with? I'm not entirely sure what is causing these to get scheduled.

I'm happy to help! To be clear: it sounds like you're saying that the Fenix performance tests do not need to depend at all on any Android builds? Eg: All of the "Bpgo" jobs in the try push linked don't need to run?

Thanks :bhearsum! Yes exactly, we download the APK file from a taskcluster index route instead of building them on the commit:

1. https://searchfox.org/mozilla-central/source/taskcluster/ci/test/raptor-fenix.yml#11-18
2. https://searchfox.org/mozilla-central/source/taskcluster/ci/test/browsertime-mobile.yml#22-45

So this is ending up trickier than it looks. Although we don't use the build, we do require the mozharness package from the build to run tests (see https://phabricator.services.mozilla.com/D83369#2564853).

I don't think there's any easy way of moving forward with this at the moment...

possibly we can be creative and use mozharness.zip from the m-c build we get the test packages from, not sure if that is possible or makes sense.

Oh that's unfortunate.

Would it be possible to force artifact builds when we have the target field set to a url? (I'm also wondering about what you think about jmaher's idea).

(In reply to Greg Mierzwinski [:sparky] from comment #7)

Oh that's unfortunate.

Would it be possible to force artifact builds when we have the target field set to a url? (I'm also wondering about what you think about jmaher's idea).

I'll try to dig into these ideas this week or next. Off the top of my head, I'm not sure the viability or difficulty of them.

Looks like bug 1628981 is going to make this easier.

:bhearsum, are you still working on this issue?

(In reply to Greg Mierzwinski [:sparky] from comment #10)

:bhearsum, are you still working on this issue?

Sorry, no. I don't have time to dig into this in the forseeable future.

Ok thanks, no problem!