New issue
Advanced search Search tips

Issue 850252 link

Starred by 1 user

Issue metadata

Status: Available
Owner: ----
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug


Sign in to add a comment

Clean up GPU test suites in src/testing/buildbot/*.pyl files

Project Member Reported by kbr@chromium.org, Jun 6 2018

Issue description

After the conversion of the GPU waterfalls to use the src/testing/buildbot version of generate_buildbot_json.py in  Issue 844815 , we should take another look at how the various GPU test suites are specified.

There's a lot of complexity in the organization of the chromium.gpu.fyi waterfall right now. The bots run top-of-tree ANGLE and an extended set of tests.

A couple of things we should consider:

1) Splitting up some bots into "extra" and "ANGLE" bots. The "extra" bots would run only extra Chromium tests, like the WebGL 2.0 conformance suite. The "ANGLE" bots might run only ANGLE-specific tests like angle_end2end_tests or angle_white_box_tests.

2) Having the optional Chromium GPU trybots mirror those "extra" waterfall bots once they're specified, rather than having a separate specification.

More ideas, and patches, welcome.

 

Comment 1 by kbr@chromium.org, Jun 8 2018

Blockedon: 844582

Comment 2 by kbr@chromium.org, Jun 22 2018

Blockedon: 854698

Comment 3 by kbr@chromium.org, Jun 26 2018

Blocking: angleproject:2689
angleproject:2689 is a situation where testing top-of-tree ANGLE during ANGLE rolls is problematic. angle_end2end_tests was failing briefly with the DEPS version of ANGLE but win_optional_gpu_tests passed in this ANGLE roll:

https://chromium-review.googlesource.com/c/chromium/src/+/1115739

because it ran with ToT ANGLE, which included this fix:

https://chromium-review.googlesource.com/c/angle/angle/+/1115770

I'm not sure what the best way to address this is. The GPU FYI waterfall has historically run ToT ANGLE everywhere, and so do the optional GPU trybots. We could consider having separate ANGLE bots which only run things like angle_end2end_tests and the dEQP tests on top of ANGLE, and which would also run ToT ANGLE, but this has significant disadvantages; it would double the load for the WebGL 2.0 conformance suite, for example, which would need to run with both DEPS ANGLE and ToT ANGLE.

Any help anyone can offer in helping redesign which bots should run what tests is greatly appreciated.

Blockedon: 822310

Sign in to add a comment