Blink Infra Owners is a group of people that work to improve Blink-related infrastructure; some of us work on Blink Infra as our primary project; others work on a particular aspect of Blink Infra part-time.
These notes are mailed monthly to blink-infra, bcc blink-dev and infra-dev.
Attendees: atotic, jeffcap, jparent, jsbell, kbr, qyearsley, skobes
Status updates and discussion:
jeffcarp:
There are some possible follow-ups, including potentially removing
kbr: Swarming tests on mac:
Try bots and continuous bots are not the same right now.
The question: which type of machine do we need to focus on: VMs vs Mac Minis?
Also, theoretically if there is a bug that can be fixed for one of these platforms so that they behave consistently then we wouldn't need to choose one.
What gpu-related flags are passed to content_shell when running layout tests? Ideally we want to run layout tests with GPU enabled but for most tests it shouldn't matter.
The bug for this is crbug.com/736177.
atotic:
Update from last time: Aleks fixed the issue he discovered with pipes being slow when running layout tests locally (crbug.com/725229), tests now run much faster locally.
Aleks's new results page is now ready to be made the default. qyearsley plans to send out a CL for this this week (related thread).
skobes:
Working on a tool for gardening flag specific expectations :-)
swarming:
Because we're currently part-way switched over to swarming, flakiness dashboard results are split (crbug.com/748195)
One possible solution might be to make both of the step names the same
Swarming layout tests on Windows is blocked by the issue with content_shell not exiting (crbug.com/717348)
Maybe on Windows the driver processes are ending, but not content_shell, and a more cooperative way to end content_shell is required.
Maybe this isn't just happening swarming, but swarming is a stricter environment; maybe we could reproduce locally, which would help with solving the problem.