772959 - [tracking bug] foopy-less setup for pandas

Status: RESOLVED WONTFIX

(Release Engineering :: General, defect, P3)

Description

[Armen [:armenzg]]

Because tegras and pandas should be at the top of the food chain and not at the bottom.

The first two bugs is to rip out of buildbot the execution of the jobs.
The second one would move us away from clientproxy/foopy model.

bug 650890 - mozharness for remote talos
bug 650881 - mozharness for remote unit tests
bug 738824 - web app to help check out available boards

Comment 1

[Aki Sasaki (not active)]

The above will allow us to cut the hard link between the build slave and the device.
However, it won't automatically allow us to ditch the foopies; if we run a 1-1 device-to-slave model, we'll overrun the test pool.

Adding a soft blocker of bug 706887 - allow for mozharness script parallelization.
Essentially, we would ask for, say, 10 devices and run 10 tests side by side on the same slave.  This would allow us to run on the test pool without overwhelming the test pool with device tests.

Alternately, we could spin up more controllers in AWS or the like, at which point a 1-1 device-to-slave model would be more scalable.

Comment 2

[Aki Sasaki (not active)]

(In reply to Aki Sasaki [:aki] from comment #1)
> The above will allow us to cut the hard link between the build slave and the
> device.
> However, it won't automatically allow us to ditch the foopies

Merely cutting this hard link would probably green up the remote tests significantly, even if we don't get rid of the foopies.
Still, not needing a class of hardware that needs to be maintained separately would probably be a win anyway.

> Alternately, we could spin up more controllers in AWS or the like, at which
> point a 1-1 device-to-slave model would be more scalable.

We would have to look hard at whether network latency/hiccups would affect the testing negatively here.

Comment 3

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

So would the buildbot slave run directly on the panda?

Comment 4

[Aki Sasaki (not active)]

(In reply to Pete Moore [:pete][:pmoore] from comment #3)
> So would the buildbot slave run directly on the panda?

No, this was to track grabbing pandas/tegras from a service like mozpool on a test slave, and run some number of them in parallel via a single script.  Not sure if this should be WONTFIX or if it's still valuable.

Comment 5

[[e-mail] (:bhearsum)]

Do we still care about this?

Comment 6

[Justin Wood (:Callek)]

foopy-less setup is now just emulators, we're planning to stop using pandas, and at this point don't have any plans to connect buildbot directly on the pandas which is what this bug was about.