Closed Bug 927937 Opened 11 years ago Closed 9 years ago

Send separate sendchanges per target test platform

Categories

(Release Engineering :: General, defect)

defect
Not set
normal

Tracking

(Not tracked)

RESOLVED WONTFIX

People

(Reporter: coop, Unassigned)

Details

(Whiteboard: [buildfaster][scheduling])

We'd like to get to a model where we don't run tests on all variants of an OS for every changeset. 

There is some sheriff tooling that needs to happen before that can become a reality, but one way to facilitate this would be to stop triggering multiple test platforms off of a single builder sendchange, and instead trigger each test platform individually with it's own sendchange.  This immediately gives us finer-grained control over re-triggering.

Once we get to a model where we're not running tests on every OS per check-in (say, only triggering testing on all OSes for PGO builds, as suggested in bug 855282), we could continue to generate the sendchanges for the OSes we skip. I imagine it wouldn't take too much extra tooling to get TBPL to display the skipped sendchanges so sheriffs could retrigger them as needed.
Something to note is that this model can only be as granular as:
* all opt tests
* all debug tests
* all talos tests
I think the right place to fix this is in the scheduling rather than the sendchanges.
(In reply to Chris AtLee [:catlee] from comment #2)
> I think the right place to fix this is in the scheduling rather than the
> sendchanges.

+1, in TC!
Status: NEW → RESOLVED
Closed: 9 years ago
Resolution: --- → WONTFIX
Component: General Automation → General
You need to log in before you can comment on or make changes to this bug.