On 06/07/2015 16:14, J. Ryan Stinnett wrote: > It seems like wakeups are something we should try to track over time, > so we find issues like this more quickly. > > Can we use some performance system like Talos / Raptor / etc. to track > these over time with each build?
I'd say mostly yes depending on the scenario. In this case (battery life
at idle) the test I did is reproducible since kernel tracing is
available on all our development & dogfooding devices and it can be
augmented by use of the battery harness. For my test I just flashed a
device clean (i.e. default profile, no apps configured), waited a while
for it to settle down then used a script to enable tracing, grabbing the
trace and analyzing it. This is definitely doable in automation and I
think :jhylands (which I'm CC'ing) is working on something similar using
the Flame battery harness which is obviously an even better approach
than looking at traces.
We had other scenarios however that are not so easy to catch with
automated tests. I remember an old issue (bug 864516 [1]) where we
leaked a timer every time we turned the screen on and off. Obviously
this was invisible with a freshly flashed phone but would become
terribly obvious to any dogfooder after a while.
Gabriele
[1] CPU load spike at each minute mark of system time
https://bugzilla.mozilla.org/show_bug.cgi?id=864516
signature.asc
Description: OpenPGP digital signature
_______________________________________________ dev-b2g mailing list [email protected] https://lists.mozilla.org/listinfo/dev-b2g
