1267717 - Figure out if we need 10s of idle on each talos run

Status: RESOLVED WONTFIX

(Testing :: Talos, defect)

Description

[Avi Halachmi (:avih)]

To speed up development cycles within talos, I was looking to reduce the initial idle time when running talos locally, which is currently 10 seconds.

I found it hardcoded here: http://hg.mozilla.org/mozilla-central/file/tip/testing/talos/talos/pageloader/chrome/memory.js#l40

Joel committed this whole file some years ago but he doesn't recall why it's 10 seconds, whether or not it's still required, etc.

For my local development purpose, changing it to 2000 was good enough and without noticeable changes at the reported test numbers, but I wasn't looking at RSS numbers.

Reducing this number could not only save valuable development time with talos, but also runtime at the infrastructure.

Comment 1

[Joel Maher ( :jmaher ) (UTC -8)]

I suspect there is a need to have a timer here- but 10s seems quite long.  My only concern with changing this is that on browser startup we do delayed initialization of many services.  Most of that probably occurs in that 10s, so by reducing the 10s we could be measuring the perf impact of those services.  Might not be an issue- but we would need to look at noise of tests before/after.

Comment 2

[Avi Halachmi (:avih)]

Note also that as far as I can tell, it's impossible to disable RSS collection and this delay.

If I run talos without --rss, it still goes through the memory collection code (and 10s delay), even if it's not set at test.py for the test I'm checking. Even more so, even if at this test I use rss = False, it still goes through the memory code and delay.

If there are tests where we don't care about memory data, then I think we still have a 10s delay from that.