Do humans look at the debug logs right now? I'm worried that this could make them so spammy as to be unreadable (if they aren't already). Is there some way to log this info to a new file that'll get bundled up into the test results?

Sure. I will log into a separate file into the results directory.

May want to checkout psutil module in python if you haven't already instead of making calls to pidstat.

In particular psutil.process_iter to loop across the pids.

Here is the initial CL for the same.
https://chromium-review.googlesource.com/#/c/444965/

Please let me know incase you have any suggestion.

This seems to add around 0.7% of load on the cpu(watched manually). One suggestion that todd mentioned was to add the page loaded  at that time. 
 
 

Discussed a bit w/ Ravi and was thinking initially we should put this feature behind an argument as to not impact existing control files and instead enable in a new control file.

Thoughts?

I'd think copying control.1hour -> control.1hour_pidstat and enabling there.

Would be interesting to validate tsproxy's impact ( crbug.com/653519 ) as one effort to weed out potential test related cpu load during PLT.

On that note, it would be useful to add a helper script or postprocessing method to aggregate the pidstat results into a hotlist ... especially for long running pids in a separate CL to that in #c4