Lighthouse: Enable sampling profiler for better `bootup-time` attribution

[patrickhulce]

Feature request summary
Enable the disabled-by-default-v8.cpu_profiler trace category.

What is the motivation or use case for changing this?
In https://github.com/GoogleChrome/lighthouse/pull/7059#issuecomment-461924036 we discussed enabling the sampling profiler to improve the attribution of scripts in bootup-time audit. Additionally, one of our smoketests that tests attribution was thought to only require async stacks, but turns out also needs sampling profiler now due to other restructuring of tasks.

https://github.com/GoogleChrome/lighthouse/blob/b4d9107263db043e6f776892571e4b9575c78a42/lighthouse-cli/test/smokehouse/tricky-metrics/expectations.js#L63-L71

How is this beneficial to Lighthouse?
Improved attribution of scripts and the elimination of "Other" as a large category in bootup-time results.

[patrickhulce]

A few other comments in #9200 and #9250 bring up the issue where scripts have parse time but 0ms of execution, which is almost definitely not the case :)

[radical-edo]

I've found a way to get those numbers. I've been loading the script with type="module". Once I've removed it the lighthouse was able to give me the scripting value for the bootup times. Maybe that's something that will help you to track down the issues with this scripting.

[patrickhulce]

Thanks @radical-edo! That's expected at the moment, #9200 outlined that specific shortcoming and is one of the cases that would be solved by the sampling profiler. It's mainly modules and specific types of async callback work.

[OZZlE]

This really needs to be prioritised I think! This "Other" is the biggest thing on a website I am looking at and we have no clue where it comes from.. Also I couldn't find this flag you mention?

Perhaps it's not a regular flag? Can I enable it somehow on my Chrome anyway?

[OZZlE]

URL | Total CPU Time | Script Evaluation | Script Parse
-- | -- | -- | --
Other | 9,145 ms | 1,593 ms | 2 ms
-- | -- | -- | --

😞

[patrickhulce]

@OZZlE if by "flag" you're referring to disabled-by-default-v8.cpu_profiler it's a trace category requested over the DevTools protocol, not a Chrome flag. And unfortunately it requires logic to then consume its output, so simply adding the flag wouldn't be enough to workaround your issue.

The lack of movement here is not because we don't want to do it or it's hard to do, it's around the current belief that this trace category would drastically impact the other performance results. Until we can show convincingly otherwise, it will not be implemented.

[connorjclark]

Until we can show convincingly otherwise, it will not be implemented.

How can we verify the impact of this? DZL??

[patrickhulce]

How can we verify the impact of this? DZL??

Yep, the last time I tried this out in DZL there was no observable impact, but that didn't seem to convince anyone at the time.

[ErisDS]

I'm also having a problem with a site that reports most of the time spent on "Other". Could someone who understands lighthouse a bit explain what known things can end up in this category?

[OZZlE]

Could it be implemented as a Flag? Perhaps enabling it requires a restart of Chrome but I would be fine with it. Even if it affects the over-all scoring of a website, you could at least see where this Other time is coming from; fix that issue, disable the flag and re-run the Audit in Chrome without the flag enabled..

As it is now you could even think that this is just some made-up thing "Other", how do we as web developer know if it's a real thing even right now? When we have no way of seeing what this actually is..

[patrickhulce]

Adding as an option is a neat idea @OZZlE I like it!

[themreza]

What's the status on this? Has it been released?

[TeoTN]

I'd love to see this ready- my page is running smoothly but it gets poor Performance score with 15seconds in "Others" section, I'm really curious what is hiding there

[patrickhulce]

We decided to proceed here and add this data to bootup time in an additive way.

[OZZlE]

@patrickhulce what does that mean? 😕 In PageSpeed I no longer see any "Other" but in Chrome I still see it.. Did you implement any flag we can enable to see what it is?

[patrickhulce]

Did you implement any flag we can enable to see what it is?

That's exactly what this issue is about :)