GitHub Actions: Integrate telemetry action

[uncleDecart]

This action will gather statistics around eden test execution in GitHub Runners, helping us identify bottlenecks

CC: @milan-zededa

Let's see what we have in Eden repo

[milan-zededa]

If this works well we could keep this beyond the current issue troubleshooting.

[uncleDecart]

If this works well we could keep this beyond the current issue troubleshooting.

Yep, I hope it works during failure as well, they should give some outputs :D

[milan-zededa]

I see error Error: [Workflow Telemetry] Resource not accessible by integration, although diagrams are provided...

[uncleDecart]

Huh, CPU load goes in the beginning high, we are not using much resources

[uncleDecart]

Seems to be working, adding sys calls information to get better visibility.

[giggsoff]

Seems to be working, adding sys calls information to get better visibility.

Can you please check Step Trace section? Looks like it contains only one (or two) steps. Are you sure that other charts contains the whole workflow?

[uncleDecart]

you're absolutely right @giggsoff, let's try this one :D

[uncleDecart]

Seems like it works time wise, still this steps are weirdly seen, could be because of reusable actions?

[giggsoff]

Seems like it works time wise, still this steps are weirdly seen, could be because of reusable actions?

I can see related issue here.

[uncleDecart]

So am I being punished for over-engineering separating things? Oh man. Well, we can create an action without this hierarchy, but still biggest question is how can we see nested virtualisation impact? It should be something like count of context change, but for that we technically need to get to host, from guest machine maybe sleep time could help?

[milan-zededa]

So am I being punished for over-engineering separating things? Oh man. Well, we can create an action without this hierarchy, but still biggest question is how can we see nested virtualisation impact? It should be something like count of context change, but for that we technically need to get to host, from guest machine maybe sleep time could help?

Maybe this could be relevant for us: https://scoutapm.com/blog/understanding-cpu-steal-time-when-should-you-be-worried

[uncleDecart]

@milan-zededa I think we should merge this and then in my separate fork I'll run this on self-hosted runner so that we can compare numbers and talk to buildjet about it