[supervisor] track cpu/memory usage analytics

[akosyakov]

Description

This PR adds analytics to track cpu and memory consumption by individual workspace instances. It adds 2 new events gitpod_cpu_changed and gitpod_memory_changed correspondingly. Each event has buckets property and reported whenever a workspace instance hit higher bucket.

• CPU buckets: 1, 2, 3, 4, 5, 6 CPUs
• Memory buckets: 1, 2, 4, 8, 16 Gb

It will allow us to segment users by perf requirements and also correlate with other event like active language to see requirements from different languages. Mixpanel is capable to breakdown arrays to individual properties. I've verified it with IDE lab project.

Related Issue(s)

Fixes #

How to test

• Start a workspace in prev env: https://ak-perf-anayltics.preview.gitpod-dev.com/#https://github.com/akosyakov/parcel-demo
• Inspect gitpod_cpu_changed and gitpod_memory_changed events in segment untrusted. I would recommend to filter by instance id. Whenever perf metrics hit a higher bucket you should see a change. If it goes down though we don't report changes.

Release Notes

NONE

Documentation

Werft options:

• /werft with-preview
• /werft analytics=segment|TEZnsG4QbLSxLfHfNieLYGF4cDwyFWoe

[akosyakov]

/werft run

👍 started the job as gitpod-build-ak-perf-anayltics.2
(with .werft/ from main)

[akosyakov]

/werft run

👍 started the job as gitpod-build-ak-perf-anayltics.5
(with .werft/ from main)

[akosyakov]

/werft run with-clean-slate-deployment=true

👍 started the job as gitpod-build-ak-perf-anayltics.6
(with .werft/ from main)

[akosyakov]

/werft run with-clean-slate-deployment=true

👍 started the job as gitpod-build-ak-perf-anayltics.7
(with .werft/ from main)

[andreafalzetti]

I took down the prev env (too much memory usage testing) 🤦‍♂️

/werft run

👍 started the job as gitpod-build-ak-perf-anayltics.8
(with .werft/ from main)

[andreafalzetti]

the prev env is back now, but werft failed so I am trying with:

/werft run with-clean-slate-deployment=true

👍 started the job as gitpod-build-ak-perf-anayltics.9
(with .werft/ from main)

[andreafalzetti]

Code changes LGTM 🍪

I've tested it and it works as advertised. I was able to get the CPU to hit the biggest bucket and the reporting worked quite well.

For the memory I also tried to stress it to see reports and I saw up to 4GB, and saw no more reporting but I think it's due to my approach for stressing the memory which is too agressive, I think: cat /dev/zero | head -c 5G | tail, and it basically very quickly kills the workspace.

If you have another way to consume more memory, I am interested to know.

/hold

in case someone wants to review/test further

[mustard-mh]

If you have another way to consume more memory, I am interested to know.

Files in /tmp folder will calc as workspace's MEM, you can generate big files here. @andreafalzetti

[akosyakov]

/unhold