Change CPU panels' units from percent to sishort (20% -> 200m)

[consideRatio]

Before	After

The groups panel "CPU Usage" isn't screenshotted, because I haven't got it to work yet due to #165.

Unchanged panels with CPU related data include:

[consideRatio]

Ooops the description of this PR is for #164, will fix it from a computer when given the chance (EDIT: DONE!)

[jnywong]

Can you explain from your perspective why changing the units from % to (nano, micro, milli)-cores would be more user-friendly?

I think it's more technically correct to express the units as (nano, micro, milli)-cores, but less intuitive for someone looking at these dashboards to get a vague sense of % compute used out of the available requested.

I'm inclined to keep this as % unless you can convince me otherwise! I think at the very least we should add a tooltip description that unit of CPU compute here is broadly the number of cores used in a given time period.

[consideRatio]

Thank you @jnywong for taking the time!!

someone looking at these dashboards to get a vague sense of % compute used out of the available requested.

There are some panels related to CPU that I didnt change in this PR that measured the fraction of CPU used up on a node - a part-divided-by-whole situation - there i agree fully on using percent!

However, for the panels updated, the unit is in actual CPU cores in absolute terms, so if we use percent it means 50% represents 0.5 CPU cores. It seems confusing to me that the percent relates to 1 CPU core specifically - unrelated to the pod containers requested CPU cores, or the node's available CPU cores for pods' containers, or the nodes total CPU capacity etc.

Is 50% and 250% to represent 0.5 CPU and 2.5 CPU is better here than 500m and 2.5? From maintaining k8s container requests and limit, i appreciate sticking with using the latter, and I figure people reading these panels may want to do just that as well. Separately, I also think there is merit to reserving the use of percent for part-divided-by-whole situations where it is very clear what the whole is from the context.

[jnywong]

Thanks for the thoughtful response!

Yes from the k8s requests/limits point of view this makes sense. There are definitely times where a user is using a Dask Gateway cluster and you get values like 1500% which look funny!

For a hub admin (who does not operate k8s) looking at the User/Group diagnostics and interested in how efficiently users are requesting CPU, maybe they can benefit from a unitless/% based panels with a time series of memory and CPU usage/requests (in a different PR :D)?

Then I would be happy to merge this PR with CPU cores, as long as the tooltip description explains that the SI units in the Y axis ($\mu$-, m-, etc.) refer to CPU cores. Does that sound good?

[consideRatio]

Sounds excellent to me @jnywong!

I added panel descriptions that appear as tooltips when you hover over the info icon next to the panel titles!

Thank you for taking the time to help out reviewing @jnywong - it makes a big difference to me!

[jnywong]

Fantastic! Thanks for the small changes, they make a big difference 🚀

[consideRatio]

Thank you @jnywong!!! ❤️ 🎉 🚀