Add help copy explaining instance metrics

[charliepark]

We added several new metrics, including CPU utilization charts that incorporate different states — running, emulating, waiting, idling. We should add some help copy to explain them, and link to relevant documentation for more.

[david-crespo]

I asked in Q&A for more info and got some.

My question

we are putting OxQL-powered CPU stats in the console and I'm thinking about how to briefly describe run, idle, waiting, and emulation for the user. we could leave idle out if it's generally 100 minus the other three, but even so that leaves the other three to describe

• definitions in propolis
• some definitions in illumos: vmm_sol_dev.c, vmm_kernel.h

run seems to correspond in the happy case to what you'd see in htop inside the guest. waiting is about contention and hopefully is near zero most of the time? I'm least clear on emulation

we will probably end up with two levels of description: short ones for right under the chart titles and longer ones for popovers or the docs site

@gjcolombo

I think that's basically right:

• "run" means "this vCPU is running guest instructions"
• "idle" means "this vCPU has halted"
• "waiting" means "the host has a thread that wants to run this vCPU, but the system isn't scheduling that thread right now," probably due to contention of some kind
• "emulation" means "this vCPU did something that forced it to exit the guest, and now the host is taking care of whatever that was"--think "did some I/O," "executed an instruction like CPUID that the host wants to intercept and handle," and so forth

Even more discussion of the emulation state: oxidecomputer/propolis#855 (comment)

Waiting is never going to be zero

I thought VMs were in some sense locked to CPUs. I asked whether that makes waiting less likely.

rm: They aren't really tied right now. Even though we want to do better allocation and core aware scheduling. Ultimately we bounce you around a lot and even if it was, you can still end up in waiting for other reasons.