cpu: calculate current CPU usage from delta instead of avg since boot

[NotAShelf]

Fixes #51, at least partially. Did not yet implement time-delta measurements yet, but it is not exactly a part of the issue I was trying to fix. Will be implementing at a later date alongside a --dry-run flag.

This PR only introduces a new expression (cpu-usage-since) for evaluating CPU usage over a specified time interval and refactors CPU usage calculation to be based on recent usage deltas rather than historical averages. The time expression is backed by the humantime crate, which allows for intuitive time formats. Also adds support for logging CPU usage history.

Signed-off-by: NotAShelf raf@notashelf.dev
Change-Id: I2088217f2c2924fa9200f43033bb92426a6a6964

[RGBCube]

A few Q's:

• is %cpu-usage still the "from first boot" one? We should remove that because people will shoot themselves in the foot again.
• We need to append to CpuLog a lot more if the config has anything that is cpu-usage-since = "1sec" or similar. Otherwise the small calcs will just be 0 (not even sure it'll be zero, it should be the None expression)
• In general we need to remove all "from boot" parts.

[NotAShelf]

is %cpu-usage still the "from first boot" one? We should remove that because people will shoot themselves in the foot again.

No, it's no longer from boot because we're changing it; cpu_log.usage now uses cpu.current_usage(), which is delta-based, and state.cpu_usage comes from cpu_log.usage so %cpu-usage returns delta-based usage

We need to append to CpuLog a lot more if the config has anything that is cpu-usage-since = "1sec" or similar. Otherwise the small calcs will just be 0 (not even sure it'll be zero, it should be the None expression)

I think the solution here is to sample at a fixed high frequency (like 100ms or 500ms) regardless of poll_interval, and then cpu_log.iter() can filter by the requested duration window.

In general we need to remove all "from boot" parts.

You're right, I'll remove those. Addressed the rest of your comments.

[RGBCube]

No, it's no longer from boot because we're changing it; cpu_log.usage now uses cpu.current_usage(), which is delta-based, and state.cpu_usage comes from cpu_log.usage so %cpu-usage returns delta-based usage

How much of a delta? I don't think we should have that, everyone should set the delta themselves.

I think the solution here is to sample at a fixed high frequency (like 100ms or 500ms) regardless of poll_interval, and then cpu_log.iter() can filter by the requested duration window.

Won't that be too intensive? Also sounds like a waste

[NotAShelf]

Do you have a better idea?

[RGBCube]

The idea I had was traverse the Expression and collect all of the cpu-usage-since = "..." durations and sample with some fraction of that delta. We can use a derive Visitor crate to make this really easy and get the min usage-since Duration

[NotAShelf]

It does not seem like I'll be receiving any more meaningful comments, so I'm merging. The issue is fixed, nitpicks can wait.

[RGBCube]

bro casually ignored the %cpu-usage variable being completely broken and useless ☠️