bug: host-metrics reporting wrong system and process CPU utilization

[david-luna]

What version of OpenTelemetry are you using?

"dependencies": {
"@opentelemetry/api": "^1.4.1",
"@opentelemetry/auto-instrumentations-node": "^0.39.2",
"@opentelemetry/host-metrics": "^0.33.1",
"@opentelemetry/sdk-metrics": "^1.15.2",
"@opentelemetry/sdk-node": "^0.41.2"
}

What version of Node are you using?

v18.16.0

What did you do?

Instrument an application with console metrics exporter and watch the CPU utilisation values. There is a repo with the app

https://github.com/david-luna/otel-metrics-cpu-utilization

What did you expect to see?

According to semantic conventions I expect to see CPU utilization values between 0 and 1.

What did you see instead?

Values are way above 1 like the sample provided in the repo I get some metric readings like this one.

{
  descriptor: {
    name: 'system.cpu.utilization',
    type: 'OBSERVABLE_GAUGE',
    description: 'Cpu usage time 0-1',
    unit: '',
    valueType: 1
  },
  dataPointType: 2,
  dataPoints: [
    {
      attributes: [Object],
      startTime: [Array],
      endTime: [Array],
      value: 157.82086149302881
    },
    {
      attributes: [Object],
      startTime: [Array],
      endTime: [Array],
      value: 143.03498303049165
    },
    {
      attributes: [Object],
      startTime: [Array],
      endTime: [Array],
      value: 1083.5302883959548
    },
    {
      attributes: [Object],
      startTime: [Array],
      endTime: [Array],
      value: 1362.709911550165
    },
    {
      attributes: [Object],
      startTime: [Array],
      endTime: [Array],
      value: 0
    }
    // ...
  ]
}

Additional context

N/A

[david-luna]

@legendecas since you're the module owner could you please have a look? I'd be happy to help by creating a PR but 1st I want your ACK on this to discard a misunderstanding from my side

Thanks :)

[legendecas]

Right, and the semantic convention the package used is outdated as well. We should update the semantic convention all together.

[david-luna]

@legendecas thanks for the confirmation :)

I'm planning on crate a PR which:

• will refer to the updated semantic convention
• will change the calculation to make diffs from the latest measurement as semantic convention specifies

One question. Semantic conventions defines system.cpu.utilization as

Difference in system.cpu.time since the last measurement, divided by the elapsed time and number of logical CPUs

Since we are using the state attribute does it makes sense to divide by the number of CPUs?