feat!: return metrics objects from register instead of updating with an options object

[achingbrain]

Removes the component update methods in favour of registering metrics once with methods that return objects which can wrap metric gathering implementations.

This lets us (for example) return a wrapped prometheus gauge and avoid concatenating strings, etc when extracting metrics.

Register methods for metrics (prometheus gauges) and counters are present, extra registerHistogram etc methods can be added at a later date when required.

Single metrics

Before:

metrics.updateComponentMetric({
  system: 'libp2p',
  component: 'connection-manager',
  metric: 'incoming-connections',
  value: 5
})

After:

const metric = metrics.registerMetric('libp2p_connection_manager_incoming_connections')

// call repeatedly
metric.update(5)

// or
metric.increment()

// or
metric.increment(5)

// or
const stopTimer = metric.timer()
// later
stopTimer()

// or
metric.reset()

// etc

Metric groups

Before:

metrics.updateComponentMetric({
  system: 'libp2p',
  component: 'connection-manager',
  metric: 'connections',
  value: {
    incoming: 5,
    outgoing: 10
  }
})

After:

const metric = metrics.registerMetricGroup('libp2p_connection_manager_connections')

// call repeatedly
metric.update({
  incoming: 5,
  outgoing: 10
})

Extracting metrics

This has got much simpler, this example is from an upcoming PR to js-ipfs:

import client from 'prom-client'

// Endpoints for handling debug metrics
export default [{
  method: 'GET',
  path: '/debug/metrics/prometheus',
  /**
   * @param {import('../../types').Request} request
   * @param {import('@hapi/hapi').ResponseToolkit} h
   */
  async handler (request, h) {
    if (!process.env.IPFS_MONITORING) {
      throw Boom.notImplemented('Monitoring is disabled. Enable it by setting environment variable IPFS_MONITORING')
    }

    return h.response(await client.register.metrics())
      .type(client.register.contentType)
  }
}]

BREAKING CHANGE: the global/per-peer moving average tracking has been removed from the interface as it's expensive and requires lots of timers - this functionality can be replicated by implementations if it's desirable. It's better to have simple counters instead and let an external system like Prometheus or Graphana calculate the values over time

[github-actions]

🎉 This PR is included in version @libp2p/interface-metrics-v4.0.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀

[github-actions]

🎉 This PR is included in version @libp2p/interface-mocks-v8.0.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀