Per-component metrics

[achingbrain]

I'd like to get some insight into what's going on inside a libp2p node. At the moment the metrics are a bit limited, only focussing on bandwidth stats for peers and protocols and few other bits, from what I can see.

It would be really useful to be able to get component-specific metrics to really see what's going on.

How about something like:

libp2p.metrics.updateComponentMetric(component: string, metric: string, value: number)

This would set per-component metrics, eg:

libp2p.metrics.updateComponentMetric('kad-dht', 'active-queries', 1)
libp2p.metrics.updateComponentMetric('dialler', 'pending-dials', 10)
// etc

Then getting metrics out:

libp2p.metrics.getComponentMetrics() => Map<string, Map<string, number>

We'd then report these metrics in the /debug/metrics/prometheus endpoint in ipfs-http-api-server which would allow graphing these stats over time when IPFS_MONITORING is set in the environment similar to how we do with the number of connected peers.

I don't think we need moving averages, or anything fancy like that - the tool we use to examine the stats can figure all of that out.

[vasco-santos]

I think this is super valuable! For further iteration, we should look into getting this closer to what we want to have with https://github.com/libp2p/observer-toolkit/tree/master/packages/proto

On previous hack week, I did a small POC in the daemon libp2p/js-libp2p-daemon#45 and I think this will likely be something we want to use in the daemon. But, there are useful things that we should extract on the libp2p repo that would make easier to consume things in IPFS and the observer toolkit