feature: Metrics for GossipSub

libp2p / go-libp2p-pubsub

The PubSub implementation for go-libp2p

https://github.com/libp2p/specs/tree/master/pubsub

Other

309 stars 178 forks source link

feature: Metrics for GossipSub #534

Open MarcoPolo opened 1 year ago

MarcoPolo commented 1 year ago

Metrics such as:

Peer prune rate per topic. How often a peer is pruning us per topic.
Peer counts per topic. How many peers do we have per topic.
Peer churn rate per topic. The sum of grafts+prunes per topic.

We should also include some grafana dashboards.

This should help node operators identify when their node is running into issues, or when GossipSub is limping along but still working.

vyzo commented 1 year ago

You can easily implement this with a tracer.

So an option to enable metrics, that hooks a tracer into you preferred metrics system is something the library could provide == see how it is handled in lotus.

Changing protocol code to directly hook metrics is not acceptable however.

nisdas commented 1 year ago

In Prysm, we have already implemented something pretty similar using a raw tracer: https://github.com/prysmaticlabs/prysm/blob/develop/beacon-chain/p2p/pubsub_tracer.go . It has been helpful in us identifying interesting spikes of certain control messages in the network for different topics. If a tailored metrics tracer for gossipsub lives inside go-libp2p-pubsub that would be a welcome addition on our end as we would no longer need to maintain the above.

MarcoPolo commented 1 year ago

If you’d consider upstreaming a patch, I’d be happy to review

nisdas commented 1 year ago

Thank you for the offer, I am bit occupied over the next few weeks, but once I get some bandwidth I can open up a PR for this to be upstreamed if no-one has done it yet.

vyzo commented 1 year ago

Please use a metrics subpackage in any pr for this, so that people who don't use the metrics don't incur any extra dependencies.

Just to be clear, I also think it would be a welcome addition to have in tree support for this.