Rework iroh-net address book

[rklaehn]

I think https://github.com/n0-computer/iroh/pull/2612 is just one out of many examples that you need some more information about nodes in the address book.

We should have a database containing for each node id some information about the provenance (where we learned about it), and maybe also simple statistics.

While doing this, we should also rework persistence. Not sure how to best do it, maybe a redb database that is in mem by default and can be pointed to a file optionally?

[divagant-martian]

we already have a provenance that we use for logging, but we don't store. So in a way we are halfway there

[flub]

@rklaehn what information do you need other than what will need to be added for #2612?

We decided a while ago to explicitly move persistence out of iroh-net.

iroh-net internally keeps a whole bunch of information in the NodeState, one major important component of that is in the PathState. This is essentially the information that's made public in some way via Endpoint::remote_info (was Endpoint::connection_info) which I assume is what you consider to be the "address book".

I think it will be natural that as our understanding, implementation and requirements evolve that this structs will evolve. I think that's fine. I don't see a particular reason why involving a database here will make life easier, on the contrary.

So I don't think I can see benefits here for a major rework along the lines you are suggesting.

---

For context what I'm currently do want to work on is cleaning up the NodeState and PathState because right now they contain a bunch of conflicting information, and claim to provide information which they can't. Some fields are updated and used in the wrong ways and other such minor disasters. That's a quite delicate cleanup because obviously things work good enough to mostly work right now, but also there are many bugs because of this.

The major change I would like this to lead to is that whenever the NodeState/PathState is updated the new send address is computed and stored. Currently this computation happens on the hot-path when sending a datagram. I believe getting things cleaned up enough to get to that stage will have significant benefits.

[rklaehn]

We don't have to make this overly complex provided that we have a way for an external entity to get every single thing that is related to discovery as an event.

Then for those cases that need something more complex than the logic we need for realtime, we can combine an event listener and a discovery mechanism to have the persistent storage external.

Note that this requires that the discovery trait allows a discovery subsystem to emit arbitrary provenances, which is currently the case. https://docs.rs/iroh-net/latest/iroh_net/discovery/struct.DiscoveryItem.html

[ramfox]

We have largely done what is mentioned in this issue:

1. we've decided the address book is ephemeral
2. users must subscribe to the discovery stream to get any passively discovered nodes
3. we now have the StaticDiscovery system to allow users to add node address information to the endpoint.