Persist peers for bootstrapping

[Winterhuman]

Description

Mirroring libp2p/go-libp2p-kad-dht#254, it would be great if long-lasting peers were persisted and used for bootstrapping later.

Motivation

Using bootstrap nodes outside of the default ones improves network resiliency, and potentially improves initial connections times when the new bootstrap nodes are closer than they otherwise would be.

Requirements

1. Save the longest lasting peer(s) from each k-bucket where it can be re-used for bootstrapping.

Open questions

• Which k-buckets are considered, and how many peers are picked from each bucket?
• Should latency be factored into picking future bootstrap nodes? If so, how?
• Other open questions mentioned at Saving previously seen nodes for later bootstrapping ipfs/kubo#3926 and Persist routing table go-libp2p-kad-dht#254.

Are you planning to do it yourself in a pull request?

No

[dariusc93]

If im understanding the issue correctly, one could implement their own store for RecordStore. Something like sled or rocksdb could be a good place to start for persistence and move up from there for different backend stores, though something to be included for rust-libp2p is hard to say. Mostly feel that it should be an outside implementation.

[Winterhuman]

Perhaps you're right, I mainly opened this to mirror libp2p/go-libp2p-kad-dht#254 which targets go-libp2p as opposed to Kubo, although ipfs/kubo#8856 being open at the same time confuses things as to where this sort of code should belong

[thomaseizinger]

In rust-libp2p, PeerIds are resolved to addresses by querying all configured behaviours via NetworkBehaviour::addresses_of_peer. It should be relatively straight-forward to implement your own NetworkBehaviour that extracts addresses from the NetworkBehaviour::inject_connection_established callback, stores them in some persistent way and queries that storage whenever addresses_of_peer is called.

To avoid latency issues, I'd be best to not do any IO in the NetworkBehaviour::poll method as that is run as part of the main event loop of a Swarm.

[Winterhuman]

Closing since this can be implemented independently