We need to implement the Kademlia DHT(https://en.wikipedia.org/wiki/Kademlia) cluster to answer queries from BitTorrent clients. This cluster will hold the DHT records and expose a query API. Data will be saved on the metadata cluster, which would be composed of all data nodes or all nodes where we keep the metadata.

Specs

• http://bittorrent.org/beps/bep_0005.html
• https://github.com/libp2p/specs/tree/master/kad-dht
• https://docs.libp2p.io/concepts/discovery-routing/kaddht/
• https://github.com/libp2p/rust-libp2p/blob/0c34d9fdd45e81892824ba5faf3875ab35f14f1b/misc/server/README.md?plain=1#L5

Crates

• https://crates.io/crates/libp2p-kad (recommended)
  • https://github.com/libp2p/rust-libp2p/tree/master/protocols/kad
  • https://github.com/libp2p/rust-libp2p/tree/master/examples/ipfs-kad
• https://crates.io/crates/dht
• https://crates.io/crates/kademlia-dht

It would help to cache the results to Redis. But this would mean to fork the crate.

libp2p-kad doesn't fully respect the standard. Plan is to fork it and change it to standard based on a feature flag so we can PR back to them.
If that is too much work then look for another crate which implements the standard, if there are none, then create our own.

Then we need a custom impl to have auth and encryption. We could use Iroh or Quinn for that to take advantage of QUIC's build in TLS.
Auth could be added with Keycloak.