Double connection established, but observed by listener only (ghost connection?)

[b-zee]

I am doing an experiment with a Kademlia network of 200 nodes, locally on my PC. I am using identify to add new adresses to the DHT. And I am using QUIC as transport layer.

I observe that between some peers, a connection gets established twice. But, the dialer does not show this.

Listener:

[2023-03-30T07:34:01.249Z DEBUG] ConnectionEstablished { peer_id: PeerId("12D3KooWNiHroU8x2zecGQXRFU3p7mZkujYxHJ5vFBWVxhyxeGXQ"), endpoint: Listener { local_addr: "/ip4/127.0.0.1/udp/12001/quic-v1", send_back_addr: "/ip4/127.0.0.1/udp/12013/quic-v1" }, num_established: 1, concurrent_dial_errors: None, established_in: 5.515427ms }
[2023-03-30T07:34:01.251Z DEBUG] ConnectionEstablished { peer_id: PeerId("12D3KooWNiHroU8x2zecGQXRFU3p7mZkujYxHJ5vFBWVxhyxeGXQ"), endpoint: Listener { local_addr: "/ip4/127.0.0.1/udp/12001/quic-v1", send_back_addr: "/ip4/127.0.0.1/udp/12013/quic-v1" }, num_established: 2, concurrent_dial_errors: None, established_in: 6.191895ms }

Dialer:

[2023-03-30T07:34:01.249Z DEBUG] ConnectionEstablished { peer_id: PeerId("12D3KooWFFwDodRUMKymsQSvr7y6vaDDEReryyynvdUw191Hf8Qs"), endpoint: Dialer { address: "/ip4/127.0.0.1/udp/12001/quic-v1/p2p/12D3KooWFFwDodRUMKymsQSvr7y6vaDDEReryyynvdUw191Hf8Qs", role_override: Dialer }, num_established: 1, concurrent_dial_errors: Some([]), established_in: 9.826861ms }

After a while the QUIC connection will timeout for one of the two connections (I've set this to 18s, so that checks out):

[2023-03-30T07:34:19.250Z INFO] lost connection to `12D3KooWNiHroU8x2zecGQXRFU3p7mZkujYxHJ5vFBWVxhyxeGXQ`, removing from routing table (cause=Some(IO(Custom { kind: Other, error: Connection(ConnectionError(TimedOut)) })), connected_peers=101, num_established=1)

---

What I am pondering is that the QUIC connection shouldn't time out, as we have keep alive messages. But probably more concerning to me is that I don't see the connection happening on the dialer side, which might explain why the connection times out for the listener as there is actually no connection at all.

[mxinden]

Can you share the code that you are using? Are you dialing the listener twice by any chance, e.g. through libp2p-kads bootstrap?

[b-zee]

https://github.com/b-zee/rust-libp2p/blob/master/examples/safe/src/main.rs

I am most probably dialing twice.

[b-zee]

Correction, I do not see a Dialing event twice. I only see it once from the dialer.

[thomaseizinger]

Mind trying if the following PR fixes this? #3454

[b-zee]

Actually, I have rebased the example on master and I do not observe the problem anymore!