mdns/fix: Failed to register opened substream

[lexnv]

This PR ensures that when MDNS encounters an error it does not terminate other litep2p components.

Previously, if MDNS failed to send a query or to handle the incoming packets it would exit.
The exit is presented by the following log line observed on kusama validator:

tokio-runtime-worker litep2p::mdns: failed to send mdns query error=IoError(NetworkUnreachable)

This situation is causing the substrate Discovery mechanism to also exit, which propagates to the litep2p kademlia handler that exits as well. This leaves the node unable to discover the network or handle incoming substreams.

Testing Done

The issue was reproduced locally with a tokio interval patch that exits the MDNS component after having connectivity in Kusama:

2024-12-11 12:50:34.425 ERROR tokio-runtime-worker litep2p::mdns: interval tick MDNS
2024-12-11 12:50:34.425 ERROR tokio-runtime-worker litep2p::mdns: interval tick expired, closing MDNS

2024-12-11 12:50:35.111 ERROR tokio-runtime-worker litep2p::tcp::connection: failed to register opened substream to protocol protocol=Allocated("/b0a8d493285c2df73290dfb7e61f870f17b41801197a149ca93654499ea3dafe/kad") peer=PeerId("12D3KooWEwh9AwKFUJKPFqmJXWByH7JKYRcfAUfPvp9f3xzj3ibJ") endpoint=Dialer { address: "/ip4/3.96.91.180/tcp/30333", connection_id: ConnectionId(200) } error=ConnectionClosed
...
2024-12-11 12:50:38.753 ERROR tokio-runtime-worker litep2p::tcp::connection: failed to register opened substream to protocol protocol=Allocated("/b0a8d493285c2df73290dfb7e61f870f17b41801197a149ca93654499ea3dafe/kad") peer=PeerId("12D3KooWJb1W7jmqDCaU3Hsh6NRfDo12gnj8hnKfGwA77vRE4jBv") endpoint=Dialer { address: "/ip4/51.38.63.126/tcp/30333", connection_id: ConnectionId(294) } error=ConnectionClosed
2024-12-11 12:50:40.389 ERROR tokio-runtime-worker litep2p::tcp::connection: failed to register opened substream to protocol protocol=Allocated("/b0a8d493285c2df73290dfb7e61f870f17b41801197a149ca93654499ea3dafe/kad") peer=PeerId("12D3KooWGXXuap75AN24aA5XP9S1X3BKqdDbYyHwBTJakMyv1P5V") endpoint=Dialer { address: "/ip4/104.243.41.217/tcp/30330", connection_id: ConnectionId(29) } error=ConnectionClosed
...

2024-12-11 12:53:15.690 ERROR tokio-runtime-worker litep2p::tcp: connection exited with error connection_id=ConnectionId(29) error=EssentialTaskClosed
2024-12-11 12:53:40.071 ERROR tokio-runtime-worker litep2p::tcp::connection: failed to register opened substream to protocol protocol=Allocated("/b0a8d493285c2df73290dfb7e61f870f17b41801197a149ca93654499ea3dafe/kad") peer=PeerId("12D3KooWGphqiEqsfR5ZnV7R2Lgubxi7eAo6MTx3tVmso8oCkvJn") endpoint=Dialer { address: "/ip4/51.163.1.153/tcp/30003", connection_id: ConnectionId(51) } error=ConnectionClosed
2024-12-11 12:53:40.233 ERROR tokio-runtime-worker litep2p::tcp::connection: failed to register opened substream to protocol protocol=Allocated("/b0a8d493285c2df73290dfb7e61f870f17b41801197a149ca93654499ea3dafe/kad") peer=PeerId("12D3KooWM5mnupyiDGtdN6qm3riQDjBbAZfFqAJfMbcbPQbkEn8u") endpoint=Dialer { address: "/ip4/168.119.149.170/tcp/30333", connection_id: ConnectionId(28) } error=ConnectionClosed
2024-12-11 12:53:41.060 ERROR tokio-runtime-worker litep2p::tcp::connection: failed to register opened substream to protocol protocol=Allocated("/b0a8d493285c2df73290dfb7e61f870f17b41801197a149ca93654499ea3dafe/kad") peer=PeerId("12D3KooWGphqiEqsfR5ZnV7R2Lgubxi7eAo6MTx3tVmso8oCkvJn") endpoint=Dialer { address: "/ip4/51.163.1.153/tcp/30003", connection_id: ConnectionId(51) } error=ConnectionClosed
2024-12-11 12:53:42.766 ERROR tokio-runtime-worker litep2p::tcp::connection: failed to register opened substream to protocol protocol=Allocated("/b0a8d493285c2df73290dfb7e61f870f17b41801197a149ca93654499ea3dafe/kad") peer=PeerId("12D3KooWM5mnupyiDGtdN6qm3riQDjBbAZfFqAJfMbcbPQbkEn8u") endpoint=Dialer { address: "/ip4/168.119.149.170/tcp/30333", connection_id: ConnectionId(28) } error=ConnectionClosed

Closes: #300

Thanks @dmitry-markin for also confirming this 🙏

cc @paritytech/networking

[skunert]

Basically execution would be stuck here forever, right? Is it a problem that self.socket will never be polled then below?

[dmitry-markin]

Would tokio::time::Interval work here? It fires immediately the first time, so there will be no need to send the first outbound request manually.

The caveat is to use Delay MissedTickBehavior, otherwise we might end up bursting many packets if for some reason other branches of select! take too long.

[lexnv]

Yep indeed, have ended up using tokio::time::Interval here 🙏