Peer discovery not working in 0.9.43

[tmpolaczyk]

Peer discovery seems to be broken in the 0.9.43 release. This affects polkadot as well as template parachains based on cumulus, etc. It is impossible to start a local network unless one of the nodes is added as a boot node.

Steps to reproduce: launch 2 nodes in 2 separate terminals and verify how the peer count stays at 0.

./polkadot build-spec --chain rococo-local --disable-default-bootnode --raw > rococo-local-cfde.json
./polkadot --chain rococo-local-cfde.json --alice --tmp
./polkadot --chain rococo-local-cfde.json --bob --tmp

Sample logs:

2023-06-26 16:31:51 💤 Idle (0 peers), best: #0 (0x1f9b…6b45), finalized #0 (0x1f9b…6b45), ⬇ 0 ⬆ 0    

This last worked in 0.9.42, can be easily reproduced by downloading the release binary and running the above commands.

Sample logs (0.9.42):

2023-06-26 16:37:16 💤 Idle (1 peers), best: #1 (0x4715…e3ed), finalized #0 (0xcbcd…e878), ⬇ 1.5kiB/s ⬆ 1.3kiB/s    

[altonen]

It looks like 0.9.43 is able to discover 0.9.42 but not vice versa so there could be something wrong in the mDNS advertisement 0.9.43 sends

[altonen]

Fixed in paritytech/substrate#14478

[ntn-x2]

The 0.9.43 branch at commit ba42b9c is still broken. Does this mean Polkadot 0.9.43 is broken for local relaychain environments? I don't see the commit being backported to the release branch, as the last commit is dated 08/06.

[altonen]

How is ba42b9c related to peer discovery? Please open a new issue if you believe there is a problem with 0.9.43

[ntn-x2]

I am just saying that the HEAD of the 0.9.43 branch is the commit I linked above, which is dated 08/06. The fix was merged into Substrate late June, so this means that 0.9.43 branch is still broken (that's how I ended up on here). So I think this issue should actually be re-opened, since it has not been fixed, really. At least that's not what the lockfile says.

[altonen]

I see. I don't believe our backporting policies apply to any kind of bug fix but somebody can correct me on that. I'm fine with backporting this to 0.9.43 though if it's something we'd want to do

cc @bkchr @coderobe

[bkchr]

Node side changes can be backported, this also looks like quite a small change. So, should be fine.