Gossipsub behaviour's add_explicit_peer with address

[lukaszrzasik]

I've been testing a standalone gossipsub behaviour. I had to take care of the peer discovery because there's no other behaviour which could do it of course. I thought that using add_explicit_peer together with Swarm::add_peer_address would solve that problem but it didn't. The problem turned out to be that the dial failed because the peer's address was unknown even though I called Swarm::add_peer_address before that.

I've ended up modifying add_explicit_peer to take an additional parameter Option<Vec<Multiaddr>>.

Maybe it makes sense to add a method something like add_explicit_peer_with_addr which could be called when handling FromSwarm::NewExternalAddrOfPeer or something similar?

If it sounds good I can prepare a PR.

[elenaf9]

The problem turned out to be that the dial failed because the peer's address was unknown even though I called Swarm::add_peer_address before that.

That shouldn't happen; if a peer address was added the peer should be dialable as long as the address is reachable. Could you please provide some logs?

[lukaszrzasik]

Yes, I will but first I want to check with the latest libp2p version. I've been using 0.54 so far.

[lukaszrzasik]

test.log

I've attached the logs. What the code does is basically this:
in a loop:

1. Checks whether the node is connected to the expected number of peers by using gossibsub behaviour's all_peers.
2. If not, tries to add peers by first calling Swarm's add_peer_address and then Behaviour's add_explicit_peer.
3. Then awaits on Swarm's select_next_some with a timeout.

We can see in the logs that the node prints: Connecting to explicit peer but fails.
As far as I can tell it fails because DialOpts constructed by check_explicit_peer_connection does not include the address and then Swarm's dial fails because there's no address.
NewExternalAddrOfPeer emitted by add_peer_address is not handled in any way by the gossipsub behaviour.

My way of making it work just for the test's sake was to modify add_explicit_peer so that it accepts a second parameter, an address, and then check_explicit_peer_connection constructs DialOpts that includes the address. This allows dialling the peer successfully.

I might be doing something wrong, I'm quite new to libp2p so I'm open for any suggestions. Thanks!

[dariusc93]

I think the problem here is the use of Swarm::add_peer_address since it will only emit an FromSwarm::NewExternalAddrOfPeer to behaviours and I dont believe libp2p-gossipsub handles that event. In this case, I think the peer-store, whenever it is available, might be of use (I dont think it is published, although you can use it by pointing to the git repo, especially if its a binary project and not a library expected to be published), however until then you can:

A. Create a small behaviour to capture the NewExternalAddrOfPeer event and store it into a map that can be read by NetworkBehaviour::handle_pending_outbound_connection; or
B. Dial the peer and ensure there is a connection before using gossipsub.

[jxs]

Yeah Darius is right, I am curious, why are you using add_explicit_peer instead of just dialing the peer via Swarm?
Explicit Peering has a specific purpose, defined in gossipsub 1.1 which you can read more about

[lukaszrzasik]

Yes, of course, calling dial directly works. I guess I misused add_explicit_peer. Thank you for the help! I'll close the discussion.