mdns storm

[ineiti]

I'm using the latest version of the webrtc crate to connect nodes between servers and webclients. Recently I had problems in my home network, and it turned out that my nodes running on my server using webrtc create mdns storms:

08:32:16.365924 fwpr100p0 Out IP fricklebox.fritz.box.mdns > mdns.mcast.net.mdns: 0 A (QM)? 53854a1e-1c02-4df1-932a-c3f8bad7784f.local. (60)
08:32:16.365924 fwpr100p0 Out IP fricklebox.fritz.box.mdns > mdns.mcast.net.mdns: 0 A (QM)? 53854a1e-1c02-4df1-932a-c3f8bad7784f.local. (60)
08:32:16.365926 eno4  Out IP fricklebox.fritz.box.mdns > mdns.mcast.net.mdns: 0 A (QM)? 53854a1e-1c02-4df1-932a-c3f8bad7784f.local. (60)
08:32:16.365928 eno1  Out IP fricklebox.fritz.box.mdns > mdns.mcast.net.mdns: 0 A (QM)? 53854a1e-1c02-4df1-932a-c3f8bad7784f.local. (60)
08:32:16.365928 eno4  Out IP fricklebox.fritz.box.mdns > mdns.mcast.net.mdns: 0 A (QM)? 53854a1e-1c02-4df1-932a-c3f8bad7784f.local. (60)

Lots and lots of them. If I correlated this correctly, it tries to find the mdns of a remote web-client, which will never be in the local network.

Is there something to do to calm down these mdns storms? I tried to look how to configure the mdns part in the documentation, but didn't find how to do that.

The nodes are running in a docker container - might this impact how mdns interacts with the network?

[ineiti]

I added the following lines to the setup of the APIBuilder from the /examples directory. I'm not really sure what the consequences of it are with regard to running it on the local network. I guess it still will manage to find the nodes locally.

Also not sure if I should keep the issue open or not. Without turning off the mdns module, I still have the flood of messages.

        // There seems to be some trouble with mdns where it can flood the local network
        // with requests - so turn it off.
        let mut setting_engine = SettingEngine::default();
        setting_engine.set_ice_multicast_dns_mode(MulticastDnsMode::Disabled);

        // Create the API object with the MediaEngine
        let api = APIBuilder::new()
            .with_media_engine(m)
            .with_interceptor_registry(registry)
            .with_setting_engine(setting_engine)
            .build();

[r-byondlabs]

I see lot of mdns error messages when a peer connection is closed. It seems like the mdns connection is not stopped when the peer connection is closed. Tracking down the call from peer_connection.stop() leads to here:

webrtc/ice/src/agent/mod.rs

Line 342 in fc3a0aa

pub async fn close(&self) -> Result<()> {

I think this should have a close_multicast_conn() here somewhere.

I'm disabling mdns for now using the above method.

[ineiti]

Ah, that might be it. My nodes do reconnect quite often, which might lead to having lots of mdns requests going on.

I didn't try it out yet, but here is a PR: #636

@r-byondlabs - is that what you think should happen?

[r-byondlabs]

@ineiti yup, seems right to be me 👍🏽

[ineiti]

Unfortunately this doesn't fix it - I just had the same problem as above with the printer not being reachable due to excessive mDNS packets from webrtc...

@rainliu - can you re-open this issue? Or shall I create a new one?