[MM-62064] Leave nodeID empty when using RTCD service

[streamer45]

Summary

This is slightly embarrassing. I introduced a regression in #908 because I cannot read my own code. The second statement in that PR (that we'd be starting up the embedded RTC server regardless of whether RTCD is used) is wrong as there was a return statement in the block so that change wasn't required at all.

And while making that change, I forgot the importance of setting p.nodeID , which should only be done when RTCD is not in use.

Unfortunately, e2e tests couldn't spot this since they run in a non-HA (created https://mattermost.atlassian.net/browse/MM-62065). I ran some basic manual test in HA myself but didn't test multiple users joining a call.

In practice, this means we'll need to prepare a couple more dot releases (1.3.2 and 0.29.6) to remedy the previous mistake, along with backports to respective MM versions.

@DHaussermann For QA, I'd like to ensure we can connect at least a couple of people to the same call in a HA installation, and ensure they can communicate both ways. Repeating this a couple of times may be beneficial since it's always possible they'd connect to the same HA node which is a case not affected by the issue we are fixing here.

Ticket Link

https://mattermost.atlassian.net/browse/MM-62064

[cpoile]

Hmm, interesting. I didn't catch that either. Thank you for the comments, that definitely helps.

[DHaussermann]

Tested and passed

I have a local HA setup in place (thanks to Agniva) where I can explicitly send users to a specific cluster instance.

• I joined a few calls from 4 different browser sessions. 2 sessions where on each instance.
• No issues using calls across the 4 sessions
• By selectively adding and removing users I was able to see that the audio is making it back and forth across the 2 instances.
• Tested with and without RTCD
• Repeated the testing on a /cloud HA server as well as a precaution. No guarantee who is on which instance but I had 4 unique sessions.

LGTM!