Anything wrong with this cluster design idea?

[Konrad Piwowar]

Hey Guys,

We're about to set this up, and I just wanted to check if there's maybe any technical reason or contraindications to such a setup. We tried it in the past back in 2.0-era - and it worked - but the daemon keep crashing on us for unrelated reasons.

The idea is as follows:

- We have 5 geographically distributed servers running Antmedia. All of these servers are monsters with 128-256gb RAM and single or dual AMD Epyc.

- We have a domain name that is running via geoip dns, so depending on location resolves to a different IP. The regions are setup to resolve to specific servers.

- The same SSL cert would be installed on all servers.

- Broadcasters would broadcast to domain.com, thus getting sent to their closest server, since in cluster mode all streams are available to all servers.

- Users would pull from domain.com, thus pulling from their closest server, due to the above.

There are no more then 30-35 broadcasters at any given time and no more then 200 viewers at any given time.

Is there anything that I maybe do not know that would cause the above to be problematic?

[Konrad Piwowar]

Just to elaborate on the above, we wouldn't be using nginx or haproxy or anything - we'd be hitting antmedia directly and spreading this all via geoip DNS without a load balancer.

 

 

[Murat Eminoglu]

Hello, thank you for this good question.

 

The structure you will set up will work and there will be no problem. However, I have suggestions for you.

 

1. Node-to-node TCP connections must be open.  For security reasons, you should assign Static IP to your instances and it should be configured to respond to requests from those IP addresses only, so you will not have any problems with opening TCP ports.

2. Likewise, for 27017 MongoDB, you need to open this port to node Public IPs.

 

If you have more questions, please don't hesitate to contact me