Managing hosts and iptables in scalable architecture
- by hakunin
Let's say I have a load balancer in front of 3 app servers.
Let's say I also have these services available at certain IPs:
Postgres server
Redis server
ElasticSearch server
Memcached server 1
Memcached server 2
Memcached server 3
So that's 6 nodes at 6 different IP addresses. Naturally, every one of my 3 app servers needs to talk to these 6 servers above.
Then, to make it a bit funkier, I also have 3 worker servers. And each worker also talks to the above 6 servers, but thankfully workers and apps never need to talk to each other.
Now's the kicker. Everything is on Digital Ocean VPS. What that means is: you have no private network, no private IPs. You only have separate, random IP address on each machine. You can't mask them or anything.
So in order to build a secure environment I would have to configure some iptables. For example:
Open app servers be accessed by load balancer server
Open redis, ES, PG, and each memcached servers to be accessed by each app's IP and each worker's IP
This means that every time I add an app or worker I have to also reconfigure iptables in those above 6 servers to welcome the new app or worker.
Is there a way to simplify this type of setup?
I was thinking — what if there was a gateway machine between apps/workers and the above 6 machines. This way all the interaction would always happen via the gateway server, and when I add a new app or worker I wouldn't need to teach the 6 servers to let it in.
If I went this route, then I'd hope a small 512mb server could handle that perhaps, and there wouldn't be almost any overhead. Or would there?
Please help with best way to handle this situation. I would appreciate an answer as concrete as possible. I don't think this is too specific, because this general architecture is very common, and Digital Ocean is becoming increasingly popular. A concrete solution here would be much appreciated by many.
