Cluster connection management - aka load balancer

[Terhiman]

When I front NodeJS with say NGINX I can specify a server pool and add/remove servers from that pool simply by saying whether it's down or not. Once removed, connections drain organically over time and when a server (NodeJS instance) has none I can restart it thus allowing me to do rolling code updates which for some purposes is an ok to do thing imo.

My question is, how could I go about doing the same thing internally to a NodeJS Cluster? My understanding is the master cluster worker load balances connections between workers using 1 of 2 methods but what if I wanted to take a worker out of the "pool", wait until existing connections drop off over time and when zero just kill the worker and re-fork a new one thus adopting any new code I may have added/changed.

Maybe there is already some code out there that allows me to do this? I guess what I am asking for is a way to tell the master worker that a new connection should not be connected to a specified worker. I guess I could set a flag in the child worker and it could redirect the incoming request and in a round robin environment that might work, but least busy might not and we get a race situation.

[cjihrig]

If I understand correctly, it sounds like you may want to look into worker.disconnect(). Take a look at the code sample for that method as well. You'll likely need to write some custom code or look for something on npm as well.

[Terhiman]

Thank you, exactly what I needed. I had read this in the docs before but was confused when the terminology started referring to the server closing when in fact that mean the individual "server" that was the worker thread itself.

"After a server is closed, it will no longer accept new connections, but connections may be accepted by any other listening worker."