ruby/rails grpc server restart without disconnecting clients

Question

I am using the following code snippet to start a grpc server which works fine. But whenever I need to deploy new code to the server, what is the right way for me to restart the server? Should I just kill the server process, and let client to handle the error message? Or is there a way for enabling master/worker mode like unicorn does?

s = GRPC::RpcServer.new
s.run_till_terminated

Answer 1

There is no such support for rolling out new deployments that's built in to the ruby-gRPC.

However, it should be possible for applications with multiple server instances to do rolling restarts. Eg, note that if gRPC connects to a server and starts to make RPC's to it and that server gets shut down, then gRPC will internally notice that the connection went bad and it will try to make its next RPC on a newly connection (the default gRPC behavior will be to perform its next RPC on the next resolved address that can be successfully connected to, and this might mean reconnecting to the same address for which the connection just broke). Note too that gRPC servers use SO_REUSEPORT by default, so one could potentially run multiple servers on the same port.