rsocket / rsocket-java

Java implementation of RSocket
http://rsocket.io
Apache License 2.0
2.35k stars 354 forks source link

Support for Multiple RSocket Handlers #974

Open rstoyanchev opened 3 years ago

rstoyanchev commented 3 years ago

RSocketConnector and RSocketServer can be configured with a SocketAcceptor that handles all requests. A framework can build on this low level mechanism and decide how to route requests internally. However there is no easy way to have multiple frameworks, e.g. rsocket-rpc-java and spring-messaging, side by side in the same application.

Given the existence of the Routing Metadata Extension, RSocket Java could provide a mechanism to delegate requests to one of multiple RSocket handlers depending on the route. This could be as simple as a route prefix that helps to decide which RSocket should handle the request.

OlegDokuka commented 3 years ago

Drafted a diagram of a possible design of the implementation image

1) Decode given metadata to find encoded route via MetadataDecoder (note, a requester side should use a MetadataEncoder to encode route in the payload metadata 2) Find a Handler Function based on string route in the Router 3) Handle found function depends on the interaction type

yschimke commented 3 years ago

Playing devil's advocate - Isn't this the point of the framework code above rsocket? Should this be part of the base library?

I know it's awkward if rsocket-rpc-java and spring-messaging are peers that are unaware of the other, but seems like something would need to configure how to decide where to delegate calls to. That logic seems likely to be similar to any framework like spring or rsocket-rpc.

No strong opinion here against, just probing.

OlegDokuka commented 3 years ago

Actually, this logic is in rsocket-rpc as of now, but, after a few discussions, @bsideup and I ended up that rsocket-router logic should sit in a different place than rpc. Thus it is either should be rsocket-router-java or rsocket-java with submodule rsocket-router

yschimke commented 3 years ago

Makes sense. So I'd change my argument to be that it shouldn't be that routing has pride of place in rsocket-java. But a module allowing frameworks to coexist, has an attraction.