Serve multiple services on the same server

[Laci556]

Hey! Is it possible to expose multiple services by the same server?
We've run into some limitations with out previous grpclib-based services, so I decided to use connect for our next service. I'd like to separate the service into multiple smaller ones and serve them all on a single server. As far as I can tell, connect generates an ASGI application per service and I can't find a way to "merge" them. grpclib's server accepts a collection of servables, is there something similar for connect?

[i2y]

Hi @Laci556!

Yes, you can absolutely serve multiple services on the same server. ASGI applications are composable, and connect-python is designed with this in mind - each generated application class (e.g., ElizaServiceASGIApplication) has a .path property specifically for mounting.

Using Starlette (recommended)

from starlette.applications import Starlette
from starlette.routing import Mount

from your_service1_connect import Service1ASGIApplication
from your_service2_connect import Service2ASGIApplication

service1_app = Service1ASGIApplication(YourService1Impl())
service2_app = Service2ASGIApplication(YourService2Impl())

app = Starlette(
    routes=[
        Mount(service1_app.path, service1_app),
        Mount(service2_app.path, service2_app),
    ]
)

Without any extra dependencies

from your_service1_connect import Service1ASGIApplication
from your_service2_connect import Service2ASGIApplication

service1_app = Service1ASGIApplication(YourService1Impl())
service2_app = Service2ASGIApplication(YourService2Impl())

apps = {
    service1_app.path: service1_app,
    service2_app.path: service2_app,
}

async def app(scope, receive, send):
    if scope["type"] != "http":
        return

    path = scope["path"]
    for prefix, service_app in apps.items():
        if path.startswith(prefix):
            await service_app(scope, receive, send)
            return

    await send({"type": "http.response.start", "status": 404, "headers": []})
    await send({"type": "http.response.body", "body": b"Not Found"})

Note: The dependency-free version skips lifespan handling for simplicity. If you need proper lifespan support (for startup/shutdown hooks), Starlette is recommended as it handles this correctly.

You can also check out example/example/eliza_service.py in the repo for a working example using Starlette.

Hope this helps!

[Laci556]

Thanks @i2y! I went with Starlette based on the CORS middleware example in the readme and was able to put it together but your example is even cleaner. We're already using FastAPI for the REST service so the dependency on Startlette is not an issue + it evens allows us to share middlewares between the two.