Calling Python across the network with gRPC

What if calling a function on another machine felt exactly like calling one in your own code? That's the promise of gRPC, Google's open-source framework for remote procedure calls. I built a small server to see how it feels in Python.

What it does

One job, done properly: ask the server for a random number and it generates one, hands it back, and writes it to a Postgres database.

It's a deliberately tiny example, so the moving parts are easy to see.

How it fits together

• The contract. Everything starts with a rng.proto file that describes the service and its messages.
• Generated code. grpcio-tools turns that proto file into Python modules the server builds on. The command lives in the Makefile, so it's just make generate.
• The server. It implements the service and stores a number in Postgres after every call.
• The packaging. Docker Compose runs the server next to its Postgres database as one small network.

Poking it

To test it, I use BloomRPC: import the proto file, hit a button, see a number. That's a testing convenience, though.

In the real world, a gRPC client would make these calls from your own services. The point of the whole exercise is that the client hands the work to the server and gets the result back. Voilà, compute offloaded.

---

Run it locally

1. Install BloomRPC and import rng.proto.
2. Install Python (I'm on 3.12).
3. Run make install.
4. Run make run.

Run it in Docker

1. Install BloomRPC and import rng.proto.
2. Copy .env.example to .env.
3. Run make docker-run, or docker-compose up --build -d.

Then the fun part

Open BloomRPC, point it at 127.0.0.1:50051, call the function, and watch the random numbers roll in.