Google’s gRPC provides a framework for implementing RPC (Remote Procedure Call) workflows. By layering on top of HTTP/2 and using protocol buffers, gRPC promises a lot of benefits over conventional REST+JSON APIs.

Considering the promised goodies, I decided to get my hands dirty and roll gRPC for some of the service-oriented environments at Semantics3.

I headed over to the official documentation, opened the section for my current language of choice (Python), and promptly got lost in the all the pre-written code and black magic that seemed to happen under the hood.

This post is an attempt to start from scratch, take a simple function and expose it via a gRPC interface.

So, let’s get building.

0. Define the function

Let’s create a function (procedure) that we want to expose (remotely call) — square_root, located in

square_root take an input x and returns the square root as y. The rest of this post will focus on how square_root can be exposed via gRPC.

1. Set up protocol buffers

Protocol buffers are a language-neutral mechanism for serializing structured data. Using it comes with the requirement to explicitly define values and their data types.

Let’s create calculator.proto, which defines the message and service structures to be used by our service.

You can think of the message and service definitions as below:

  • Number.value will be used to contain variablesx andy
  • Calculator.SquareRoot will be used for the function square_root

2. Generate gRPC classes for Python

This section is possibly the most “black-boxed” part of the whole process. We will be using special tools to automatically generate classes.

New files (and classes), following certain naming conventions, will be generated when running these commands. (You can refer to the documentation on the various flags used. In this post, all files are located in a single folder and the commands are run in that same folder.)

$ pip install grpcio
$ pip install grpcio-tools

$ python -m grpc_tools.protoc -I. --python_out=. --grpc_python_out=. calculator.proto

The files generated will be as follows: — contains message classes

  • calculator_pb2.Number for request/response variables (x and y) — contains server and client classes

  • calculator_pb2_grpc.CalculatorServicer for the server
  • calculator_pb2_grpc.CalculatorStub for the client

3. Create a gRPC server

We now have all the pieces required to create a gRPC server, as below. Comments, inline, should explain each section.

We can start the server using the command,

$ python
Starting server. Listening on port 50051.

Now we have a gRPC server, listening on port 50051.

4. Create a gRPC client

With the server setup complete, we create — which simply calls the function and prints the result.

That’s it!

With the server already listening, we simply run our client.

$ python

Taking it from the top

All the files can be found on GitHub at ramananbalakrishnan/basic-grpc-python. For quick reference, here is what each file is used for.

├──          # module containing a function
├── calculator.proto       # protobuf definition file
├── # generated class for server/client
├──      # generated class for message
├──              # a server to expose the function
└──              # a sample client

This post, using a very simple example to convert a function into a remote procedure, just scratches the surface.

Of course, gRPC can be used in more advanced modes (request-streaming, response-streaming, bidirectional-streaming) with additional features such as error-handling and authentication. But hey, we all have to begin somewhere and I hope this post serves as a good reference for those just starting out.