mprpc

mprpc is a lightweight MessagePack RPC library. It enables you to easily build a distributed server-side system by writing a small amount of code. It is built on top of gevent and MessagePack.

This repo adapt to the newest msgpack with a little bit speedup and add numpy support.

Also, this repo has unified the RPCServer and RPCClient usage, making it easier to use.

Installation

To install mprpc, simply:

$ pip install git+https://github.com/zylo117/mprpc

Examples

RPC server

New Style (no need to inherit RPCServer explicitly)

from mprpc import RPCServer

class SumServer:
    def sum(self, x, y):
        return x + y

server = RPCServer('127.0.0.1', 6000, SumServer())
server.serve_forever()

Original Style (Not Recommended)

from gevent.server import StreamServer
from mprpc._server import _RPCServer

class SumServer(_RPCServer):
    def sum(self, x, y):
        return x + y

server = StreamServer(('127.0.0.1', 6000), SumServer())
server.serve_forever()

RPC client

from mprpc import RPCClient

client = RPCClient('127.0.0.1', 6000)
print client.call('sum', 1, 2)

RPC client with connection pooling (generally you don't need it)

import gsocketpool.pool
from mprpc import RPCPoolClient

client_pool = gsocketpool.pool.Pool(RPCPoolClient, dict(host='127.0.0.1', port=6000))

with client_pool.connection() as client:
    print client.call('sum', 1, 2)

Performance

mprpc significantly outperforms the official MessagePack RPC (1.8x faster), which is built using Facebook's Tornado and MessagePack, and ZeroRPC (14x faster), which is built using ZeroMQ and MessagePack.

While this repo has adapt to the newest msgpack with a few of extra features, it's a little bit faster than the original mprpc.

Personal perspective: zerorpc was my personal favorite before this repo. Though zerorpc use zeromq as middleware, which provides a more stable communication for rpc and even better, auto load balancing, it brings a lots of overhead along with it.

Besides, this repo is cython optimized, yet zerorpc is not. So this benchmark it's not entirely fair.

While using mprpc, you should pay attention to the load balancing and job distribution, which might be the real bottleneck someday.

Results

Environment:

Intel i5-8400

Ubuntu 19.10 Desktop x64 (5.3.0-21-generic)

Python3.7

Documentation

Documentation is available at http://mprpc.readthedocs.org/.