all of the popular rpc framework today, is fast and stable.
but it is kind of not good enough for you to easily start. you have to accept all the rpc's definition,learn their structure, and you have to adapt your code to them.
so if you dont like them,and you havn't have other alternatives. may be SimRpc is suit for you
SimRpc is an rpc tools for you to easily transfer your code to rpc service(both client-side and server-side),built on top of zeromq and msgpack.
-
zeromq
assures the message transfer extremely fast and stable
-
msgpack
enables the message blazing small
class a:
def minus(self, x, y):
print("x+y=", x - y)
return x - y
def add(x, y):
return x + y
from simrpc.client import SimRpcClient
rpc_client = SimRpcClient()
or maybe you want to start a specific address
from simrpc.client import SimRpcClient
rpc_client = SimRpcClient(server_address="tcp://localhost:6666")
the server_address is a zeromq address.
for more information you can refer to zeromq.
# response_only=True means you only want the result
class a:
@rpc_client.task(response_only=True)
def minus(self, x, y):
print("x+y=", x - y)
return x - y
# func=True means this is a funcion,not a class service
@rpc_client.task(response_only=True,func=True)
def add(x, y):
return x + y
a().minus(4,5,rpc_client=True)
you must point it out this a rpc client(as use rpc_client=True prameter), otherwise will use local code
from server import SimRpcServer
server = SimRpcServer()
like client,you can also give specific addresses to server.
parameter | value | instruction |
---|---|---|
device_front_address | "tcp://127.0.0.1:5559" | address that receive message. used for client to connect. |
device_backend_address | "tcp://127.0.0.1:5560" | address that send message to service worker. |
worker_address | "tcp://127.0.0.1:5560" | address that worker connect to. usually the same with device_backend_address.for advanced infomation,you can refer to zeromq |
more ways of address(like ipc) are supported, you can refer to zeromq
- initial your class service as usual
service = a()
server.register(service)
- or you let server to initial your service(just pass your init arguments to server.)
settings = {
"a": {
"kwargs": {
"name": "test"
}
}
}
service = a()
server.register(service)
# or you can have this class init with server
# server.register_with_init(cls_list=[a], settings=settings)
server.start_broker()
# if you also want to start worker in process ,you could use server.run(),this will start a few workers.
# server.run()
with this broker,you can easily expand your service from both client-side and server-side
# the same as above but without server.start_broker()
server.run()
- like this project, star it
- any suggestion is welcome
- this project is under MIT lisense
more docs are writing.