rap(par[::-1]) is advanced and fast python async rpc
rap
achieves very fast communication through msgpack
and Python asyncio
and multiplexing transport, while supporting high concurrency.
Implement the protobuf
of Grpc
through Python functions and TypeHint.
Note: The current rap
API may change significantly in subsequent versions
The rap first version feature idea comes from aiorpc
Warning There will be an architectural change in version 0.6
pip install rap
import asyncio
from typing import AsyncIterator
from rap.server import Server
def sync_sum(a: int, b: int) -> int:
return a + b
async def async_sum(a: int, b: int) -> int:
await asyncio.sleep(1) # mock io
return a + b
async def async_gen(a: int) -> AsyncIterator[int]:
for i in range(a):
yield i
loop = asyncio.new_event_loop()
rpc_server = Server() # init service
# register func
rpc_server.register(sync_sum)
rpc_server.register(async_sum)
rpc_server.register(async_gen)
# run server
loop.run_until_complete(rpc_server.create_server())
try:
loop.run_forever()
except KeyboardInterrupt:
# stop server
loop.run_until_complete(rpc_server.shutdown())
The client supports to invoke the service by invoke_by_name
and invoke
methods, but this can not fully use the functions of TypeHint, it is recommended to use @client.register
to register the function and then invoke it.
Note: For rap.client
there is no distinction between async def
and def
, but functions registered with @client.register
can be used directly by the user, so functions decorated with @client.register
should be similar to:
async def demo(): pass
example:
import asyncio
from typing import AsyncIterator
from rap.client import Client
client: "Client" = Client() # init client
# Declare a function with no function. The function name, function type and return type must be the same as the server side function (async def does not differ from def)
def sync_sum(a: int, b: int) -> int:
pass
# The decorated function must be an async def function
@client.register()
async def sync_sum(a: int, b: int) -> int:
pass
# The decorated function must be the async def function, because the function is a generator syntax, to `yield` instead of `pass`
@client.register()
async def async_gen(a: int) -> AsyncIterator:
yield
async def main():
client.add_conn("localhost", 9000)
await client.start()
# Call the invoke method; read the function name and then invoke `invoke_by_name`.
print(f"invoke result: {await client.invoke(sync_sum, {'a': 1, 'b': 2})}")
# Basic calls to rap.client
print(f"raw invoke result: {await client.invoke_by_name('sync_sum', {'a': 1, 'b': 2})}")
# Functions registered through `@client.register` can be used directly
# await async_sum(1,3) == await client.invoke_by_name('async_sum', 1, 2)
# It is recommended to use the @client.register method, which can be used by tools such as IDE to determine whether the parameter type is wrong
print(f"decorator result: {await sync_sum(1, 3)}")
async_gen_result: list = []
# Example of an asynchronous generator, which by default opens or reuses the current session of the rap (about the session will be mentioned below)
async for i in async_gen(10):
async_gen_result.append(i)
print(f"async gen result:{async_gen_result}")
asyncio.run(main())
The server side supports def
and async def
, if it is a def
function, it will be run with multiple threads. When registering, the TypeHints of the function's parameters and return value will be checked, and an error will be reported if the type does not match the type specified by json.
The server comes with a registration library. If there are duplicate registrations in the same group, an error will be reported. You can use the group
parameter to define the group to be registered or redefine the name of the registration with the name
parameter (you also need to specify the corresponding group when the client calls it).
In addition, you can set is_private
to True when registering, so that the function can only be called by the local rap.client.
import asyncio
from typing import AsyncIterator
from rap.server import Server
def demo1(a: int, b: int) -> int:
return a + b
async def demo2(a: int, b: int) -> int:
await asyncio.sleep(1)
return a + b
async def demo_gen(a: int) -> AsyncIterator[int]:
for i in range(a):
yield i
server: Server = Server()
server.register(demo1) # register def func
server.register(demo2) # register async def func
server.register(demo_gen) # register async iterator func
server.register(demo2, name='demo2-alias') # Register with the value of `name`
server.register(demo2, group='new-correlation_id') # Register and set the groups to be registered
server.register(demo2, group='root', is_private=True) # Register and set the correlation_id to be registered, and set it to private
For clients, it is recommended to use client.register
instead of client.invoke
, client.invoke_by_name
.
client.register
uses Python syntax to define function names, arguments, parameter types, and return value types,
It allows the caller to invoke the function as if it were a normal function, and the function can be checked through tools using the TypeHint feature.
Note: When using client.register
, be sure to use async def ...
.
from typing import AsyncIterator
from rap.client import Client
client: Client = Client()
# register func
@client.register()
async def demo1(a: int, b: int) -> int: pass
# register async iterator fun, replace `pass` with `yield`
# Since `async for` will make multiple requests to the same transport over time, it will check if the session is enabled and automatically reuse the current session if it is enabled, otherwise it will create a new session and use it.
@client.register()
async def demo_gen(a: int) -> AsyncIterator: yield
# Register the general function and set the name to demo2-alias
@client.register(name='demo2-alias')
async def demo2(a: int, b: int) -> int: pass
# Register the general function and set the correlation_id to new-correlation_id
@client.register(group='new-correlation_id')
async def demo2(a: int, b: int) -> int: pass
rap
client support session function, after enabling the session, all requests will only be requested through the current session's transport to the corresponding server, while each request, the session_id in the header will set the current session id, convenient for the server to identify.
rap
sessions support explicit and implicit settings, each with its own advantages and disadvantages, without mandatory restrictions.
from typing import AsyncIterator
from rap.client import Client
client = Client()
def sync_sum(a: int, b: int) -> int:
pass
@client.register()
async def async_sum(a: int, b: int) -> int:
pass
@client.register()
async def async_gen(a: int) -> AsyncIterator[int]:
yield
async def no_param_run():
# The rap internal implementation uses the session implicitly via the `contextvar` module
print(f"sync result: {await client.invoke(sync_sum, 1, 2)}")
print(f"async result: {await async_sum(1, 3)}")
# The asynchronous generator detects if a session is enabled, and if so, it automatically reuses the current session, otherwise it creates a session
async for i in async_gen(10):
print(f"async gen result:{i}")
async def param_run(session: "Session"):
# By explicitly passing the session parameters in
print(f"sync result: {await client.invoke(sync_sum, 1, 2, session=session)}")
print(f"sync result: {await client.invoke_by_name('sync_sum', 1, 2, session=session)}")
# May be a bit unfriendly
print(f"async result: {await async_sum(1, 3, session=session)}")
# The asynchronous generator detects if a session is enabled, and if so, it automatically reuses the current session, otherwise it creates a session
async for i in async_gen(10):
print(f"async gen result:{i}")
async def execute(session: "Session"):
# The best way to invoke a session explicitly, using a method similar to the mysql cursor
# execute will automatically recognize the type of invoke
print(f"sync result: {await session.execute(sync_sum, arg_list=[1, 2])}")
print(f"sync result: {await session.execute('sync_sum', arg_list=[1, 2])}")
print(f"async result: {await session.execute(async_sum(1, 3))}")
# The asynchronous generator detects if a session is enabled, and if so, it automatically reuses the current session, otherwise it creates a session
async for i in async_gen(10):
print(f"async gen result:{i}")
async def run_once():
client.add_conn("localhost", 9000)
await client.start()
# init session
async with client.session as s:
await no_param_run()
await param_run(s)
await execute(s)
await client.stop()
channel supports client-server interaction in a duplex manner, similar to Http's WebSocket, it should be noted that the channel does not support group settings.
Only @client.register
is supported on the client side to register the channel function, which is characterized by a single argument of type Channel
.
The channel will maintain a session and will only communicate with the server via a transport from the time the channel is enabled to the time it is closed.
To avoid the use of 'while True', the channel supports the use of 'async for' syntax and the use of 'while await channel.loop()` syntax instead of 'while True
from rap.client import Channel, Client
from rap.client.model import Response
client = Client()
@client.register()
async def async_channel(channel: Channel) -> None:
await channel.write_to_conn("hello") # send data
cnt: int = 0
while await channel.loop(cnt < 3):
cnt += 1
print(await channel.read_body()) # read data
@client.register()
async def echo_body(channel: Channel) -> None:
await channel.write_to_conn("hi!")
# Reads data, returns only when data is read, and exits the loop if it receives a signal to close the channel
async for body in channel.iter_body():
print(f"body:{body}")
await channel.write_to_conn(body)
@client.register()
async def echo_response(channel: Channel) -> None:
await channel.write_to_conn("hi!")
# Read the response data (including header data), and return only if the data is read, or exit the loop if a signal is received to close the channel
async for response in channel.iter():
response: Response = response # help IDE check type....
print(f"response: {response}")
await channel.write_to_conn(response.body)
Due to the high degree of encapsulation of the Python asyncio
module, rap
can be used very easily with ssl
# Quickly generate ssl.crt and ssl.key
openssl req -newkey rsa:2048 -nodes -keyout rap_ssl.key -x509 -days 365 -out rap_ssl.crt
client.py
from rap.client import Client
client = Client(ssl_crt_path="./rap_ssl.crt")
server.py
from rap.server import Server
rpc_server = Server(
ssl_crt_path="./rap_ssl.crt",
ssl_key_path="./rap_ssl.key",
)
The server side supports start_event
and stop_event
for event handling before start and after shutdown respectively.
from rap.server import Server
async def mock_start():
print('start event')
async def mock_stop():
print('stop event')
# example 1
server = Server(start_event_list=[mock_start()], stop_event_list=[mock_stop()])
# example 2
server = Server()
server.load_before_start_event([mock_start()])
server.load_after_stop_event([mock_stop()])
rap
currently supports 2 types of middleware::
- Conn middleware: Used when creating transport, such as limiting the total number of links, etc...
reference block.py,
The
dispatch
method will pass in a transport object, and then determine whether to release it according to the rules (return await self.call_next(transport)) or reject it (await transport.close) - Message middleware: only supports normal function calls (no support for
Channel
), similar to the use ofstarlette
middleware reference access.py Message middleware will pass in 4 parameters: request(current request object), call_id(current invoke id), func(current invoke function), param(current parameter) and request to return call_id and result(function execution result or exception object)
In addition, the middleware supports start_event_handle
and stop_event_handle
methods, which are called when the Server
starts and shuts down respectively.
example:
from rap.server import Server
from rap.server.plugin.middleware import ConnLimitMiddleware
rpc_server = Server()
rpc_server.load_middleware([ConnLimitMiddleware()])
The rap
processor is used to handle inbound and outbound traffic, where on_request
is for inbound traffic and on_response
is for outbound traffic.
The methods of rap.client
and rap.server
processors are basically the same, rap.server
supports start_event_handle
and stop_event_handle
methods, which are called when Server
starts and shuts down respectively
server crypto processor example
client crypto processor example
client load processor example
from rap.client import Client
from rap.client.processor import CryptoProcessor
client = Client()
client.load_processor([CryptoProcessor('key_id', 'xxxxxxxxxxxxxxxx')])
server load processor example
from rap.server import Server
from rap.server.plugin.processor import CryptoProcessor
server = Server()
server.load_processor([CryptoProcessor({'key_id': 'xxxxxxxxxxxxxxxx'})])
rap supports plug-in functionality through middleware
and processor
, middleware
only supports the server side, processor
supports the client and server side
Encrypted transmission only encrypts the request and response body content, not the header etc. While encrypting, the nonce parameter is added to prevent replay, and the timestamp parameter is added to prevent timeout access.
client example:
from rap.client import Client
from rap.client.processor import CryptoProcessor
client = Client()
# The first parameter is the id of the secret key, the server determines which secret key is used for the current request by the id of the secret key
# The second parameter is the key of the secret key, currently only support the length of 16 bits of the secret key
# timeout: Requests that exceed the timeout value compared to the current timestamp will be discarded
# interval: Clear the nonce interval, the shorter the interval, the more frequent the execution, the greater the useless work, the longer the interval, the more likely to occupy memory, the recommended value is the timeout value is 2 times
client.load_processor([CryptoProcessor("demo_id", "xxxxxxxxxxxxxxxx", timeout=60, interval=120)])
server example:
from rap.server import Server
from rap.server.plugin.processor import CryptoProcessor
server = Server()
# The first parameter is the secret key key-value pair, key is the secret key id, value is the secret key
# timeout: Requests that exceed the timeout value compared to the current timestamp will be discarded
# nonce_timeout: The expiration time of nonce, the recommended setting is greater than timeout
server.load_processor([CryptoProcessor({"demo_id": "xxxxxxxxxxxxxxxx"}, timeout=60, nonce_timeout=120)])
Server-side use only, you can limit the maximum number of links on the server side, more than the set value will not handle new requests
from rap.server import Server
from rap.server.plugin.middleware import ConnLimitMiddleware, IpMaxConnMiddleware
server = Server()
server.load_middleware(
[
# max_conn: Current maximum number of transport
# block_timeout: Access ban time after exceeding the maximum number of transport
ConnLimitMiddleware(max_conn=100, block_time=60),
# ip_max_conn: Maximum number of transport per ip
# timeout: The maximum statistics time for each ip, after the time no new requests come in, the relevant statistics will be cleared
IpMaxConnMiddleware(ip_max_conn=10, timeout=60),
]
)
Support restrict single ip or whole segment ip, support both whitelist and blacklist mode, if whitelist is enabled, blacklist mode is disabled by default
from rap.server import Server
from rap.server.plugin.middleware import IpFilterMiddleware
server = Server()
# allow_ip_list: whitelist, support network segment ip, if filled with allow_ip_list, black_ip_list will be invalid
# black_ip_list: blacklist, support network segment ip
server.load_middleware([IpFilterMiddleware(allow_ip_list=['192.168.0.0/31'], block_ip_list=['192.168.0.2'])])
TODO, This feature is not yet implemented
TODO, Document is being edited
TODO, Document is being edited