手把手教你写一个自己的 master-slave 架构的 TCP 服务器

nginx、uvicorn、gunicorn 这些 HTTP sever 都是 master-slave 架构

你好奇他们是怎么实现的吗？

如果你是也是一个好奇宝宝，就接着往下看吧

多进程

参考：python 进程池的两种不同实现

使用 multiprocessing.Pool 的实现

import os
import socket
import sys
import time
import threading
from loguru import logger
from concurrent.futures import ThreadPoolExecutor
from concurrent.futures._base import Future
import multiprocessing

default_encoding: str = 'utf-8'

pool = ThreadPoolExecutor(
    max_workers=20,
    thread_name_prefix='simple-work-thread-pool'
)


def init_serversocket() -> socket.socket:
    serversocket = socket.socket(
        family=socket.AF_INET,
        type=socket.SOCK_STREAM
    )

    # 获取本地主机名
    host = socket.gethostname()

    logger.debug(f'host {host}')

    port = 6001

    # 绑定端口号
    serversocket.bind(('0.0.0.0', port))

    # 设置最大连接数，超过后排队
    serversocket.listen(2048)

    return serversocket


def send_response(clientsocket: socket.socket, addr: tuple, response_body: bytes) -> int:
    send_len: int = clientsocket.send(response_body)
    clientsocket.close()
    return send_len


def start_request(clientsocket: socket.socket, addr: tuple) -> int:
    try:
        pid = os.getpid()
        logger.debug(f'pid: {pid}, get message from {addr}')
        request_body: bytes = clientsocket.recv(2048)
        request_text: str = request_body.decode(encoding=default_encoding)

        response_text: str = f'server get message: {request_text}'

        response_body: bytes = response_text.encode(default_encoding)
        # time.sleep(1)
        send_len = send_response(
            clientsocket=clientsocket, addr=addr, response_body=response_body)
        logger.debug(f'发送了响应')
        return send_len
    except Exception as error:
        logger.exception(error)


def start_request_callback(future: Future) -> None:
    send_len: int = future.result()
    logger.debug(
        f'{threading.current_thread().name}, send payload len is {send_len}')


if __name__ == "__main__":
    serversocket = init_serversocket()

    pool = multiprocessing.Pool(processes=16)

    while True:
        clientsocket, addr = serversocket.accept()

        clientsocket: socket.socket
        addr: tuple

        # future: Future = pool.submit(start_request, clientsocket, addr)
        # future.add_done_callback(start_request_callback)

        pool.apply_async(start_request, (clientsocket, addr))

    pool.close()
    pool.join()

使用 ProcessPoolExecutor 的实现

import os
import socket
import sys
import time
import threading
from loguru import logger
from concurrent.futures._base import Future
import multiprocessing
from concurrent.futures import ProcessPoolExecutor

default_encoding: str = 'utf-8'


def init_serversocket() -> socket.socket:
    serversocket = socket.socket(
        family=socket.AF_INET,
        type=socket.SOCK_STREAM
    )

    # 获取本地主机名
    host = socket.gethostname()

    logger.debug(f'host {host}')

    port = 6001

    # 绑定端口号
    serversocket.bind(('0.0.0.0', port))

    # 设置最大连接数，超过后排队
    serversocket.listen(2048)

    return serversocket


def send_response(clientsocket: socket.socket, addr: tuple, response_body: bytes) -> int:
    send_len: int = clientsocket.send(response_body)
    clientsocket.close()
    return send_len


def start_request(clientsocket: socket.socket, addr: tuple) -> int:
    try:
        pid = os.getpid()
        logger.debug(f'pid: {pid}, get message from {addr}')
        request_body: bytes = clientsocket.recv(2048)
        request_text: str = request_body.decode(encoding=default_encoding)

        response_text: str = f'server get message: {request_text}'

        response_body: bytes = response_text.encode(default_encoding)
        # time.sleep(1)
        send_len = send_response(
            clientsocket=clientsocket, addr=addr, response_body=response_body)
        logger.debug(f'发送了响应')
        return send_len
    except Exception as error:
        logger.exception(error)


def start_request_callback(future: Future) -> None:
    send_len: int = future.result()
    logger.debug(
        f'{threading.current_thread().name}, send payload len is {send_len}')


if __name__ == "__main__":
    serversocket = init_serversocket()

    # pool = multiprocessing.Pool(
    #     processes=16,
    #     mp_context=multiprocessing.get_context('spawn')
    # )

    pool = ProcessPoolExecutor(
        max_workers=multiprocessing.cpu_count(),
        mp_context=multiprocessing.get_context('spawn')
    )

    while True:
        clientsocket, addr = serversocket.accept()

        clientsocket: socket.socket
        addr: tuple

        # future: Future = pool.submit(start_request, clientsocket, addr)
        # future.add_done_callback(start_request_callback)
        
        pool.submit(start_request,clientsocket, addr)

        # pool.apply_async(start_request, (clientsocket, addr))

    pool.close()
    pool.join()

上诉两种实现的潜在问题

问题一：无法完美运行在 mac 平台

上面两种方式在 Linux 上都可以工作的良好，但是在 mac 上却不行

服务端会有很大概率报错（客户端请求的时候，随机出现报错）：

• ConnectionRefusedError: [Errno 61] Connection refused
• concurrent.futures.process.BrokenProcessPool: A child process terminated abruptly, the process pool is not usable anymore

问题二：master 负载太高，容易成为瓶颈，无法实现横向扩展

master 进程都干冒烟了，slave 都闲着

为什么呢？因为 slave 干的事情太少了，而 master 进程负责了 serversocket.accept 等等操作，压力山大

改进版：支持横向扩展的 master-slave 结构