python的多进程共享内存操作出现的同步不一致问题 —— 多进程操作的缓存不一致问题

点击查看代码

from multiprocessing import Process
from multiprocessing import Queue
import numpy as np
from numpy import ndarray
from typing import Callable, Dict, List, Tuple

"""目前主要支持gymnasium环境，gym环境未成功支持"""
from gymnasium import Env
# try:
#     import gymnasium as gym
# except ImportError:
#     import gym

# from gym import Env


"""
    多进程实现：
        以子进程的方式运行游戏环境并与主进程交互数据
"""
class ChildEnv(Process):

    def __init__(self, id:int, env_fns:List[Callable[[], Env]], shared_data:Dict[str, ndarray], queue:Queue, barrier:Queue):
        super(ChildEnv, self).__init__()
        self.id = id    # 子进程的编号
        self.env_fns = env_fns

        self.actions = shared_data["actions"]
        self.rewards = shared_data["rewards"]
        self.next_obs = shared_data["next_obs"]
        self.next_obs_truncated = shared_data["next_obs_truncated"]
        self.terminated = shared_data["terminated"]
        self.truncated = shared_data["truncated"]

        self.queue = queue
        self.barrier = barrier

    def run(self):
        super(ChildEnv, self).run()

        envs = [env_fn() for env_fn in self.env_fns]
        for i, env in enumerate(envs):
            self.next_obs[i], _ = env.reset()
            # print(self.id, i, env)
        self.barrier.put(True)

        while True:
            instruction = self.queue.get()
            if instruction is None:
                break
            for i, (env, action) in enumerate(zip(envs, self.actions)):
                next_obs, reward, terminated, truncated, _ = env.step(action)
                if terminated:
                    self.next_obs[i], _ = env.reset()
                elif truncated:
                    self.next_obs_truncated[i] = next_obs
                    self.next_obs[i], _ = env.reset()
                else:
                    self.next_obs[i] = next_obs

                self.rewards[i] = reward
                self.terminated[i] = terminated
                self.truncated[i] = truncated

            self.barrier.put(True)

缓存不一致问题（Cache Coherency）是多核/多CPU系统中多个处理器缓存之间的数据一致性问题。在Python多进程编程中，由于每个进程可能运行在不同的CPU核心上，当它们访问同一块共享内存时，可能会出现缓存不一致的情况。以下是解决缓存不一致问题的几种方法：

解决缓存不一致问题的深度解析
缓存不一致问题是多核系统中数据同步的核心挑战，特别是在多进程共享内存的场景下。要彻底解决这个问题，需要从硬件到软件的多层协同设计。

缓存不一致的本质原因