分布式系统入门：CAP 理论与一致性算法详解

CAP 理论核心概念

CAP 理论指出分布式系统最多只能同时满足一致性（Consistency）、可用性（Availability）、分区容错性（Partition Tolerance）中的两项。

• 一致性 (C)：所有节点访问同一份最新数据。
• 可用性 (A)：非故障节点在合理时间内返回响应。
• 分区容错性 (P)：网络分区时系统仍能运作。

一致性算法实现（Paxos 简化版）

以下代码模拟 Paxos 算法的提案阶段，使用 Python 实现基础逻辑：

class PaxosNode: def __init__(self, node_id): self.node_id = node_id self.proposed_value = None self.accepted_value = None self.promised_proposal_id = 0 def prepare(self, proposal_id): if proposal_id > self.promised_proposal_id: self.promised_proposal_id = proposal_id return {"status": "promised", "accepted_value": self.accepted_value} return {"status": "rejected"} def accept(self, proposal_id, value): if proposal_id >= self.promised_proposal_id: self.accepted_value = value self.promised_proposal_id = proposal_id return {"status": "accepted"} return {"status": "rejected"} # 模拟三个节点的集群 nodes = [PaxosNode(i) for i in range(3)] # 提案阶段 proposal_id = 10 value = "data_v1" for node in nodes: response = node.prepare(proposal_id) print(f"Node {node.node_id} prepare response: {response}") # 接受阶段（仅半数以上节点响应时） if sum(1 for node in nodes if node.prepare(proposal_id)["status"] == "promised") > len(nodes)/2: for node in nodes: response = node.accept(proposal_id, value) print(f"Node {node.node_id} accept response: {response}")

https://www.zhihu.com/zvideo/1993908941005035175/
https://www.zhihu.com/zvideo/1993908941005035175
https://www.zhihu.com/zvideo/1993908940069699915/
https://www.zhihu.com/zvideo/1993908940069699915
https://www.zhihu.com/zvideo/1993908934109573278/
https://www.zhihu.com/zvideo/1993908934109573278
https://www.zhihu.com/zvideo/1993908933191037953/
https://www.zhihu.com/zvideo/1993908933191037953
https://www.zhihu.com/zvideo/1993908931861446751/
https://www.zhihu.com/zvideo/1993908931861446751
https://www.zhihu.com/zvideo/1993908925066650081/
https://www.zhihu.com/zvideo/1993908925066650081
https://www.zhihu.com/zvideo/1993908923359580411/
https://www.zhihu.com/zvideo/1993908923359580411
https://www.zhihu.com/zvideo/1993908922080313621/
https://www.zhihu.com/zvideo/1993908922080313621
https://www.zhihu.com/zvideo/1993908907127641011/
https://www.zhihu.com/zvideo/1993908907127641011
https://www.zhihu.com/zvideo/1993908900202840925/
https://www.zhihu.com/zvideo/1993908900202840925
https://www.zhihu.com/zvideo/1993908885128508038/
https://www.zhihu.com/zvideo/1993908885128508038

代码功能说明

1. PaxosNode 类：模拟单个节点的提案（prepare）和接受（accept）逻辑。
2. prepare 方法：节点承诺不接受比当前提案 ID 更小的请求。
3. accept 方法：节点在提案 ID 有效时接受值并更新状态。
4. 集群模拟：通过半数以上节点的承诺达成一致性。

输出示例

Node 0 prepare response: {'status': 'promised', 'accepted_value': None} Node 1 prepare response: {'status': 'promised', 'accepted_value': None} Node 2 prepare response: {'status': 'promised', 'accepted_value': None} Node 0 accept response: {'status': 'accepted'} Node 1 accept response: {'status': 'accepted'} Node 2 accept response: {'status': 'accepted'}

关键注意事项

• 实际分布式系统需处理网络延迟、节点故障等复杂场景。
• Paxos 变种（如 Raft）通过领导者选举简化实现。
• CAP 权衡：选择 CP（如 ZooKeeper）或 AP（如 Cassandra）取决于业务需求。