《分布式》：paxos算法和一致性Hash算法相关记录

《分布式》：paxos算法和一致性Hash算法相关记录

今天无意之间看到了这个关于分布式的协调一致的算法：paxos，于是找了一些相关博文看了下，说没有弄懂吧，感觉还懂点，说懂了吧，但是又感觉差点。哎，还需要多看看相关博文呀。

莱斯利·兰伯特（Leslie Lamport，paxos算法的奠基人，）的Paxos Made Simple 是学习paxos最好的文档，文档连接在这里：http://research.microsoft.com/en-us/um/people/lamport/pubs/paxos-simple.pdf。

关于讲解paxos算法的博文很多，如下：

1、http://iunknown.iteye.com/blog/2246484

2、http://www.cnblogs.com/endsock/p/3480093.html

3、http://blog.csdn.net/cutesource/article/details/5813379

一个常见的paxos疑问

源地址：http://stackoverflow.com/questions/14435646/paxos-value-choice/14472334#14472334

问题：

Now, let’s take a hypothetical example that people often give, and which they think breaks Paxos. Suppose we have three Acceptors A1, A2, and A3. A1 and A2 have both accepted value ABC at round 1 and A3 has chosen XYZ at round 2 (ie. from a different proposer). We can see that A1 and A2 form a majority and that ABC has been “chosen.”

Continuing along this hypothetical example, a proposer sends Prepare(3) and receives back responses from A2 and A3, viz Promise(ABC @ 1) and Promise(XYZ @ 2). The Proposer sees XYZ has the highest round, and sends that along in the Accept phase, overwriting ABC on the other hosts. And viola, Paxos is broken, Right?

解答：

No. The problem is with the start state, which is impossible. Let me show you why.

First, some propositions, which are key to Paxos running correctly:

Proposition A: For A1 and A2 to have the value ABC @ 1, a proposer must have sent Accept(ABC @ 1) which means it must have received a majority of Promises in response to sending Prepare(1).

Proposition B: For A3 to have the value XYZ @ 2, a proposer must have sent Accept(XYZ @ 2) which means it must have received a majority of Promises in response to sending Prepare(2).

一致性Hash算法

一致性Hash算法：http://blog.csdn.net/sparkliang/article/details/5279393