zookeeper选举机制

全新集群选举

zookeeper 全新集群选举机制网上资料很多说法很模糊，仔细思考了一下，应该是这样

得到票数最多的机器>机器总数半数

具体启动过程中的哪个节点成为 leader 与 zoo.cfg 中配置的节点数有关，下面以3个举例

选举过程如下

1. server1 启动，给自己投票，然后群发投票信息，由于其他 server 没有启动，所有收不到任何反馈信息，状态处于 LOOKING
2. server2 启动，给自己投票，由于myid文件内容比server1的大，自己得票两票，机器总数半数是1.5，符合上面的规则所以胜出，server2 成为 leader，server1 成为 follower
3. server3 启动，给自己投票，由于myid文件内容比server2的大，自己得票三票，机器总数半数是1.5，符合上面的规则，但是 leader 已经有了，所以只能成为 follower（有点委屈，有点类似于国内结婚每个男人只有一个老婆，哈哈哈）

机器总数对应在 zoo.cfg 中配置如下形式配置节点信息

server.myid文件中的id=节点ip:2888:3888

集群节点数量计算公式

y=2x+1，x为正整数，y为节点数

为什么是奇数？之前写过一篇文章

redis集群最少使用三个主节点和使用16384个槽以及主节点数量不超过1000的原因_zlpzlpzyd的博客-CSDN博客

很多说法是按照5个节点来讲的，按照公式来讲三个应该可以。

下面就来试一下

在自己的机器上安装 zookeeper

三个节点

分别为 master、node1、node2

zoo.cfg

sudo tee conf/zoo.cfg <<-'EOF'
tickTime=2000
initLimit=10
syncLimit=5
dataDir=/opt/zookeeper
clientPort=2181
server.1=master:2888:3888
server.2=node1:2888:3888
server.3=node2:2888:3888
EOF

各个节点的 zoo.cfg 文件内容一致。

master

echo 1 > /opt/zookeeper/myid

node1

echo 2 > /opt/zookeeper/myid

node2

echo 3 > /opt/zookeeper/myid

启动 zookeeper

先启动两个节点，看看情况

[root@master apache-zookeeper-3.9.0-bin]# ./bin/zkServer.sh start
ZooKeeper JMX enabled by default
Using config: /opt/apache-zookeeper-3.9.0-bin/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED

[root@node1 apache-zookeeper-3.9.0-bin]# ./bin/zkServer.sh start
ZooKeeper JMX enabled by default
Using config: /opt/apache-zookeeper-3.9.0-bin/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED

节点状态

[root@master apache-zookeeper-3.9.0-bin]# ./bin/zkServer.sh status
ZooKeeper JMX enabled by default
Using config: /opt/apache-zookeeper-3.9.0-bin/bin/../conf/zoo.cfg
Client port found: 2181. Client address: localhost. Client SSL: false.
Mode: follower

[root@node1 apache-zookeeper-3.9.0-bin]# ./bin/zkServer.sh status
ZooKeeper JMX enabled by default
Using config: /opt/apache-zookeeper-3.9.0-bin/bin/../conf/zoo.cfg
Client port found: 2181. Client address: localhost. Client SSL: false.
Mode: leader

配置了三个节点，只启动了两个节点的情况下，正常选举，符合想法。说明3个节点中有两个节点启动就可以完成选举流程。

下面再启动 node2

[root@node2 apache-zookeeper-3.9.0-bin]# ./bin/zkServer.sh start
ZooKeeper JMX enabled by default
Using config: /opt/apache-zookeeper-3.9.0-bin/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
[root@node2 apache-zookeeper-3.9.0-bin]# ./bin/zkServer.sh status
ZooKeeper JMX enabled by default
Using config: /opt/apache-zookeeper-3.9.0-bin/bin/../conf/zoo.cfg
Client port found: 2181. Client address: localhost. Client SSL: false.
Mode: follower

符合上面的推测

突发奇想，两个节点会是什么情况？

个人猜测，跟上面两个节点启动一致。

两个节点

分别为 master、node1

zoo.cfg

sudo tee conf/zoo.cfg <<-'EOF'
tickTime=2000
initLimit=10
syncLimit=5
dataDir=/opt/zookeeper
clientPort=2181
server.1=master:2888:3888
server.2=node1:2888:3888
server.3=node2:2888:3888
EOF

各个节点的 zoo.cfg 文件内容一致。

启动 zookeeper

[root@master apache-zookeeper-3.9.0-bin]# ./bin/zkServer.sh start
ZooKeeper JMX enabled by default
Using config: /opt/apache-zookeeper-3.9.0-bin/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED

[root@node1 apache-zookeeper-3.9.0-bin]# ./bin/zkServer.sh start
ZooKeeper JMX enabled by default
Using config: /opt/apache-zookeeper-3.9.0-bin/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED

节点状态

[root@master apache-zookeeper-3.9.0-bin]# ./bin/zkServer.sh status
ZooKeeper JMX enabled by default
Using config: /opt/apache-zookeeper-3.9.0-bin/bin/../conf/zoo.cfg
Client port found: 2181. Client address: localhost. Client SSL: false.
Mode: follower

[root@node1 apache-zookeeper-3.9.0-bin]# ./bin/zkServer.sh status
ZooKeeper JMX enabled by default
Using config: /opt/apache-zookeeper-3.9.0-bin/bin/../conf/zoo.cfg
Client port found: 2181. Client address: localhost. Client SSL: false.
Mode: leader

最终结果如上面猜测。

为什么最少三个节点？

去看一下官方文档

ZooKeeper: Because Coordinating Distributed Systems is a Zoo

For replicated mode, a minimum of three servers are required, and it is strongly recommended that you have an odd number of servers. If you only have two servers, then you are in a situation where if one of them fails, there are not enough machines to form a majority quorum. Two servers are inherently less stable than a single server, because there are two single points of failure.

对于集群节点，最少的 server 数量是三个， 强烈建议是奇数。如果是两个 server，当其中一个节点有故障，没有足够的机器进行优先级处理。

符合上面的最少三个节点的理论。

非全新集群选举