kafka的消费者分区分配策略

kafka有三种分区分配策略

1.RoundRobin

2.Range

3.Sticky

一、RoundRobin

RoundRobin策略很简单~假设我们有三个Topic10个Partition，上图！

假设顺序为A-0,A-1,A-2...C-2

 不难看出轮询策略是将partition当做最小分配单位，将所有topic的partition都看作一个整体。然后为消费者轮询分配partition。当然得到此结果的前提是Consumer Group种的消费者订阅信息是一致的，如果订阅信息不一致，得到的结果也不均匀，下面举个例子：

 如图，Consumer0订阅Topic-A、B，Consumer1订阅Topic-B、C

顺序注意图中的Seq，先分配TopicA

第一轮 : Consumer-0: Topic-A-Partition0

由于Consumer-1没有订阅Topic-A，所以只能找到Topic-B给Consumer-1分配

于是 Consumer-1: Topic-B-Partition0

------------------------------------------------------------------------------------------------------

第二轮: Consumer-0: Topic-A-Partition0,Topic-A-Partition1

             Consumer-1: Topic-B-Partition0,Topic-B-Partition1

------------------------------------------------------------------------------------------------------

第三轮: Consumer-0: Topic-A-Partition0,Topic-A-Partition1，Topic-A-Partition2

             Consumer-1: Topic-B-Partition0,Topic-B-Partition1，Topic-B-Partition2

------------------------------------------------------------------------------------------------------

第四、五、六轮：

Consumer-0: Topic-A-Partition0,Topic-A-Partition1，Topic-A-Partition2

Consumer-1: Topic-B-Partition0,Topic-B-Partition1，Topic-B-Partition2,Topic-C-Partition-0,Topic-C-Partition-1,Topic-C-Partition-2

------------------------------------------------------------------------------------------------------

可以看到