一、kafka-preferred-replica-election使用背景

在建立一個topic時，kafka儘量將partition均分在所有的brokers上，並且將replicas也均分在不同的broker上。

每個partitiion的所有replicas叫做"assigned replicas"，"assigned replicas"中的第一個replicas叫"preferred replica"，剛建立的topic一般"preferred replica"是leader。leader replica負責所有的讀寫。

但隨著時間推移，broker可能會停機，會導致leader遷移，導致機群的負載不均衡。我們期望對topic的leader進行重新負載均衡，讓partition選擇"preferred replica"做為leader。
 

kafka-preferred-replica-election命令便是用於對Leader進行重新負載均衡

二、kafka-preferred-replica-election用法

用法1： 觸發對所有的topic Leader進行負載均衡

kafka-preferred-replica-election --zookeeper cdh-002/kafka

用法2： 對某個topic Leader觸發負載均衡

kafka-preferred-replica-election --zookeeper cdh-002/kafka --path-to-json-file xx.json

其中xx.json格式見如下解釋：
[
 
[email protected] ~]# kafka-preferred-replica-election
This tool causes leadership for each partition to be transferred back to the 'preferred replica', it can be used to balance leadership among the servers
Option                                 Description                           
------                                 -----------                           
--path-to-json-file <String: list of   The JSON file with the list of        
  partitions for which preferred         partitions for which preferred      
  replica leader election needs to be    replica leader election should be   
  triggered>                             done, in the following format -     
                                       {"partitions":                        
                                        [{"topic": "foo", "partition": 1},   
                                         {"topic": "foobar", "partition": 2}]
                                       }      
 

三、生產環境常開啟topic Leader的自動負載均衡

可通過配置auto.leader.rebalance.enable=true實現topic Leader的自動負載均衡