如何使用kafka增加topic的備份數量，讓業務更上一層樓

本文由雲+社區發表

一、困難點

建立topic的時候，可以通過指定參數 --replication-factor 設置備份數量。但是，一旦完成建立topic，則無法通過kafka-topic.sh 或者 命令修改replica數量。

二、解決辦法

​ 實際上，我們可以考慮一種 “另類” 的辦法：可以利用 kafka-reassign-partitions.sh 命令對所有分區進行重新分布，在做分區重新分布的時候，通過增加每個分區的replica備份數量來達到目的。

​ 本文將介紹如何利用 kafka-reassign-partitions.sh 命令增加topic的備份數量。

注意：以下命令使用到的topic名稱、zookeeper的ip和port，需要讀者替換成為實際集群的參數。

（假設kafka集群有4個broker，id分別為：1001,1002,1003,1004）

2.1、獲取當前topic的所有分區分布在broker的情況

[root@tbds bin]# ./kafka-topics.sh --zookeeper 172.16.32.13:2181 --topic ranger_audits --describe
Topic:ranger_audits     PartitionCount:10       ReplicationFactor:1     Configs:
        Topic: ranger_audits    Partition: 0    Leader: 1001    Replicas: 1001  Isr: 1001
        Topic: ranger_audits    Partition: 1    Leader: 1002    Replicas: 1002  Isr: 1002
        Topic: ranger_audits    Partition: 2    Leader: 1001    Replicas: 1001  Isr: 1001
        Topic: ranger_audits    Partition: 3    Leader: 1002    Replicas: 1002  Isr: 1002
        Topic: ranger_audits    Partition: 4    Leader: 1001    Replicas: 1001  Isr: 1001
        Topic: ranger_audits    Partition: 5    Leader: 1002    Replicas: 1002  Isr: 1002
        Topic: ranger_audits    Partition: 6    Leader: 1001    Replicas: 1001  Isr: 1001
        Topic: ranger_audits    Partition: 7    Leader: 1002    Replicas: 1002  Isr: 1002
        Topic: ranger_audits    Partition: 8    Leader: 1001    Replicas: 1001  Isr: 1001
        Topic: ranger_audits    Partition: 9    Leader: 1002    Replicas: 1002  Isr: 1002

可以看出，ranger_audits 這個topic有10個分區，每個分區只有一個feplica備份，分布在1001和1002兩台broker上面。

下面我們需要將ranger_audits 的每個分區數據都增加到2個replica備份，且分布到4個broker上面。

2.2、創建增加replica備份數量的配置文件

（注意：盡量保持topic的原有每個分區的主備份不變化。因此，配置文件的每個分區的第一個broker保持不變。）

[root@tbds bin]# vim ../config/increase-replication-factor.json
{"version":1,
"partitions":[
{"topic":"ranger_audits","partition":0,"replicas":[1001,1003]},
{"topic":"ranger_audits","partition":1,"replicas":[1002,1004]},
{"topic":"ranger_audits","partition":2,"replicas":[1001,1003]},
{"topic":"ranger_audits","partition":3,"replicas":[1002,1004]},
{"topic":"ranger_audits","partition":4,"replicas":[1001,1003]},
{"topic":"ranger_audits","partition":5,"replicas":[1002,1004]},
{"topic":"ranger_audits","partition":6,"replicas":[1001,1003]},
{"topic":"ranger_audits","partition":7,"replicas":[1002,1004]},
{"topic":"ranger_audits","partition":8,"replicas":[1001,1003]},
{"topic":"ranger_audits","partition":9,"replicas":[1002,1004]}
]}

上面的配置文件說明，我們將topic的每個分區都增加了一個replica，且保持每個分區原有的主備份所在broker不變化，將每個分區新增的replica備份數據放到到1003和1004兩個broker上面。

2.3、開始執行增加分區

[root@tbds bin]# ./kafka-reassign-partitions.sh -zookeeper 172.16.32.13:2181 --reassignment-json-file ../config/increase-replication-factor.json --execute
Current partition replica assignment
{"version":1,"partitions":[{"topic":"ranger_audits","partition":3,"replicas":[1002]},{"topic":"ranger_audits","partition":9,"replicas":[1002]},{"topic":"ranger_audits","partition":8,"replicas":[1001]},{"topic":"ranger_audits","partition":1,"replicas":[1002]},{"topic":"ranger_audits","partition":4,"replicas":[1001]},{"topic":"ranger_audits","partition":2,"replicas":[1001]},{"topic":"ranger_audits","partition":5,"replicas":[1002]},{"topic":"ranger_audits","partition":0,"replicas":[1001]},{"topic":"ranger_audits","partition":6,"replicas":[1001]},{"topic":"ranger_audits","partition":7,"replicas":[1002]}]}

Save this to use as the --reassignment-json-file option during rollback
Successfully started reassignment of partitions 
{"version":1,"partitions":[{"topic":"ranger_audits","partition":0,"replicas":[1001,1003]},{"topic":"ranger_audits","partition":8,"replicas":[1001,1003]},{"topic":"ranger_audits","partition":5,"replicas":[1002,1004]},{"topic":"ranger_audits","partition":2,"replicas":[1001,1003]},{"topic":"ranger_audits","partition":9,"replicas":[1002,1004]},{"topic":"ranger_audits","partition":1,"replicas":[1002,1004]},{"topic":"ranger_audits","partition":3,"replicas":[1002,1004]},{"topic":"ranger_audits","partition":4,"replicas":[1001,1003]},{"topic":"ranger_audits","partition":7,"replicas":[1002,1004]},{"topic":"ranger_audits","partition":6,"replicas":[1001,1003]}]}

2.4、查看執行進度

[root@tbds bin]# ./kafka-reassign-partitions.sh -zookeeper 172.16.32.13:2181 --reassignment-json-file ../config/increase-replication-factor.json --verify
Status of partition reassignment:
Reassignment of partition [ranger_audits,0] completed successfully
Reassignment of partition [ranger_audits,8] completed successfully
Reassignment of partition [ranger_audits,5] completed successfully
Reassignment of partition [ranger_audits,2] completed successfully
Reassignment of partition [ranger_audits,9] completed successfully
Reassignment of partition [ranger_audits,1] completed successfully
Reassignment of partition [ranger_audits,3] completed successfully
Reassignment of partition [ranger_audits,4] completed successfully
Reassignment of partition [ranger_audits,7] completed successfully
Reassignment of partition [ranger_audits,6] completed successfully

上面顯示增加分區操作成功

2.5、再次查看topic的情況

[root@tbds bin]# ./kafka-topics.sh --zookeeper 172.16.32.13:2181 --topic ranger_audits --describe
Topic:ranger_audits     PartitionCount:10       ReplicationFactor:2     Configs:
        Topic: ranger_audits    Partition: 0    Leader: 1001    Replicas: 1001,1003     Isr: 1001,1003
        Topic: ranger_audits    Partition: 1    Leader: 1002    Replicas: 1002,1004     Isr: 1002,1004
        Topic: ranger_audits    Partition: 2    Leader: 1001    Replicas: 1001,1003     Isr: 1001,1003
        Topic: ranger_audits    Partition: 3    Leader: 1002    Replicas: 1002,1004     Isr: 1002,1004
        Topic: ranger_audits    Partition: 4    Leader: 1001    Replicas: 1001,1003     Isr: 1001,1003
        Topic: ranger_audits    Partition: 5    Leader: 1002    Replicas: 1002,1004     Isr: 1002,1004
        Topic: ranger_audits    Partition: 6    Leader: 1001    Replicas: 1001,1003     Isr: 1001,1003
        Topic: ranger_audits    Partition: 7    Leader: 1002    Replicas: 1002,1004     Isr: 1002,1004
        Topic: ranger_audits    Partition: 8    Leader: 1001    Replicas: 1001,1003     Isr: 1001,1003
        Topic: ranger_audits    Partition: 9    Leader: 1002    Replicas: 1002,1004     Isr: 1002,1004

從上面可以看出，備份數量增加成功

三、進一步思考

​ 利用上述介紹的辦法，除了可以用來增加topic的備份數量之外，還能夠處理以下幾個場景：

1、對topic的所有分區數據進行整體遷移。怎么理解呢？假如集群有N個broker，后來新擴容M個broker。由於新擴容的broker磁盤都是空的，原有的broker磁盤占用都很滿。那么我們可以利用上述方法，將存儲在原有N個broker的某些topic整體搬遷到新擴容的M個broker，進而實現kafka集群的整體數據均衡。

​ 具體使用方法就是：通過編寫2.2章節的配置文件，將topic的所有分區都配置到新的M個broker上面去，再執行excute，即可完成topic的所有分區數據整體遷移到新擴容的M個broker節點。

*2、broker壞掉的情況。*導致某些topic的某些分區的replica數量減少，可以利用kafka-reassign-partitions.sh增加replica；

*3、kafka 某些broker磁盤占用很滿，某些磁盤占用又很少。*可以利用kafka-reassign-partitions.sh遷移某些topic的分區數據到磁盤占用少的broker，實現數據均衡；

*4、kafka集群擴容。*需要把原來broker的topic數據整體遷移到新的broker，合理利用新擴容的broker，實現負載均衡。

此文已由作者授權騰訊雲+社區在各渠道發布

獲取更多新鮮技術干貨，可以關注我們騰訊雲技術社區-雲加社區官方號及知乎機構號