Kafka - 10 Kafka副本 | 分区副本分配 | 手动调整分区副本 | Leader Partition 负载平衡 | 增加副本因子

1. 分区副本分配

如果 kafka 服务器只有 4 个节点，那么设置 kafka 的分区数大于服务器台数，在 kafka底层如何分配存储副本呢？

① 创建 16 分区，3 个副本

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-topics.sh --bootstrap-server hadoop101:9092 --create --partitions 16 --replication-factor 3 --topic test4
1

② 查看分区和副本情况

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-topics.sh --bootstrap-server hadoop101:9092 --describe --topic test4
1

这样是为了leader和follower均匀的分配在每个服务器上，尽量不让每个分区的leader和follower都分配在某几个服务器上，比如只分配在broker0、broker1、broker2上，那么这三个服务器挂掉后，数据就丢失了，这样分配的好处是除非四个服务器都挂掉，不然就还可以从broker3同步数据。

2. 手动调整分区副本

在生产环境中，每台服务器的配置和性能不一致，但是Kafka只会根据自己的代码规则创建对应的分区副本，就会导致个别服务器存储压力较大。所有需要手动调整分区副本的存储。

创建一个新的topic，4个分区，两个副本，将该topic的所有副本都存储到broker0和broker1两台服务器上。

手动调整分区副本存储的步骤如下：

① 创建一个新的topic，4个分区，两个副本

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-topics.sh --bootstrap-server hadoop101:9092 --create --partitions 4 --replication-factor 2 --topic test5
1

② 查看分区副本存储情况

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-topics.sh --bootstrap-server hadoop101:9092 --describe --topic test5
1

③ 创建副本存储计划（所有副本都指定存储在 broker0、broker1 中）

[root@hadoop101 kafka_2.12-2.2.1]# vi increase-replication-factor.json
{
    "version":1,
    "partitions":[{"topic":"test5","partition":0,"replicas":[0,1]},
                  {"topic":"test5","partition":1,"replicas":[0,1]},
                  {"topic":"test5","partition":2,"replicas":[1,0]},
                  {"topic":"test5","partition":3,"replicas":[1,0]}]
}
1
2
3
4
5
6
7
8

④ 执行副本存储计划

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-reassign-partitions.sh --bootstrap-server hadoop102:9092 --zookeeper hadoop101:2181,hadoop102:2181,hadoop103:2181/kafka_cluster --reassignment-json-file increase-replication-factor.json --execute
Current partition replica assignment

{"version":1,"partitions":[{"topic":"test5","partition":2,"replicas":[3,1],"log_dirs":["any","any"]},{"topic":"test5","partition":1,"replicas":[2,3],"log_dirs":["any","any"]},{"topic":"test5","partition":3,"replicas":[1,0],"log_dirs":["any","any"]},{"topic":"test5","partition":0,"replicas":[0,2],"log_dirs":["any","any"]}]}

Save this to use as the --reassignment-json-file option during rollback
Successfully started reassignment of partitions.
1
2
3
4
5
6
7

⑤ 验证副本存储计划

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-reassign-partitions.sh --bootstrap-server hadoop102:9092 --zookeeper hadoop101:2181,hadoop102:2181,hadoop103:2181/kafka_cluster --reassignment-json-file increase-replication-factor.json --verify
Status of partition reassignment:
Reassignment of partition test5-0 completed successfully
Reassignment of partition test5-1 completed successfully
Reassignment of partition test5-2 completed successfully
Reassignment of partition test5-3 completed successfully
1
2
3
4
5
6

⑥ 查看分区副本存储情况

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-topics.sh --bootstrap-server hadoop101:9092 --describe --topic test5
1

3. Leader Partition 负载平衡

正常情况下，Kafka本身会自动把Leader Partition均匀分散在各个机器上，来保证每台机器的读写吞吐量都是均匀的。但是如果某些broker宕机，会导致Leader Partition过于集中在其他少部分几台broker上，这会导致少数几台broker的读写请求压力过高，其他宕机的broker重启之后都是follower partition，读写请求很低，造成集群负载不均衡。

① auto.leader.rebalance.enable，默认是true。自动Leader Partition 平衡

② eader.imbalance.per.broker.percentage，默认是10%。每个broker允许的不平衡的leader的比率。如果每个broker超过了这个值，控制器会触发leader的平衡。

③ leader.imbalance.check.interval.seconds，默认值300秒。检查leader负载是否平衡的间隔时间。

下面拿一个主题举例说明，假设集群只有一个主题如下图所示：

针对broker0节点，分区2的AR优先副本是0节点，但是0节点却不是Leader节点，所以不平衡数加1，AR副本总数是4，所以broker0节点不平衡率为1/4>10%，需要再平衡。

broker2和broker3节点和broker0不平衡率一样，需要再平衡，broker1的不平衡数为0，不需要再平衡。

4. 增加副本因子

在生产环境当中，由于某个主题的重要等级需要提升，我们考虑增加副本。副本数的增加需要先制定计划，然后根据计划执行。

① 创建 topic

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-topics.sh --bootstrap-server hadoop101:9092 --create --partitions 3 --replication-factor 1 --topic test6
1

② 创建副本存储计划（所有副本都指定存储在 broker0、broker1、broker2 中）

[root@hadoop101 kafka_2.12-2.2.1]# vi increase-replication-factor.json
{
    "version":1,
    "partitions":[{"topic":"test6","partition":0,"replicas":[0,1,2]},
                  {"topic":"test6","partition":1,"replicas":[0,1,2]},
                  {"topic":"test6","partition":2,"replicas":[0,1,2]}]
}
1
2
3
4
5
6
7

③ 执行副本存储计划

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-reassign-partitions.sh --bootstrap-server hadoop102:9092 --zookeeper hadoop101:2181,hadoop102:2181,hadoop103:2181/kafka_cluster --reassignment-json-file increase-replication-factor.json --execute
Current partition replica assignment

{"version":1,"partitions":[{"topic":"test6","partition":2,"replicas":[0],"log_dirs":["any"]},{"topic":"test6","partition":1,"replicas":[1],"log_dirs":["any"]},{"topic":"test6","partition":0,"replicas":[3],"log_dirs":["any"]}]}

Save this to use as the --reassignment-json-file option during rollback
Successfully started reassignment of partitions.
1
2
3
4
5
6
7

⑤ 验证副本存储计划

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-reassign-partitions.sh --bootstrap-server hadoop102:9092 --zookeeper hadoop101:2181,hadoop102:2181,hadoop103:2181/kafka_cluster --reassignment-json-file increase-replication-factor.json --verify
Status of partition reassignment:
Reassignment of partition test6-0 completed successfully
Reassignment of partition test6-1 completed successfully
Reassignment of partition test6-2 completed successfully
1
2
3
4
5

⑥ 查看分区副本存储情况

[root@hadoop101 kafka_2.12-2.2.1]# bin/kafka-topics.sh --bootstrap-server hadoop101:9092 --describe --topic test6
Topic:test6    PartitionCount:3    ReplicationFactor:3   Configs:segment.bytes=1073741824
Topic: test6    Partition: 0    Leader: 0       Replicas: 0,1,2 Isr: 0,2,1
Topic: test6    Partition: 1    Leader: 1       Replicas: 0,1,2 Isr: 1,2,0
Topic: test6    Partition: 2    Leader: 0       Replicas: 0,1,2 Isr: 0,2,1
1
2
3
4
5