一 Kafka在zookeeper中存储结构图
二 分析
2.1 topic注册信息
/brokers/topics/[topic] :
存储某个topic的partitions所有分配信息
[zk: localhost:2181(CONNECTED) 1] get /brokers/topics/firstTopic
Schema:
{ "version": "版本编号目前固定为数字1", "partitions": {"partitionId编号": [ 同步副本组brokerId列表],"partitionId编号": [同步副本组brokerId列表], .......}}
Example:
{"version": 1,"partitions": {"2": [2, 1, 0],"1": [1, 0, 2],"0": [0, 2, 1],}}
2.2 partition状态信息
/brokers/topics/[topic]/partitions/[0...N] 其中[0..N]表示partition索引号
/brokers/topics/[topic]/partitions/[partitionId]/state
Schema:
{"controller_epoch": 表示kafka集群中的中央控制器选举次数,"leader": 表示该partition选举leader的brokerId,"version": 版本编号默认为1,
"leader_epoch": 该partition leader选举次数,"isr": [同步副本组brokerId列表]}
Example:
{"controller_epoch":61,"leader":0,"version":1,"leader_epoch":0,"isr":[0,2,1]}
2.3 Broker注册信息
/brokers/ids/[0...N]
每个broker的配置文件中都需要指定一个数字类型的id(全局不可重复),此节点为临时znode(EPHEMERAL)
Schema:
{"jmx_port": jmx端口号,"timestamp": kafka broker初始启动时的时间戳,"host": 主机名或ip地址,"version": 版本编号默认为1,
"port": kafka broker的服务端端口号,由server.properties中参数port确定}
Example:
{"jmx_port":-1,"host":"192.168.100.21","timestamp":"1533452008040","port":9092,"version":4}
2.4 Controller epoch
/controller_epoch --> int (epoch)
此值为一个数字,kafka集群中第一个broker第一次启动时为1，以后只要集群中center controller中央控制器所在broker变更或挂掉，就会重新选举新的center controller，每次center controller变更controller_epoch值就会 + 1;
2.5 Controller注册信息
/controller -> int (broker id of the controller) 存储center controller中央控制器所在kafka broker的信息
Schema:
{"version": 版本编号默认为1,"brokerid": kafka集群中broker唯一编号,"timestamp": kafka broker中央控制器变更时的时间戳}
Example:
{"version": 1,"brokerid": 0,"timestamp": "1533452008692"}
2.6 Consumer and Consumer group
a.每个consumer客户端被创建时,会向zookeeper注册自己的信息;
b.此作用主要是为了"负载均衡".
c.同一个Consumer Group中的Consumers，Kafka将相应Topic中的每个消息只发送给其中一个Consumer。
d.Consumer Group中的每个Consumer读取Topic的一个或多个Partitions，并且是唯一的Consumer；
e.一个Consumer group的多个consumer的所有线程依次有序地消费一个topic的所有partitions,如果Consumer group中所有consumer总线程大于partitions数量，则会出现空闲情况;
举例说明：
kafka集群中创建一个topic为report-log 4 partitions 索引编号为0,1,2,3，假如有目前有三个消费者node：注意-->一个consumer中一个消费线程可以消费一个或多个partition.如果每个consumer创建一个consumer thread线程,各个node消费情况如下，node1消费索引编号为0,1分区，node2费索引编号为2,node3费索引编号为3，如果每个consumer创建2个consumer thread线程，各个node消费情况如下(是从consumer node先后启动状态来确定的)，node1消费索引编号为0,1分区；node2费索引编号为2,3；node3为空闲状态。
总结：
从以上可知，Consumer Group中各个consumer是根据先后启动的顺序有序消费一个topic的所有partitions的。如果Consumer Group中所有consumer的总线程数大于partitions数量，则可能consumer thread或consumer会出现空闲状态。
2.7 Consumer均衡算法
当一个group中,有consumer加入或者离开时,会触发partitions均衡.均衡的最终目的,是提升topic的并发消费能力.
1) 假如topic1,具有如下partitions: P0,P1,P2,P3
2) 加入group中,有如下consumer: C0,C1
3) 首先根据partition索引号对partitions排序: P0,P1,P2,P3
4) 根据(consumer.id + '-'+ thread序号)排序: C0,C1
5) 计算倍数: M = [P0,P1,P2,P3].size / [C0,C1].size,本例值M=2(向上取整)
6) 然后依次分配partitions: C0 = [P0,P1],C1=[P2,P3],即Ci = [P(i * M),P((i + 1) * M -1)]
2.8 Consumer注册信息
每个consumer都有一个唯一的ID(consumerId可以通过配置文件指定,也可以由系统生成),此id用来标记消费者信息.
/consumers/[groupId]/ids/[consumerIdString]
是一个临时的znode,此节点的值为请看consumerIdString产生规则,即表示此consumer目前所消费的topic + partitions列表.
consumerId产生规则：
StringconsumerUuid = null;
if(config.consumerId!=null && config.consumerId)
consumerUuid = consumerId;
else {
String uuid = UUID.randomUUID()
consumerUuid = "%s-%d-%s".format(
InetAddress.getLocalHost.getHostName, System.currentTimeMillis,
uuid.getMostSignificantBits().toHexString.substring(0,8));
}
String consumerIdString = config.groupId + "_" + consumerUuid;
[zk: localhost:2181(CONNECTED) 11] get /consumers/console-consumer-2304/ids/console-consumer-2304_hadoop2-1525747915241-6b48ff32
Schema:
{"version": 版本编号默认为1,"subscription": { //订阅topic列表"topic名称": consumer中topic消费者线程数},"pattern": "static",
"timestamp": "consumer启动时的时间戳"}
Example:
{"version": 1,"subscription": {"topic2": 1},"pattern": "white_list","timestamp": "1525747915336"}
2.9 Consumer owner
/consumers/[groupId]/owners/[topic]/[partitionId] -> consumerIdString + threadId索引编号
a) 首先进行"Consumer Id注册";
b) 然后在"Consumer id 注册"节点下注册一个watch用来监听当前group中其他consumer的"退出"和"加入";只要此znode path下节点列表变更,都会触发此group下consumer的负载均衡.(比如一个consumer失效,那么其他consumer接管partitions).
c) 在"Broker id 注册"节点下,注册一个watch用来监听broker的存活情况;如果broker列表变更,将会触发所有的groups下的consumer重新balance.
2.10 Consumer offset
/consumers/[groupId]/offsets/[topic]/[partitionId] -> long (offset)
用来跟踪每个consumer目前所消费的partition中最大的offset
此znode为持久节点,可以看出offset跟group_id有关,以表明当消费者组(consumer group)中一个消费者失效,
重新触发balance,其他consumer可以继续消费.
2.11 Re-assign partitions
/admin/reassign_partitions
{
"fields":[
{ "name":"version","type":"int","doc":"version id" },
{ "name":"partitions",
"type":{"type":"array", "items":{
"fields":[{"name":"topic","type":"string","doc":"topic of the partition to be reassigned"},
{"name":"partition", "type":"int","doc":"the partition to be reassigned"},
{"name":"replicas","type":"array","items":"int","doc":"a list of replica ids"}],
}
"doc":"an array of partitions to be reassigned to new replicas"
}
}
]
}
Example:
{"version": 1,"partitions": [{"topic": "Foo","partition": 1, "replicas": [0, 1, 3] }] }
2.12 Preferred replication election
/admin/preferred_replica_election
{
"fields":[
{
"name":"version",
"type":"int",
"doc":"version id"
},
{
"name":"partitions",
"type":{
"type":"array",
"items":{
"fields":[
{
"name":"topic",
"type":"string",
"doc":"topic of the partition for which preferred replica election should be triggered"
},
{
"name":"partition",
"type":"int",
"doc":"the partition for which preferred replica election should be triggered"
}
],
}
"doc":"an array of partitions for which preferred replica election should be triggered"
}
}
]
}
例子:
{
"version": 1,
"partitions":
[
{
"topic": "Foo",
"partition": 1
},
{
"topic": "Bar",
"partition": 0
}
]
}
2.13 删除topics
/admin/delete_topics
Schema:
{ "fields":
[ {"name": "version", "type": "int", "doc": "version id"},
{"name": "topics",
"type": { "type": "array", "items": "string", "doc": "an array of topics to be deleted"}
} ]
}
例子:
{ "version": 1,"topics": ["foo", "bar"]}
2.14 Topic配置
/config/topics/[topic_name]
