一 Kafka在zookeeper中存儲結構圖
二 分析
2.1 topic注冊信息
/brokers/topics/[topic] :
存儲某個topic的partitions所有分配信息
[zk: localhost:2181(CONNECTED) 1] get /brokers/topics/firstTopic
Schema:
{ "version": "版本編號目前固定為數字1", "partitions": {"partitionId編號": [ 同步副本組brokerId列表],"partitionId編號": [同步副本組brokerId列表], .......}}
Example:
{"version": 1,"partitions": {"2": [2, 1, 0],"1": [1, 0, 2],"0": [0, 2, 1],}}
2.2 partition狀態信息
/brokers/topics/[topic]/partitions/[0...N] 其中[0..N]表示partition索引號
/brokers/topics/[topic]/partitions/[partitionId]/state
Schema:
{"controller_epoch": 表示kafka集群中的中央控制器選舉次數,"leader": 表示該partition選舉leader的brokerId,"version": 版本編號默認為1,
"leader_epoch": 該partition leader選舉次數,"isr": [同步副本組brokerId列表]}
Example:
{"controller_epoch":61,"leader":0,"version":1,"leader_epoch":0,"isr":[0,2,1]}
2.3 Broker注冊信息
/brokers/ids/[0...N]
每個broker的配置文件中都需要指定一個數字類型的id(全局不可重復),此節點為臨時znode(EPHEMERAL)
Schema:
{"jmx_port": jmx端口號,"timestamp": kafka broker初始啟動時的時間戳,"host": 主機名或ip地址,"version": 版本編號默認為1,
"port": kafka broker的服務端端口號,由server.properties中參數port確定}
Example:
{"jmx_port":-1,"host":"192.168.100.21","timestamp":"1533452008040","port":9092,"version":4}
2.4 Controller epoch
/controller_epoch --> int (epoch)
此值為一個數字,kafka集群中第一個broker第一次啟動時為1，以后只要集群中center controller中央控制器所在broker變更或掛掉，就會重新選舉新的center controller，每次center controller變更controller_epoch值就會 + 1;
2.5 Controller注冊信息
/controller -> int (broker id of the controller) 存儲center controller中央控制器所在kafka broker的信息
Schema:
{"version": 版本編號默認為1,"brokerid": kafka集群中broker唯一編號,"timestamp": kafka broker中央控制器變更時的時間戳}
Example:
{"version": 1,"brokerid": 0,"timestamp": "1533452008692"}
2.6 Consumer and Consumer group
a.每個consumer客戶端被創建時,會向zookeeper注冊自己的信息;
b.此作用主要是為了"負載均衡".
c.同一個Consumer Group中的Consumers，Kafka將相應Topic中的每個消息只發送給其中一個Consumer。
d.Consumer Group中的每個Consumer讀取Topic的一個或多個Partitions，並且是唯一的Consumer；
e.一個Consumer group的多個consumer的所有線程依次有序地消費一個topic的所有partitions,如果Consumer group中所有consumer總線程大於partitions數量，則會出現空閑情況;
舉例說明：
kafka集群中創建一個topic為report-log 4 partitions 索引編號為0,1,2,3，假如有目前有三個消費者node：注意-->一個consumer中一個消費線程可以消費一個或多個partition.如果每個consumer創建一個consumer thread線程,各個node消費情況如下，node1消費索引編號為0,1分區，node2費索引編號為2,node3費索引編號為3，如果每個consumer創建2個consumer thread線程，各個node消費情況如下(是從consumer node先后啟動狀態來確定的)，node1消費索引編號為0,1分區；node2費索引編號為2,3；node3為空閑狀態。
總結：
從以上可知，Consumer Group中各個consumer是根據先后啟動的順序有序消費一個topic的所有partitions的。如果Consumer Group中所有consumer的總線程數大於partitions數量，則可能consumer thread或consumer會出現空閑狀態。
2.7 Consumer均衡算法
當一個group中,有consumer加入或者離開時,會觸發partitions均衡.均衡的最終目的,是提升topic的並發消費能力.
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) 首先根據partition索引號對partitions排序: P0,P1,P2,P3
4) 根據(consumer.id + '-'+ thread序號)排序: C0,C1
5) 計算倍數: 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注冊信息
每個consumer都有一個唯一的ID(consumerId可以通過配置文件指定,也可以由系統生成),此id用來標記消費者信息.
/consumers/[groupId]/ids/[consumerIdString]
是一個臨時的znode,此節點的值為請看consumerIdString產生規則,即表示此consumer目前所消費的topic + partitions列表.
consumerId產生規則：
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": 版本編號默認為1,"subscription": { //訂閱topic列表"topic名稱": consumer中topic消費者線程數},"pattern": "static",
"timestamp": "consumer啟動時的時間戳"}
Example:
{"version": 1,"subscription": {"topic2": 1},"pattern": "white_list","timestamp": "1525747915336"}
2.9 Consumer owner
/consumers/[groupId]/owners/[topic]/[partitionId] -> consumerIdString + threadId索引編號
a) 首先進行"Consumer Id注冊";
b) 然后在"Consumer id 注冊"節點下注冊一個watch用來監聽當前group中其他consumer的"退出"和"加入";只要此znode path下節點列表變更,都會觸發此group下consumer的負載均衡.(比如一個consumer失效,那么其他consumer接管partitions).
c) 在"Broker id 注冊"節點下,注冊一個watch用來監聽broker的存活情況;如果broker列表變更,將會觸發所有的groups下的consumer重新balance.
2.10 Consumer offset
/consumers/[groupId]/offsets/[topic]/[partitionId] -> long (offset)
用來跟蹤每個consumer目前所消費的partition中最大的offset
此znode為持久節點,可以看出offset跟group_id有關,以表明當消費者組(consumer group)中一個消費者失效,
重新觸發balance,其他consumer可以繼續消費.
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 刪除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]
