注：本文涉及到的環(huán)境版本為kafka 0.10.0
在使用kafka 0.10.0 API獲取consumer offset的時(shí)候，遇到卡住的問(wèn)題，類(lèi)似KAFKA-3296-All consumer reads hang indefinately。

分析步驟

• 打開(kāi)debug日志，可以看到下面這樣的日志內(nèi)容

responseBody={error_code=15,coordinator={node_id=-1,host=,port=-1}}

• 運(yùn)行kafka-topics.sh --describe --zookeeper localhost:2181 --topic __consumer_offsets，得到結(jié)果

Topic:__consumer_offsets        PartitionCount:50       ReplicationFactor:3     Configs:segment.bytes=104857600,cleanup.policy=compact,compression.type=producer
        Topic: __consumer_offsets       Partition: 0    Leader: -1    Replicas: 1003,1004,1007        Isr: 
        Topic: __consumer_offsets       Partition: 1    Leader: 1005    Replicas: 1005        Isr: 1005
        Topic: __consumer_offsets       Partition: 2    Leader: 1005    Replicas: 1005        Isr: 1005
        Topic: __consumer_offsets       Partition: 3    Leader: 1001    Replicas: 1001        Isr: 1001,1003,1002
        Topic: __consumer_offsets       Partition: 4    Leader: 1002    Replicas: 1002,1001,1003        Isr: 1001
        Topic: __consumer_offsets       Partition: 5    Leader: -1    Replicas: 1002,1007,1008        Isr: 
        Topic: __consumer_offsets       Partition: 6    Leader: -1    Replicas: 1003,1008,1009        Isr:       

可以看到有很多partition已經(jīng)沒(méi)有l(wèi)eader。

解決方法

1. 停止kafka集群

2. 在zk上刪除/brokers/topics/__consumer_offsets

3. 啟動(dòng)kafka集群

4. 啟動(dòng)應(yīng)用程序

5. 在通過(guò)kafka-topics.sh --describe命令可以看到__consumer_offsets的所有l(wèi)eader已經(jīng)正常