1004 broker 挂掉后,
执行以下命令可以查看,不可用分区,如图:
# /usr/hdf/current/kafka-broker/bin/kafka-topics.sh --describe --zookeeper 10.32.40.39:2181 | grep 'Isr: 1004' | grep 'Leader: 1004'
解决办法:
强行修改zk元数据isr leader,并重新选举。如再不行,滚动重启kafka broker
进入zk:
# /usr/hdf/current/kafka-broker/bin/zookeeper-shell.sh 10.32.x.x:2181
获取分区zk元数据:以action主题0分区为例:
get /brokers/topics/action/partitions/0/state
重新设置isr为可用节点,leader为可用节点,leader_epoch+1
# set /brokers/topics/action/partitions/0/state {"controller_epoch":57,"leader":1001,"version":1,"leader_epoch":98,"isr":[1003,1001]}
重新选举:
编写preferred-leader-plan.json文件,内容如下:
{"partitions":[{"topic":"action","partition": 0}]
执行以下命令:
/usr/hdf/current/kafka-broker/bin/kafka-preferred-replica-election.sh --zookeeper 10.32.x.x:2181 --path-to-json-file preferred-leader-plan.json
检查:
/usr/hdf/current/kafka-broker/bin/kafka-run-class.sh kafka.tools.GetOffsetShell --broker-list 10.32.x.x:9092 --topic action --time -1
并查看日志:/var/log/kafka/server.log
如未出现:Error: partition 0 does not have a leader. Skip getting offsets 则 成功,若出现,重启机器。
正常如下:
# /usr/hdf/current/kafka-broker/bin/kafka-run-class.sh kafka.tools.GetOffsetShell --broker-list 10.32.x.x:9092 --topic action --time -1
action:2:2964861
action:1:2964843
action:3:2948928
action:0:2964749
每个分区offset 正常,如有消费,数值会有增加。