kafka分区leader切换?报错

kafka+storm构架
kafka 版本kafka_2.10-0.8.2.1，topic 副本数为2。zk版本3.4.6，storm版本apache-storm-0.9.4

近期storm消费总发现下面这样的报错：storm.kafka.FailedFetchException: Error fetching data from [Partition{host=xx.xx.xx.xx:9092, partition=11}] for topic [test]: [NOT_LEADER_FOR_PARTITION] at storm.kafka.KafkaUtils.fetchMessages
通过kafka-topics.sh --zookeeper localhost --describe，发现主从分区总有切换，如原本partition 0的leader 为broker 1，会变为broker 2。查看kafka日志，发现了下面的信息，但是没看太明白，网上也没找到解决办法。有没有人遇到过类似的情况？

[2015-09-01 13:45:12,402] WARN [Replica Manager on Broker 24]: Fetch request with correlation id 1553237 from client ReplicaFetcherThread-0-24 on partition [fortest2,7] failed due to Leader not local for partition [fortest2,7] on broker 24 (kafka.server.ReplicaManager)
[2015-09-01 13:45:16,938] WARN [KafkaApi-24] Produce request with correlation id 1843066 from client  on partition [fortest2,7] failed due to Leader not local for partition [fortest2,7] on broker 24 (kafka.server.KafkaApis)
[2015-09-01 13:45:12,406] ERROR [ReplicaFetcherThread-0-26], Error for partition [nginxlogtopic,10] to broker 26:class kafka.common.NotLeaderForPartitionException (kafka.server.ReplicaFetcherThread)
[2015-09-01 13:45:12,406] ERROR [ReplicaFetcherThread-0-26], Error for partition [udpToRead,6] to broker 26:class kafka.common.NotLeaderForPartitionException (kafka.server.ReplicaFetcherThread)
[2015-09-01 13:45:16,938] WARN [KafkaApi-24] Produce request with correlation id 1843066 from client  on partition [fortest2,15] failed due to Leader not local for partition [fortest2,15] on broker 24 (kafka.server.KafkaApis)
[2015-09-01 13:45:16,941] WARN [KafkaApi-24] Produce request with correlation id 1837853 from client  on partition [fortest2,7] failed due to Leader not local for partition [fortest2,7] on broker 24 (kafka.server.KafkaApis)
[2015-09-01 13:45:16,941] WARN [KafkaApi-24] Produce request with correlation id 1837853 from client  on partition [fortest2,15] failed due to Leader not local for partition [fortest2,15] on broker 24 (kafka.server.KafkaApis)
[2015-09-01 13:45:16,945] WARN [KafkaApi-24] Produce request with correlation id 1836177 from client  on partition [fortest2,7] failed due to Leader not local for partition [fortest2,7] on broker 24 (kafka.server.KafkaApis)
[2015-09-01 13:45:16,945] WARN [KafkaApi-24] Produce request with correlation id 1836177 from client  on partition [fortest2,15] failed due to Leader not local for partition [fortest2,15] on broker 24 (kafka.server.KafkaApis)
[2015-09-01 13:45:16,946] WARN [KafkaApi-24] Produce request with correlation id 1840172 from client  on partition [fortest2,7] failed due to Leader not local for partition [fortest2,7] on broker 24 (kafka.server.KafkaApis)
[2015-09-01 13:45:16,946] WARN [KafkaApi-24] Produce request with correlation id 1840172 from client  on partition [fortest2,15] failed due to Leader not local for partition [fortest2,15] on broker 24 (kafka.server.KafkaApis)

感觉应该是在某个broker down掉后，从分区接手时才应该切换。但是broker并没有down