flink 1.7.2集群异常退出

Hi，ALL

写了个测试程序，大概跑了不到三个小时，flink集群就挂了，所有节点退出，报错如下：

2019-03-12 20:45:14,623 INFO org.apache.flink.runtime.executiongraph.ExecutionGraph - Job Tbox from Kafka Sink To Kafka And Print (21949294d4750b869b341c5d2942d499) switched from state RUNNING to FAILING. org.apache.hadoop.ipc.RemoteException(org.apache.hadoop.hdfs.protocol.FSLimitException$MaxDirectoryItemsExceededException): The directory item limit of /tmp/ha is exceeded: limit=1048576 items=1048576

hdfs count结果：

2097151 4 124334563 hdfs://banma/tmp/ha

下面是flink-conf.yaml的配置：

[hdfs@qa-hdpdn06 flink-1.7.2]$ cat conf/flink-conf.yaml |grep ^[^#] jobmanager.rpc.address: 10.4.11.252 jobmanager.rpc.port: 6123 jobmanager.heap.size: 1024m taskmanager.heap.size: 1024m taskmanager.numberOfTaskSlots: 10 parallelism.default: 1 high-availability: zookeeper high-availability.storageDir: hdfs://banma/tmp/ha high-availability.zookeeper.quorum: qa-hdpdn05.ebanma.com:2181 rest.port: 8081

flink版本：官方最新的flink 1,7.2

为什么 high-availability.storageDir的目录会产生如此多的子目录？里面存的都是什么？什么情况下回触发这些存储操作？如何避免这个问题？

谢谢！*来自志愿者整理的flink邮件归档