分布式训练时 总是zk 相关错误

错误日志如下：
请求帮助
E1219 13:43:45.176839 17771 zk_server_monitor.cc:150] ZK error when checking root node: connection loss.

Start by inspecting zk_server_monitor.cc at line 150 and the surrounding ZooKeeper monitoring path. Reproduce the distributed-training failure if possible and trace the reported connection-loss error to its cause. Done means identifying and fixing the ZK error or documenting the required configuration or recovery behavior with a regression test.

Written by the indexing model from the issue text.