【HDFS】DatanodeAdminBackoffMonitor退役节点极慢的问题定位

作者：凡人多烦事01 | 2024-03-21 17:30:57

踩

一、现象：

下节点特别慢。10台节点，每台大约需要退役60w个块。但是3个小时才退役了3000多个块。

NN侧如下日志，可以看到30秒只退役了512-494 = 20个块，这要是退役600w个块，得猴年马月？

2024-03-19 14:44:42,952 INFO org.apache.hadoop.hdfs.server.blockmanagement.DatanodeAdminBackoffMonitor: There are 10000 blocks pending replication and the limit is 10000. A further 6174331 blocks are waiting to be processed. The replication queue currently has 112518 blocks

2024-03-19 14:45:12,972 INFO org.apache.hadoop.hdfs.server.blockmanagement.DatanodeAdminBackoffMonitor: There are 10000 blocks pending replication and the limit is 10000. A further 6174331 blocks are waiting to be processed. The replication queue currently has 112494 blocks

1
2
3
4

NN监控里，也能发现不少块复制超时了。

DN侧监控，发现只有一台节点在处理复制任务，它的XmitsInProgress线程数不为0，其它的DN都为0ÿ

声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/凡人多烦事01/article/detail/282183