Liang Xie created HDFS-6284:
-------------------------------
Summary: don't let checkDiskError() run concurrently
Key: HDFS-6284
URL: https://issues.apache.org/jira/browse/HDFS-6284
Project: Hadoop HDFS
Issue Type: Improvement
Components: datanode
Affects Versions: 2.4.0, 3.0.0
Reporter: Liang Xie
Assignee: Liang Xie
In current codebase, there're amount of datanode.checkDiskError calls, so it
possible occurs that not a few threads call checkDiskError concurrently. IMHO,
it'll be more reasonable if we avoid the concurrent running behavior.
1. FsDatasetImpl.checkDataDir has a "synchronized" keyword, so the concurrent
checking will be done one by one internally.
2. checkDir() operation will take lengthy time once hitting a sick disk. if we
have more than one thread to check disk in series, the cost from upper invoker
like BlockReceiver will be unacceptable.
Patch will be attached later, any comments are welcome !
--
This message was sent by Atlassian JIRA
(v6.2#6252)
