Hi guys, I always hear that DataNode throughput could not improve recently because of the global coarse-grain read-write lock on DataNod especially for federation deployment. Actually about more than one years ago, we split it into a fine-grain read-write lock for BlockPool/Volume grain in my internal branch[1], it works well for a long time and the performance is very good as expected, I invited my colleague Li Mingxiang to contribute this improvement back to the community and now this work is continuing[2].
Now I try to involve more guys which is interested or who are experts on DataNode to give some more comments or help to review the PR[3]. Welcome any suggestions or discussion here. Thanks. Regards - He Xiaoqiao [1] https://issues.apache.org/jira/browse/HDFS-15382 [2] https://issues.apache.org/jira/browse/HDFS-15180 [3] https://github.com/apache/hadoop/pull/3941
