Jing Zhao created HDFS-9789:
-------------------------------
Summary: Correctly update DataNode's scheduled block size when
writing small EC file
Key: HDFS-9789
URL: https://issues.apache.org/jira/browse/HDFS-9789
Project: Hadoop HDFS
Issue Type: Sub-task
Components: namenode
Affects Versions: 3.0.0
Reporter: Jing Zhao
Assignee: Jing Zhao
Currently DatanodeDescriptor's scheduled block size is increased when
allocating a new block. It is then decreased when NN receives IBR from the DN.
When writing an erasure-coded file whose size is less a stripe, we can have
less than 6 internal data blocks. Then we never decrease the scheduled block
size for Datanodes that are actually not used for storing any internal block.
When writing large amount of small EC files within a short period, this may
cause NN to wrongly exclude DN for writing due to high scheduled block size.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
