Add a test to make sure that node decomission does not get blocked by
underreplicated blocks in an unclosed file
----------------------------------------------------------------------------------------------------------------
Key: HDFS-694
URL: https://issues.apache.org/jira/browse/HDFS-694
Project: Hadoop HDFS
Issue Type: Test
Affects Versions: 0.21.0
Reporter: Hairong Kuang
Fix For: 0.21.0
We have a cluster that took much longer time to decommission datanodes than
usual. It turned out to be caused by an open file. In HDFS 0.20 when a file is
open, NN does not schedule any block belonged to this file to replicate. So if
an open file has an under-replicated block with a replica on a decommission
datanode, the decommission won't complete until the file is closed.
The new append implementation changed the replication strategy that any
complete block in an unclosed file will be scheduled to be replicated if it
becomes under-replicated. This jira aims to add a test to make sure that node
decommission does not get blocked by under-replicated blocks in an open file.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
