[
https://issues.apache.org/jira/browse/HDFS-6133?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14037378#comment-14037378
]
Daryn Sharp commented on HDFS-6133:
-----------------------------------
Agreed on pinning even if not all favored nodes granted to keep the impl
simpler. I think there's a lot of value in a change like this since there are
multiple jiras attempting different ways to stop the balancer from moving hbase
files. We can always improve later.
Allowing the user to pin/unpin would be useful, but would be much more
complicated so I'd suggest deferring to another jira. Complicated in the sense
that the NN has to be pinning aware so it can issue commands to DNs to pin,
compatibility concerns during rolling upgrades, etc. Only allowing pinning
after the fact instead of during creation does create races where the balancer
got to the block before the pin request; either via tight race, or rpc
connection dropped, failover, etc.
I like pinning during creation. What do others think?
> Make Balancer support exclude specified path
> --------------------------------------------
>
> Key: HDFS-6133
> URL: https://issues.apache.org/jira/browse/HDFS-6133
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: balancer, namenode
> Reporter: zhaoyunjiong
> Assignee: zhaoyunjiong
> Attachments: HDFS-6133-1.patch, HDFS-6133.patch
>
>
> Currently, run Balancer will destroying Regionserver's data locality.
> If getBlocks could exclude blocks belongs to files which have specific path
> prefix, like "/hbase", then we can run Balancer without destroying
> Regionserver's data locality.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
