[
https://issues.apache.org/jira/browse/HDFS-6318?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14028974#comment-14028974
]
Vinayakumar B commented on HDFS-6318:
-------------------------------------
Thanks [~azuryy] for reporting the issue and patch.
As of now, there are multiple Jiras [HDFS-6507, HDFS-5147] discussing
same/similar problem you can also provide your inputs, i.e. handling certail
DFSAdmin commands with HA setup.
Coming to your patch, current approach looks good.
I have one question, how will you handle the failures here? In your current
patch, if the command fails at first namenode it wont go to next.
What if standby namenode is made down for maintainance ..? then this command
may not even go to active namenode, if the standby namenode is first in the
list.
Any thoughts.?
> refreshServiceAcl cannot affect both active NN and standby NN
> -------------------------------------------------------------
>
> Key: HDFS-6318
> URL: https://issues.apache.org/jira/browse/HDFS-6318
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: ha, namenode
> Affects Versions: 2.4.0
> Reporter: Fengdong Yu
> Assignee: Fengdong Yu
> Attachments: HDFS-6318.patch, HDFS-6318.patch
>
>
> refreshServiceAcl cannot affect both active NN and standby NN, it only select
> one NN to reload the ACL configuration. but we should reload Acl on both
> active NN and standby NN.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
