[
https://issues.apache.org/jira/browse/HDFS-6395?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14012941#comment-14012941
]
Andrew Wang commented on HDFS-6395:
-----------------------------------
Cool, that's fine with me. I thought a bit more after I posted my HDFS-2006
comment. Basically, I want to avoid the configuration complexity of per-NS
confs, ensure that future internal HDFS features (i.e. HDFS-6134) can use
XAttrs without depending on XAttrs being enabled, but still give admins
flexibility to control user usage of XAttrs. What do you think about the
following?
- XAttr enable/disable config flag and limits configs only affects
user-settable namespaces (i.e. user and trusted)
- System and security namespace have no limits, and are still available even if
xattrs are disabled
Thanks Liu Yi!
> Assorted improvements to xattr limit checking
> ---------------------------------------------
>
> Key: HDFS-6395
> URL: https://issues.apache.org/jira/browse/HDFS-6395
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: namenode
> Affects Versions: 3.0.0
> Reporter: Andrew Wang
> Assignee: Yi Liu
>
> It'd be nice to print messages during fsimage and editlog loading if we hit
> either the # of xattrs per inode or the xattr size limits.
> We should also consider making the # of xattrs limit only apply to the user
> namespace, or to each namespace separately, to prevent users from locking out
> access to other namespaces.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
