[
https://issues.apache.org/jira/browse/HBASE-1154?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12667315#action_12667315
]
stack commented on HBASE-1154:
------------------------------
Two things come to mind:
+ Hadoop Configuration is hard to customize. Its all final methods and data
members. Somehow, we'd need to rig the HBaseConfiguration so that it ran a
custom classloader, one that read content from zookeeper.
+ Up on IRC, thought was that edits to the hbase-site.xml could be seen
immediately by the cluster. Since classloading happens once in a (class)
lifetime, then maybe we need something extra to loading configuration off the
classpath?
> Store hbase-site.xml in ZooKeeper
> ---------------------------------
>
> Key: HBASE-1154
> URL: https://issues.apache.org/jira/browse/HBASE-1154
> Project: Hadoop HBase
> Issue Type: Improvement
> Affects Versions: 0.19.0
> Environment: All
> Reporter: ryan rawson
> Fix For: 0.21.0
>
>
> Right now hbase-site.xml is stored in the classpath of every individual
> machine. To update parameters of the cluster, one has to copy files to every
> cluster machine then restart the daemons.
> Instead of storing config files in a local disk, store them in Zoo Keeper.
> Regionservers can then put a watch on the file, and do something appropriate
> when the config changes. Restarting en-masse is probably not the best
> solution however.
> A similiar change could be done for HDFS/Hadoop for their own hadoop-site.xml
> file. That is a little out of scope :-)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
