[
https://issues.apache.org/jira/browse/ACCUMULO-1766?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13805882#comment-13805882
]
Adam Fuchs commented on ACCUMULO-1766:
--------------------------------------
Seems like #1 is only bad in the case that an HDFS instance is removed, which
can only happen if there is some migration path. #2 would probably be less
vulnerable to breaking if the migration of namenode data is managed outside of
Accumulo. Does anyone know if such a thing is possible. If we do go with #1
then we should have a script to handle the root tablet directory pointer
migration and documentation around it.
> Changing Accumulo config can prevent locating root table files.
> ---------------------------------------------------------------
>
> Key: ACCUMULO-1766
> URL: https://issues.apache.org/jira/browse/ACCUMULO-1766
> Project: Accumulo
> Issue Type: Sub-task
> Components: master, tserver
> Reporter: Keith Turner
> Priority: Blocker
> Fix For: 1.6.0
>
>
> Looking at the code related to the root tablet I think it currently chooses
> the first uri listed in instance.volumes and uses that to store root table
> files. If this config changes it seems like Accumulo will not see its root
> table files.
> Two possible ways of handling this are
>
> # Have init choose a uri from instance.volumes and store that in zookeeper,
> much like each tablet uri is chosen and stored in the metadata table.
> # user all namenodes listed in instance.volumes to store root table files.
> #1 seems like the best option. #2 is still vulnerable to config changes
> causing a problem
--
This message was sent by Atlassian JIRA
(v6.1#6144)
