[
https://issues.apache.org/jira/browse/HBASE-8131?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13604689#comment-13604689
]
Matteo Bertozzi commented on HBASE-8131:
----------------------------------------
Yeah zk should be "persistent", maybe we have all the znode as ephemeral
because we don't want hbase depend on zookeeper state.. the idea is you can
wipe zookeeper and hbase recreate the state.
but in this case is not a real "persistent" information... you can wipe
zookeeper and you're in the situation that you've now, where you've to run hbck
to fix the half create table.
HBASE-7245 was trying to implement in a quick & dirty way the
create/delete/restore/clone rollback on master restart. it was more a draft
that I've abandoned since Enis started working on the table lock and it was the
proper way to go, but still we don't have it.. and this jira seems a good
candidate to reviving the rollback on failures
> Create table handler needs to handle failure cases.
> ---------------------------------------------------
>
> Key: HBASE-8131
> URL: https://issues.apache.org/jira/browse/HBASE-8131
> Project: HBase
> Issue Type: Bug
> Components: master
> Affects Versions: 0.98.0
> Reporter: ramkrishna.s.vasudevan
> Assignee: ramkrishna.s.vasudevan
> Attachments: HBASE-8131_trunk.patch
>
>
> In CreateTable Handler there are number of failure cases.
> IOExceptions are common while creation of regioninfos, htableDescriptors etc.
> After this exception if i try to recreate the table using admin, we need to
> remove the acquired table lock and also clear the ZKTable in memory cache so
> that the operation can be retried.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
