[
https://issues.apache.org/jira/browse/HBASE-18?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12568395#action_12568395
]
Bryan Duxbury commented on HBASE-18:
------------------------------------
Whenever we find a row like this, we should just immediately delete it. There's
no way that it will ever be salvaged.
Moreover, if this is caused because of an admin concurrency issue, can we find
some way to make it non-concurrent? IE, is there a reason that we can't make
creating and dropping tables atomic and blocking? If you're deleting a 10TB
table, I think you should be prepared to a wait a few minutes in the shell. I'd
much rather know it happened successfully than have it exit quickly.
> [hbase] A sequence of create/delete/create/etc. generates 'info:regioninfo is
> empty; has keys:' and eventually fails
> --------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-18
> URL: https://issues.apache.org/jira/browse/HBASE-18
> Project: Hadoop HBase
> Issue Type: Bug
> Reporter: stack
> Priority: Minor
> Attachments: hbase-dave-master-hendrix.log
>
>
> Marking as minor issue since its a concurrency issue in admin function.
> Though that said, this is a guaranteed way to generate rows in META that are
> without their regioninfo (Looks like we might add startcode and servername
> after a row has had its regioninfo removed).
> Issue was found by Dave Simpson running a create/delete/create sequence.
> Attaching log.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
