hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bryan Duxbury (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HBASE-478) offlining of table does not run reliably
Date Wed, 30 Apr 2008 03:00:55 GMT

    [ https://issues.apache.org/jira/browse/HBASE-478?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12593210#action_12593210

Bryan Duxbury commented on HBASE-478:

It seems whatever strategy we take for this issue it'll be easier if we do HBASE-451. Maybe
we should do that first?

In general, the problem is that its impossible (or at least impractical) for the client to
block while a table is being offlined. So, instead, I think HTableDescriptor should get a
new flag, something like isGoingDown that indicates it's not offline yet but it's supposed
to be coming down. Then, whenever the client tries to do a schema operation, if the table
is going down but isn't offline yet, we know to reject attempts. Regions for tables that are
offline or going down won't get assigned. 

I'd go one step further and say that we should require explicit disable of table before allowing
schema operations, rather than doing it automatically. It'd be an extra step but I think it
would set expectations a little better about what should happen during schema operations.

> offlining of table does not run reliably
> ----------------------------------------
>                 Key: HBASE-478
>                 URL: https://issues.apache.org/jira/browse/HBASE-478
>             Project: Hadoop HBase
>          Issue Type: Bug
>            Reporter: stack
>            Priority: Critical
>             Fix For: 0.2.0
> I have a table of 4 regions made w/ PE.  I cannot reliably offline it.  I'm using 'disable
TestTable' and have traced it to ensure its not a problem in hql.    What I see is that one
region will get the offlined mark or maybe two.. but never all.
> Jim in IRC suggested that if we did the .TABLE. catalog table, offlining the entry there
might be more reliable than trying to offline all regions in a table.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message