[ 
https://issues.apache.org/jira/browse/HBASE-13550?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14510049#comment-14510049
 ] 

Andrew Purtell commented on HBASE-13550:
----------------------------------------

We are doing something that's a bit ugly but should be supportable, which is: 
copy data from an old 0.94 based cluster to a new 0.98 based one, run the 
upgrade procedure on the new, stand up service on the new, complicated by a 
concomitant Phoenix upgrade. The upgrade from 0.94 to 0.98 competes 
successfully but we can't deploy regions afterward because the table 
coprocessor descriptors are invalid, Phoenix coprocessor class names change 
between versions 3.x and 4.x. We must change them. In order to change them, we 
need to get (at least) META deployed, but we cannot get a RS to stay up unless 
we set 'hbase.coprocessor.abortonerr' to "false". This is reasonable enough. We 
set that. Now we can get META deployed but cannot remove the old/invalid 
coprocessor attributes from the table descriptors because alter only lets you 
delete one attribute at a time. There are multiple coprocessor attributes in 
the table descriptor, and although we remove one the remaining invalid 
attributes fail the sanity check, failing the modifyTable action. Seems the 
cleanest way past this roadblock is a new generally useful shell modification.

> [Shell] Support unset of a list of table attributes
> ---------------------------------------------------
>
>                 Key: HBASE-13550
>                 URL: https://issues.apache.org/jira/browse/HBASE-13550
>             Project: HBase
>          Issue Type: Improvement
>    Affects Versions: 0.98.10.1
>            Reporter: Andrew Purtell
>            Assignee: Andrew Purtell
>             Fix For: 2.0.0, 1.1.0, 0.98.13, 1.0.2
>
>
> Introduce a backwards compatible shell option for unsetting a list of table 
> attributes at once. It's a trivial change. In admin.rb, if method is 
> 'table_att_unset', check if 'name' is an array, and if so handle that case.  



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to