[
https://issues.apache.org/jira/browse/HBASE-15641?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15251784#comment-15251784
]
Matt Warhaftig commented on HBASE-15641:
----------------------------------------
Just tested, looked good on a fully distributed cluster - alters affected all
nodes as expected.
Also, to be clear, by batching the table updates the alter command has become
atomic. Prior to this patch the below command would still apply the 'cf'
version change even if the table attribute unset failed:
{noformat}
alter 'tableCluster', { NAME => 'cf', VERSIONS => 5 }, { METHOD =>
'table_att_unset', NAME => 'REGION_REPLICATION'}
{noformat}
With this patch the command fails wholesale, which IMHO makes more sense.
> Shell "alter" should do a single modifyTable operation
> ------------------------------------------------------
>
> Key: HBASE-15641
> URL: https://issues.apache.org/jira/browse/HBASE-15641
> Project: HBase
> Issue Type: Improvement
> Components: shell
> Reporter: Gary Helmling
> Assignee: Matt Warhaftig
> Attachments: hbase-15641-v1.patch
>
>
> When performing an "alter" on multiple column families in a table, then shell
> will perform a separate {{Admin.modifyColumn()}} call for each column family
> being modified, with all of the table regions being bulk-reopened each time.
> It would be much better to simply apply all the changes to the table
> descriptor, then do a single call to {{Admin.modifyTable()}}.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
