[jira] [Commented] (HBASE-21892) If a new table is created inside a namespace then Namespace space quota violation is not working for first time

["Sakthi (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HBASE-21892?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16902538#comment-16902538 ] 

Sakthi commented on HBASE-21892:
--------------------------------

Ah! This is a nice catch. So this is actually an expected behavior. Here is what's happening:
 * Namespace quota set & violated. (with 'nfs1' table usage at 12 mb vs the set quota of 10 mb)
 * A new table 'nfs2' is created. (so by default 1 region initially if no split keys). Let's assume this was just a simple create command rather than using PE again.
 * The quotaobserverchore at master side lists out this new table under potential list of tables for the set namespace quota
 * But then because it hasn't yet received usagereport from rs about the only region/ or threshold amount of regions of this table, hence filters it out from considerationand continues to stop writes from nfs1 as the quota is violated. But nfs2 is not observed under quota and hence a put/flush command meanwhile would work fine.
 * This can go on, until the RS sends in it's first report and the table is not filtered out at master side and the quota violation for nfs2 table is persisted to the quota table. 
 * Even then, till spacerefresherchore from the rs side runs and detects this table being in the list of violated tables from hbase:quota, the puts etc can still carry on.
 * And finally, when it is detected by that chore, the appropriate policy is enforced.

 

[~elserj], please correct me if I am wrong. 

But I think it's worthwhile to discuss if we should allow '*create*' of a table when we know that it's for a namespace which is already under violation, and if continue to allow creation anyway then I don't there is any way to completely stop the writes etc till the above mention steps are completed. We can at best try to better the notification mechanism but still it would not be as spontaneous as expected. Josh, what do you think? :) 

> If a new table is created inside a namespace then Namespace space quota violation is not working for first time
> ---------------------------------------------------------------------------------------------------------------
>
>                 Key: HBASE-21892
>                 URL: https://issues.apache.org/jira/browse/HBASE-21892
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Ajeet Rai
>            Assignee: Sakthi
>            Priority: Minor
>              Labels: Quota, Space
>
> If a new table is created inside a namespace then NS quota violation is not working for first time.
> Steps to reproduce:
> 1: Set a space quota at NameSpace level
> set_quota TYPE => SPACE, NAMESPACE => 'game', LIMIT => '10M', POLICY => NO_WRITES
> 2: create a table and add insert data such that NS quota is violated
>  create_namespace 'game'
> ./hbase pe --table="game:nfs" --nomapred --rows=1200  sequentialWrite  10
> 3: Try to load data in table again and observe that quota is applicable
> 4: Now create another table under same NS and put some data
> ./hbase pe --table="game:nfs2" --nomapred --rows=120  sequentialWrite  10
> 5: Observe that data is inserted in new table which should not happen due to namespace quota is already violated



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)