[jira] [Comment Edited] (HBASE-11974) When a disabled table is scanned, NotServingRegionException is thrown instead of TableNotEnabledException

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

    [ https://issues.apache.org/jira/browse/HBASE-11974?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14141593#comment-14141593 ] 

Enis Soztutar edited comment on HBASE-11974 at 9/20/14 12:14 AM:
-----------------------------------------------------------------

Attaching v8 patch, which is very similar to Ted's v7 patch, but does not contain the changes in RegionServerCallable which was introduced because of hanging test TestReplicasClient. 

The test caused a hang because the test itself shuts down the master, so any relocateRegion() call hangs waiting to do the isTableDisabled() call. v8 patch instead mocks the Connection to not rely on master for isTableDisabled() call. 

I think we can get this patch in which makes the behavior on par with 0.98 in branch-1. In master instead of going to zk, we go to hmaster. The hmaster RPC is already there even w/o this patch (in regular get). If we want to eliminate hmaster dependency, we can either get rid of relocateRegion() call all together (which will cause an rpc to a disable table take very long via retries), or somehow mark the table state in meta. My proposal is we commit this patch, and do a follow up issue for this.  


was (Author: enis):
Attaching v8 patch, which is very similar to Ted's v8 patch, but does not contain the changes in RegionServerCallable which was introduced because of hanging test TestReplicasClient. 

The test caused a hang because the test itself shuts down the master, so any relocateRegion() call hangs waiting to do the isTableDisabled() call. v8 patch instead mocks the Connection to not rely on master for isTableDisabled() call. 

I think we can get this patch in which makes the behavior on par with 0.98 in branch-1. In master instead of going to zk, we go to hmaster. The hmaster RPC is already there even w/o this patch (in regular get). If we want to eliminate hmaster dependency, we can either get rid of relocateRegion() call all together (which will cause an rpc to a disable table take very long via retries), or somehow mark the table state in meta. My proposal is we commit this patch, and do a follow up issue for this.  

> When a disabled table is scanned, NotServingRegionException is thrown instead of TableNotEnabledException
> ---------------------------------------------------------------------------------------------------------
>
>                 Key: HBASE-11974
>                 URL: https://issues.apache.org/jira/browse/HBASE-11974
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Ted Yu
>            Assignee: Ted Yu
>         Attachments: 11974-test.patch, 11974-v1.txt, 11974-v2.txt, 11974-v3.txt, 11974-v4.txt, 11974-v5.txt, 11974-v6.txt, 11974-v7.txt, hbase-11974_v8.patch
>
>
> When a disabled table is scanned, TableNotEnabledException should be thrown.
> However, currently NotServingRegionException is thrown.
> Thanks to Romil Choksi who discovered this problem.



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