[ 
https://issues.apache.org/jira/browse/HBASE-13076?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14349619#comment-14349619
 ] 

Andrey Stepachev commented on HBASE-13076:
------------------------------------------

[~Apache9] That is strongly related to what I did with table states. Before 
that it was possible to create table descriptor and then add regions. 
During working on my patches for moving states into meta (and into descriptors 
before) I cleaned up those places and all clients now create table through rpc 
to master. Also hbasefsck works with states more carefully, it recreates 
missing states for tables.
So basicly I think that most places are cleaned up and should work.

That is for 2.0. For branch-1 we recently rolled back meta states, so it seems 
that issues can arise, because it still possible to have table without state.


> Table can be forcibly enabled in AssignmentManager during table disabling.
> --------------------------------------------------------------------------
>
>                 Key: HBASE-13076
>                 URL: https://issues.apache.org/jira/browse/HBASE-13076
>             Project: HBase
>          Issue Type: Bug
>          Components: master, Region Assignment
>    Affects Versions: 2.0.0
>            Reporter: Andrey Stepachev
>            Assignee: Andrey Stepachev
>             Fix For: 2.0.0
>
>         Attachments: 23757f039d83f4f17ca18815eae70b28.log, HBASE-13076.patch
>
>
> Got situation where region can be opened while table is disabling by 
> DisableTableHandler. Here is relevant log for such situation. There is no 
> clues who issued OPEN to region.
> Log file attached.
> UPD: A bit more details. It seems that even in case of new state put into 
> meta, it still possible to get previous state.
> That leads to one more round of assignment invoked in 
> AssignmentManager#onRegionClosed.
> UPD: Table become ENABLED, thats leads to regions instructed to assign 
> immediately on onRegionClosed. BulkDisabler will not know about that and will 
> wait indefinitely, because it will not issue unassign for newly opened 
> regions.



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

Reply via email to