hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-18102) [SHELL] Purge commands that allow by-pass of Master; e.g. close of a region sent to regionserver explicitly
Date Sat, 17 Jun 2017 03:30:01 GMT

    [ https://issues.apache.org/jira/browse/HBASE-18102?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16052665#comment-16052665
] 

stack commented on HBASE-18102:
-------------------------------

See over in HBASE-18227. [~uagashe] found that the Admin has #closeRegion and calling it does
damage to your cluster now AMv2 is in because it skirts the master to ask the regionserver
to close the region. Instead we should be calling unassign. Filed HBASE-18231 to do the cleanup.

This issue then depends on HBASE-18231. It needs to happen before this. Shell then needs to
work w/ HBASE-18231 by providing alternatives or making the thrown exception go down easier.

> [SHELL] Purge commands that allow by-pass of Master; e.g. close of a region sent to regionserver
explicitly
> -----------------------------------------------------------------------------------------------------------
>
>                 Key: HBASE-18102
>                 URL: https://issues.apache.org/jira/browse/HBASE-18102
>             Project: HBase
>          Issue Type: Sub-task
>          Components: Operability, shell
>            Reporter: stack
>             Fix For: 2.0.0
>
>
> In AMv2, if a RS is not aligned with Master notions of how the world is, then the Master
will kill the deviant RS (TODO: is forcing compliance via less radical means -- but that is
how it is currently).
> The shell currently allows by-passing the Master to make cluster modifications such as
our being able to send a close directly to a RegionServer for it to execute locally. This
facility was used in the past to do fix-up when Master lost account of Region locations. In
the new regime, such mis-accounting should no longer happen and, should a user mistakenly
do an explicit close against a RS, the consequences will be more than the user bargained for;
the Master will shut down the RS as soon as it reports close of a Region the master thinks
should be open (No independence allowed!).
> This issue is to review shell Region and Table manipulation commands to purge those that
by-pass Master or at least to add big warning.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message