hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HBASE-490) Doubly-assigned .META.; master uses one and clients another
Date Tue, 04 Mar 2008 05:38:51 GMT

     [ https://issues.apache.org/jira/browse/HBASE-490?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

stack updated HBASE-490:
------------------------

    Attachment: 490.patch

Patch for 0.1.  Here is commit message:

HBASE-490  Doubly-assigned .META.; master uses one and clients another

On start, the first two scans run back-to-back.  The first scan of -ROOT-
on restart of a decent size hbase install of more than a single server
will mark the .META. assignment as bad and assign the .META. to a
regionserver to deploy.  The logic in checkingAssignment fixed in the below
was allowing that the second scan would also assign the .META.,
usually to adifferent region server.  Both assignments would report back
in as successful opens.  Both would be added to metaRegionsToScan but only
the first addition would be acted on (I changed code here so we exhaust
metaRegionsToScan before we move out of initialScan).  The second would
never addition would never get promoted to being an online region even
though it being the second open, it would be the .META. that was written
into -ROOT-; it would be what clients would get while the onlineRegions
would have the first .META. in it (Master was getting its .META. info
from this deploy)..

M  src/java/org/apache/hadoop/hbase/HMaster.java
  Mild formatting. Javadoc.  Above described bug fixes.

> Doubly-assigned .META.; master uses one and clients another
> -----------------------------------------------------------
>
>                 Key: HBASE-490
>                 URL: https://issues.apache.org/jira/browse/HBASE-490
>             Project: Hadoop HBase
>          Issue Type: Bug
>          Components: master
>    Affects Versions: 0.1.0, 0.2.0
>            Reporter: stack
>            Priority: Blocker
>             Fix For: 0.1.0, 0.2.0
>
>         Attachments: 490.patch
>
>
> Internal cluster has two .META.,,1 regions up (Its possible for a region to be added
twice to the unassigned map if meta scans run close together).  Worse is that the master is
working with one .META. but when clients come in, they're being give the other.  Makes for
odd results.
> Made it a blocker.  Still trying to track down how master doesn't see subsequent update
of .META. info in -ROOT-.....

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message