hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From st...@duboce.net
Subject Re: Review Request: HBASE-2819 HBCK repair problems (new master remix)
Date Fri, 22 Oct 2010 20:05:11 GMT

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://review.cloudera.org/r/1036/#review1622
-----------------------------------------------------------



trunk/src/main/java/org/apache/hadoop/hbase/master/AssignmentManager.java
<http://review.cloudera.org/r/1036/#comment5479>

    Should we do a notifyall on rit after this remove?  I'm looking at regionOnline.  It does
a notify all if the remove actually removed something.  In fact, can we do a little refactoring
so we share the first part of regionOnline between regionOnline and here?



trunk/src/main/java/org/apache/hadoop/hbase/master/AssignmentManager.java
<http://review.cloudera.org/r/1036/#comment5480>

    Is this safe?  We need a sync on regions above, but not here?  And if you sync, is there
danger of deadlock?



trunk/src/main/java/org/apache/hadoop/hbase/master/HMaster.java
<http://review.cloudera.org/r/1036/#comment5474>

    this is an odd check?  why not have the check done inside in clearRegionFromTransition...
Its a little racy?


- stack


On 2010-10-20 17:47:11, Jonathan Gray wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> http://review.cloudera.org/r/1036/
> -----------------------------------------------------------
> 
> (Updated 2010-10-20 17:47:11)
> 
> 
> Review request for hbase and stack.
> 
> 
> Summary
> -------
> 
> Migration of latest HBCK patch made against old master by Alex.
> 
> I'm not sure this really makes sense anymore on the new master.  Dupe assignment and
non-assignment in the "traditional" cases should not happen anymore, at least not in the way
it used to happen.
> 
> This is mostly a straight migration and adds stuff to clear things out of transition
on master, close a region on an RS w/o using ZK (already supported just exposed via rpc now),
etc...
> 
> I don't think it hurts to have this code in there but mostly the detection code should
be used for now.  Once we trip over assignment issues / hbck fixable issues, we can add fixup
to this new stuff.
> 
> Can just rip out the repair for now but figured would be simpler just to leave it and
continue improving it rather than going back a step.
> 
> 
> This addresses bug HBSAE-2819.
>     http://issues.apache.org/jira/browse/HBSAE-2819
> 
> 
> Diffs
> -----
> 
>   trunk/src/main/java/org/apache/hadoop/hbase/client/HBaseFsck.java 1025790 
>   trunk/src/main/java/org/apache/hadoop/hbase/client/HBaseFsckRepair.java PRE-CREATION

>   trunk/src/main/java/org/apache/hadoop/hbase/ipc/HRegionInterface.java 1025790 
>   trunk/src/main/java/org/apache/hadoop/hbase/master/AssignmentManager.java 1025790 
>   trunk/src/main/java/org/apache/hadoop/hbase/master/HMaster.java 1025790 
>   trunk/src/main/java/org/apache/hadoop/hbase/regionserver/HRegionServer.java 1025790

> 
> Diff: http://review.cloudera.org/r/1036/diff
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Jonathan
> 
>


Mime
View raw message