Mailing-List: contact hbase-dev-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: hbase-dev@hadoop.apache.org
Message-ID: <1272165709.1222984184283.JavaMail.jira@brutus>
Date: Thu, 2 Oct 2008 14:49:44 -0700 (PDT)
From: "stack (JIRA)" <jira@apache.org>
To: hbase-dev@hadoop.apache.org
Subject: [jira] Commented: (HBASE-902) Add force major compaction and force
 split to UI
In-Reply-To: <950836415.1222407824377.JavaMail.jira@brutus>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/HBASE-902?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12636480#action_12636480 ] 

stack commented on HBASE-902:
-----------------------------

Its interesting how you carry then internal split message in HRI.  Thats fine I suppose least I cannot come up with anything more elegant (better to have HStore go to HRI than to its hosting HRegion.  There is a presumption that HRegion and HStore both share HRI instances.   Maybe add a comment to this effect on the HStore HRI data member?

Perhaps name the HRI method 'split' rather than shouldSplit?

Should RegionManager.startSplit and endSplit instead be scheduleSplit and removeSplit or some such?  start and end don't seem to describe whats going on here.

Does getMetaRegionForTableClosest belong in metautils? It looks generally useful?

If an HStore still holds references, will we split (We shouldn't -- or rather, we can't).

This is really great functionality.  I like the way you've made it easily extensible too.

> Add force major compaction and force split to UI
> ------------------------------------------------
>
>                 Key: HBASE-902
>                 URL: https://issues.apache.org/jira/browse/HBASE-902
>             Project: Hadoop HBase
>          Issue Type: Improvement
>            Reporter: stack
>            Assignee: Andrew Purtell
>         Attachments: 902-1.patch
>
>
> Would help debugging or at start of big upload or if we have a hot region we want to distribute over the cluster if we could just manually force the splitting of regions.
> I thought it would be just a little messing in jsp but its a bit more than that.
> Actual decision to split is made down in guts of the Store.  Won't split if any references still around and a file must be > maxsize.
> Chatting w/ Jon Gray, could set a flag on the region that we want to split.  Could do this from UI (Can get to HRegion instance from HRS instance in HRS jsp).  Stores have references to their hosting regions (I think).  When figuring if we're to split, check the force split flag up in the hosting region.
> To get the flag checked, safest way is to just force a flush (Otherwise, have to mess w/ locks in the compact/splitter thread).  Registering a flush request on a region is easy enough to do.  Its a public method on HRS (or HR).

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