hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Tao Xiao <xiaotao.cs....@gmail.com>
Subject Re: How to make a given table spread evenly across the cluster
Date Sun, 05 Oct 2014 09:28:50 GMT
Hi all, I checked what all you said.


#1.  Balancer
I used the command "*balance_switch true*" in the HBase shell and that
command returned true, indicating that the load balancer was running. Then
I used the command "*balancer*" to prompt HBase to load balance its
regions. However after more than 5 minutes, region assignment for the table
I mentioned in the first email did not change at all, see this screen shot
<http://imgbin.org/index.php?page=image&id=19618>.  I checked the master
log since the time I manaully trigger a balance operation by the command "
*balancer*", and here is the master log <http://pastebin.com/PagJeB1f>.
Since the load balancer is running all the time, it's weird to see that
regions of the table in concern were never balanced.


#2. Balancer configuration
I'm using CDH 5.1.0 and I did not find where to set balance configuration
in Cloudera Manager. Could anybody tell me how to ?


#3. HBASE-12139 StochasticLoadBalancer
My HBase cluster has about ten tables, and none of these tables has over 20
regions. The HBase cluster consisted of 5 nodes so the issue HBASE-12139
does not apply to my problem.



Thanks





2014-10-03 20:31 GMT+08:00 Jean-Marc Spaggiari <jean-marc@spaggiari.org>:

> It depends on the time between the 2 screen-shots.  But if there is more
> than 5 minutes (default config), then you are correct and it should have
> been a minimum balanced.
>
> 2014-10-03 3:30 GMT-04:00 Qiang Tian <tianq01@gmail.com>:
>
> > according to the 2 pictures, looks the balancer not run?
> >
> > On Fri, Oct 3, 2014 at 1:45 AM, Ted Yu <yuzhihong@gmail.com> wrote:
> >
> > > See also
> > > HBASE-12139 StochasticLoadBalancer doesn't work on large lightly loaded
> > > clusters
> > >
> > > Cheers
> > >
> > > On Thu, Oct 2, 2014 at 10:35 AM, Jean-Marc Spaggiari <
> > > jean-marc@spaggiari.org> wrote:
> > >
> > > > Also, What is the balancer configured on your cluster? Which HBase
> > > version?
> > > > If you have a 200 region table in a cluster with another pretty small
> > 13
> > > > region table like the one you have, chances are high for this 2nd
> table
> > > to
> > > > not be perfectly balanced over the cluster compared to the first one.
> > > >
> > > > JM
> > > >
> > > > 2014-10-02 13:30 GMT-04:00 Nick Dimiduk <ndimiduk@gmail.com>:
> > > >
> > > > > Is the balancer running? If it's running but unable to act, you
> > should
> > > > see
> > > > > some explanation in the Master logs.
> > > > >
> > > > > On Wed, Oct 1, 2014 at 8:27 PM, Tao Xiao <xiaotao.cs.nju@gmail.com
> >
> > > > wrote:
> > > > >
> > > > > > Hi all,
> > > > > >
> > > > > > I have a HBase table containing 40 million records. Checking
> HBase
> > > UI,
> > > > I
> > > > > > could see that this table was not spread evenly across the
> cluster.
> > > Two
> > > > > > nodes, *ec1.hadoop.com:60020 <http://ec1.hadoop.com:60020>*
and
> > > > > > *ec5.hadoop.com:60020
> > > > > > <http://ec5.hadoop.com:60020>*, for example, has only
one region
> > > each,
> > > > > > while other nodes have more regions. Here is the screen shot
> > > > > > <http://imgbin.org/index.php?page=image&id=19571>.
> > > > > >
> > > > > > I tried to split that table by clicking the button "Split".
> Later,
> > > each
> > > > > > region of that table was split into two regions, but all regions
> > for
> > > > that
> > > > > > table as a whole were still unevenly spread across the cluster.
> > Here
> > > > > > is a screen
> > > > > > shot after I click the button "Split"
> > > > > > <http://imgbin.org/index.php?page=image&id=19572>.
> > > > > >
> > > > > > So I'd like to know how to adjust the regions for a table so
that
> > all
> > > > > > regions for that table can be evenly spread across the cluster.
> > > > > >
> > > > > > Thanks
> > > > > >
> > > > >
> > > >
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message