Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Accumulo, mail # user - tablet distribution


Copy link to this message
-
Re: tablet distribution
ameet kini 2012-07-16, 13:34
Thanks Billie/John/Eric.

TableLoadBalancer does the trick.

Ameet

On Fri, Jul 13, 2012 at 5:21 PM, Billie J Rinaldi <[EMAIL PROTECTED]
> wrote:

> On Friday, July 13, 2012 5:14:53 PM, "John Vines" <[EMAIL PROTECTED]>
> wrote:
> > Sounds like you want to use the TableLoadBalancer which will try to
> > even balance all tablets on a table-by-table basis.
>
> To use this, set "master.tablet.balancer" to
> "org.apache.accumulo.server.master.balancer.TableLoadBalancer".  This has
> been made the default balancer as of 1.4.1.
>
> Billie
>
>
> > John
> >
> >
> > On Fri, Jul 13, 2012 at 5:09 PM, ameet kini < [EMAIL PROTECTED] >
> > wrote:
> >
> >
> > Thanks, I was looking for something of the equivalent:
> >
> http://hbase.apache.org/apidocs/org/apache/hadoop/hbase/client/HBaseAdmin.html#move(byte[]
> ,
> > byte[])
> >
> >
> > The use case is common enough that there may be some other way to do
> > what I want. I have a table that doesn't have its tablets distributed
> > evenly. At the system (instance?) level, they are evenly distributed,
> > but not at this particular table level.
> >
> >
> > The docs seem to suggest that pre-splitting a table would
> > automatically guarantee that the tablets would be distributed evenly
> > across the nodes of the cluster. I see that pre-splitting guarantees
> > that you have that many tablets, but not necessarily that they are
> > evenly distributed. Maybe if the instance had only one table, then it
> > would be forced to distribute its tablets.
> >
> >
> >
> > Thanks,
> > Ameet
> >
> >
> > PRE-SPLITTING TABLES
> >
> >
> > Accumulo will balance and distribute tables accross servers. Before a
> > table gets large, it will be maintained as a single tablet on a single
> > server. This limits the speed at which data can be added or queried to
> > the speed of a single node. To improve performance when the a table is
> > new, or small, you can add split points and generate new tablets.
> >
> > In the shell:
> > root @myinstance > createtable newTable root @myinstance > addsplits -
> > t newTable g n t
> >
> > This will create a new table with 4 tablets. The table will be split
> > on the letters g'', n'', and ``t'' which will work nicely if the row
> > data start with lower-case alphabetic characters. If your row data
> > includes binary information or numeric information, or if the
> > distribution of the row information is not flat, then you would pick
> > different split points. Now ingest and query can proceed on 4 nodes
> > which can improve performance.
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > On Fri, Jul 13, 2012 at 3:04 PM, Eric Newton < [EMAIL PROTECTED] >
> > wrote:
> >
> >
> > Yes, you need to write your own tablet balancer.
> >
> > -Eric
> >
> >
> >
> > On Fri, Jul 13, 2012 at 2:48 PM, ameet kini < [EMAIL PROTECTED] >
> > wrote:
> > >
> > > Hi,
> > >
> > > Is there a way to force a tablet to move to a particular tablet
> > > server?
> > >
> > > Thanks,
> > > Ameet
>