hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Vivek Krishna <vivekris...@gmail.com>
Subject Re: Manual Region Splitting Question.
Date Tue, 22 Mar 2011 19:07:55 GMT
For eg., lets assume I have keys in range of aa, ab,ac..zz

Using the sample data I create regions like this

aa-ba region 1
ca-da region 2 etc.,

The reason why I did not create region bb-bz because I did not encounter in
the sample.q
But when I encounter a key like bb, it does not fall in the region I created
and hence follow the normal procedure I guess. ?

Also, I have incoming  data distributed almost evenly.
Viv



On Tue, Mar 22, 2011 at 2:55 PM, Jean-Daniel Cryans <jdcryans@apache.org>wrote:

> It depends if you are also inserting in an ordered fashion right? Even
> if you have regions a through z, but you start inserting only keys
> with starting with "a", then you'll only hit the first regions.
>
> J-D
>
> On Tue, Mar 22, 2011 at 11:46 AM, Vivek Krishna <vivekrishna@gmail.com>
> wrote:
> > I have GBs of data to be dumped to HBase.  After lots of trials and
> reading
> > through the mailing list, I figured out creating regions manually is a
> good
> > option because all data was hitting one node initially...
> >
> > My approach to creating regions is as follow.
> >    - I sampled like about 1% of the actual data and created say 'n'
> regions
> > based on this sample.
> >
> > Now while doing the insertions, it still hits one node first and then
> > spreads out.
> >
> > Our theory is that, the key it encounters while inserting does'nt fall in
> > the region that we created(using the sample) and hence it inserts as it
> > would do normally.
> >
> > So, has anyone approached this problem in a smarter way ?
> >
> > Viv
> >
>

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