hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Yu <yuzhih...@gmail.com>
Subject Re: hbase config for 0.90
Date Sat, 22 Jan 2011 15:44:46 GMT
I encountered different kinds of failure w.r.t. hbase 0.90 running our flow
over 200GB input data.
Some values from my config were derived from processing 500GB data using
0.20.6 on our staging cluster.

I couldn't find reference to 'hfile.min.blocksize.size' in 0.90 source code.


'hbase.zookeeper.property.maxClientCnxns' is high because of zookeeper issue
I mentioned in another thread.

I will report back after further experiments.

Thanks Stack.

On Fri, Jan 21, 2011 at 9:56 PM, Stack <stack@duboce.net> wrote:

> Remove:
>
> hbase.master
> hbase.regions.slop
> hbase.regionserver.restart.on.zk.expire
> hbase.regions.percheckin
> hbase.zookeeper.property.tickTime
>
> None of the above are pertinent any more.
>
> Your 'hbase.hregion.memstore.flush.size' is very high.  Why so high?
> Same for your hfile blocksize (Though IIRC, you have lots of small
> records so you are probably trying to keep down the size of your
> indices?)
>
> Your 'hbase.regionserver.msginterval' is odd... Default is 3seconds
> and you have 23seconds.  Any reason?
>
> Your zk timeout is very high, higher than default 3 minutes in 0.90.
> Maybe its OK though?
>
> The 'hbase.zookeeper.property.maxClientCnxns' setting of 400 is
> abnormal.  It shouldn't usually have to be this high if you reuse your
> Configuration making HTables... but leave it as is.  Probably does no
> harm.
>
> Hope that helps Ted,
> St.Ack
>
>
>
>
> On Fri, Jan 21, 2011 at 8:09 PM, Ted Yu <yuzhihong@gmail.com> wrote:
> > Hi,
> > I copied hbase-site.xml from our staging cluster running 0.20.6 to dev
> > cluster which runs 0.90
> >
> > Can someone go over the attached file and see if any parameter is
> > incorrectly set ?
> >
> > Thanks
> >
>

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