cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From aaron morton <aa...@thelastpickle.com>
Subject Re: Tips for using OrderedPartitioner
Date Fri, 10 Feb 2012 09:45:39 GMT
> Also, if there's hot spot is there any way out of it, other than restarting from scratch…
A cluster with a changed partitioner is like a mule with a spinning wheel. No one knows how
it changed and danged if it knows how to return your data . 
(You cannot change it.)

By uniform I meat evenly distributed across the range of values. That is what the RandomPartitioner
does by using the MD5 transform (also means we know that the tokens have finite range).

Cheers


-----------------
Aaron Morton
Freelance Developer
@aaronmorton
http://www.thelastpickle.com

On 10/02/2012, at 8:31 AM, Tharindu Mathew wrote:

> That sounds like writing a DB... indexing the index row.... :)
> 
> By making the keys uniform.... Do you mean like keep the initial X characters the same
or the last Y the same... Could you elaborate, please?
> 
> Also, if there's hot spot is there any way out of it, other than restarting from scratch...
> 
> On Tue, Jan 24, 2012 at 3:50 PM, R. Verlangen <robin@us2.nl> wrote:
> If you would like to index your rows in an "index-row", you could also choose for indexing
the "index-rows". This will scale up for any needs and create a tree structure.
> 
> 
> 2012/1/24 aaron morton <aaron@thelastpickle.com>
> Nothing I can thin of other than making the keys uniform.
> 
> Having a single index row with the RP can be a pain. Is there a way to partition it ?
> 
> Cheers
> 
> 
> -----------------
> Aaron Morton
> Freelance Developer
> @aaronmorton
> http://www.thelastpickle.com
> 
> On 23/01/2012, at 11:42 PM, Tharindu Mathew wrote:
> 
>> Hi,
>> 
>> We use Cassandra in a way we always want to range slice queries. Because, of the
tendency to create hotspots with OrderedPartioner we decided to use RandomPartitioner. Then
we would use, a row as an index row, holding values of the other row keys of the CF.
>> 
>> I feel this has become a burden and would like to move to an OrderedPartioner to
avoid this work around. The index row workaround which has become cumbersome when we query
the data store.
>> 
>> Is there any tips we can follow to allow for lesser amount of hot spots?
>> 
>> -- 
>> Regards,
>> 
>> Tharindu
>> 
>> blog: http://mackiemathew.com/
>> 
> 
> 
> 
> 
> 
> -- 
> Regards,
> 
> Tharindu
> 
> blog: http://mackiemathew.com/
> 


Mime
View raw message