You are viewing a plain text version of this content. The canonical link for it is here.

Posted to user@cassandra.apache.org by Rana Aich <ai...@gmail.com> on 2010/07/29 19:09:05 UTC

Date loading pattern with OrderPreservingPartiton

Hi All,

We are working with a Cassandra Cluster consisting 3 nodes with each having
storage capacity of 0.5 Terabytes.
We are loading the data into the cluster with OrderPreservingPartition and
with Replication Factor 2.

The Data that has been loaded so far looks as follows:

Address       Status     Load          Range
     Ring
                                             vHexMxAF1Te2bOvu

Server1 Up         292.98 GB     UHYqNsmZpD8DtZvw
 |<--|
Server3 Up         292.77 GB     ghbzYZ6U6uYQbK6g
|   |
Server2 Up         917 bytes       vHexMxAF1Te2bOvu
  |-->|

As you can see the server2 is still empty, we guess it is for
OrderPreservingPartion that we are using.

Is this an expected behavior of Cassandra?

Raich

Re: Date loading pattern with OrderPreservingPartiton

Posted by Aaron Morton <aa...@thelastpickle.com>.

Yes the OPP will could give you a distribution like that.

Given that only two nodes have data, and they seem to have same amount of data. I wonder if all your keys are falling into the key range of the last node? So with RF 2 they go to the last and first node only.

As an experiment you could try running the nodetool loadbalance command against node 2 and see how the distribution and tokens change. Then perhaps try against node 1.   http://wiki.apache.org/cassandra/Operations

You'll need to consider how your keys are distributed when using the OPP and set the tokens to get the best distribution. 

Aaron

On 30 Jul, 2010,at 05:09 AM, Rana Aich <ai...@gmail.com> wrote:

> Hi All,
>
> We are working with a Cassandra Cluster consisting 3 nodes with each having storage capacity of 0.5 Terabytes.
> We are loading the data into the cluster with OrderPreservingPartition and with Replication Factor 2.
>
> The Data that has been loaded so far looks as follows:
>
> Address       Status     Load          Range                                      Ring
>                                              vHexMxAF1Te2bOvu                         
> Server1 Up         292.98 GB     UHYqNsmZpD8DtZvw                          |<--|
> Server3 Up         292.77 GB     ghbzYZ6U6uYQbK6g                           |   |
> Server2 Up         917 bytes       vHexMxAF1Te2bOvu                           |-->|
>
> As you can see the server2 is still empty, we guess it is for OrderPreservingPartion that we are using.
>
> Is this an expected behavior of Cassandra?
>
> Raich