Replication system_distributed

[Marcel Jakobi <ma...@ordix.de>]

Hi,

the default definition of the keyspace system_distributed is:
CREATE KEYSPACE system_distributed WITH replication = {'class': 'SimpleStrategy', 'replication_factor': '3'}  AND durable_writes = true;


If I understand correctly, every repair information will be replicated on three servers in the cluster. I have changed the RF to '1'. Once i stop the entire cluster, the replication factor changes again to 3. It seems Cassandra wants it to be 3.

Doesn`t that reduce performance on repair operation?
Why is the RF changed again after restart the cluster?
Are there reasons why you shouldn`t change the replication factor to 1?

Thanks,
Marcel

Re: Replication system_distributed

[Chris Splinter <ch...@gmail.com>]

Hi Marcel,

The RF for that keyspace is currently hardcoded, see CASSANDRA-11098
<https://issues.apache.org/jira/browse/CASSANDRA-11098>. I am not sure why
your RF switched from 1 to 3 after you restarted a cluster, I tried the
same and it remained at 1 for me.

The tables in that keyspace are used to store history about the repair
operations, having it as RF=3 shouldn't affect the performance of the
repair. See CASSANDRA-5839
<https://issues.apache.org/jira/browse/CASSANDRA-5839> for when / why it
was introduced.

Changing the replication from 3 to 1 for system_distributed is not a good
idea for the same reasons why changing the replication of *any* keyspace to
1 is not a good idea. You lose the ability to query that data if a single
node goes down.

Hope this helps,

Chris

On Wed, Jan 8, 2020 at 1:23 AM Marcel Jakobi <ma...@ordix.de> wrote:

> Hi,
>
>
>
> the default definition of the keyspace system_distributed is:
>
> CREATE KEYSPACE system_distributed WITH replication = {'class':
> 'SimpleStrategy', 'replication_factor': '3'}  AND durable_writes = true;
>
>
>
>
>
> If I understand correctly, every repair information will be replicated on
> three servers in the cluster. I have changed the RF to ‚1‘. Once i stop the
> entire cluster, the replication factor changes again to 3. It seems
> Cassandra wants it to be 3.
>
>
>
> Doesn`t that reduce performance on repair operation?
>
> Why is the RF changed again after restart the cluster?
>
> Are there reasons why you shouldn`t change the replication factor to 1?
>
>
>
> Thanks,
>
> Marcel
>