cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Robert Coli <rc...@eventbrite.com>
Subject Re: Should replica placement change after a topology change?
Date Tue, 29 Sep 2015 17:40:31 GMT
On Tue, Sep 29, 2015 at 9:35 AM, Dongfeng Lu <dlu66061@yahoo.com> wrote:

> Reading Richard's 6 step plan, I felt that I may just need step 1 for my
> case, that is "Switch to a different snitch (possibly GPFS), and the
> keyspace from SimpleStrategy to NTS but keep it all in one rack." Since I
> only have 3 nodes with repliation_factor of 3 and every node has a replica
> of all the data already, such a switch should be a NO-OP. Even if I assign
> each node to different racks, this is still a NO-OP. Is this the correct
> understanding? Do I need his steps 2 - 6?
>

If you have RF=N and you have repaired your cluster and have not missed any
replication since, all topology changes are a NO-OP, because all data is
everywhere.

While I have lost thread context in my email client, you can generally
treat RF=N as if it is a single node.

A related operation question. When we change the snitch settings in
> cassandra.yaml file, can we do it one node a time? I just wonder what would
> happen if the cluster has different settings on different nodes. Maybe we
> need to stop all the nodes, and apply the changes to cassandra.yaml on all
> nodes, and bring them back altogether.
>

You can do it one at a time.

=Rob

Mime
View raw message