cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Robert Coli <rc...@eventbrite.com>
Subject Re: GossipingPropertyFileSnitch vs Ec2Snitch
Date Fri, 21 Aug 2015 18:22:42 GMT
On Fri, Aug 21, 2015 at 2:37 AM, Alain RODRIGUEZ <arodrime@gmail.com> wrote:

> So what would be the procedure to go from EC2Snitch to GPFS ?
>
> I see 2 possibilities and I am not sure of the result of each of these:
>
> - Node by node, on a rolling restart way, and making sure that you use the
> former DC names and racks. Copying the EC2 network. Would this be doable or
> would that need a full cluster restart at once ?
> - Add the new DC using directly GPFS on all the node without modifying the
> old one (hybride: 1 DC EC2Snitch, 1 DC GPFS), again copying the EC2 DCs and
> racks. Would this work ?
>

I believe both of these would work. If you have the extra nodes I probably
would do the additional-DC method.

I also recommend :

1) get a list of one or more partition keys for all ranges in your cluster
2) test them with nodetool getendpoints before changing snitch
3) test them with nodetool getendpoints after changing snitch


> Also, Is this need of mirroring the exact DC and racks name mandatory in
> those 2 procedures as I think it is ?
>

In the spoof-ec2snitch method, yes. If you don't, Very Bad Things will
happen.

In the additional-DC method, it doesn't matter what you call the new DC or
racks, just that they are configured properly.

=Rob

Mime
View raw message