brett at bretthoerner.com
Thu Aug 4 11:38:31 EDT 2011
This isn't really about scenarios in general, just the example you gave:
On Thu, Aug 4, 2011 at 9:33 AM, Jonathan Langevin <
jlangevin at loomlearning.com> wrote:
> One such scenario that I'd be interested in seeing, is a setup where you
> have a Riak cluster with a specific node that is used for all reads/writes
> of an application.
> The idea is to ensure that the data is always online (via cluster), but
> also ensure that you always have the latest version of your data (specific
> single node read/write).
> If this primary node goes down, then just switch the application to a new
> primary node. That could occur via application logic or other solution that
> exists between application and riak cluster.
Riak has no real notion of "primary" node(s), and you don't want to change
that. (Though I guess the first node in the preference list for a given hash
could be considered the "most primary"... but that's a bit different.)
Anyway, you should be able to achieve strong consistency (read your last
write) by keeping R + W > N. You shouldn't think of Riak in "master slave"
terms you may be used to from your relational database of choice.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the riak-users