NoSQL combinations - what works best?

Les Mikesell lesmikesell at
Tue Aug 9 14:56:09 EDT 2011

On 8/9/2011 1:33 PM, Matt Ranney wrote:
>     The main point of memcache is that it is distributed and the clients
>     automatically handle failover at any reasonably large scale.  How do
>     you arrange that with redis?
> You can achieve this with Redis with a little bit of client-side magic.
>   Antirez suggests this:
> I've implemented a similar scheme with Redis, except also added a
> replica for every shard.  The clients in this case know about all shards
> and all replicas and do failover themselves.
> I really like Redis, but once Riak gets secondary indicies, I hope to be
> able to scrap this big cluster of Redis servers and keep everything in Riak.
> Redis is really good for being just one of something.  You can use it
> for locks, unique ID generation, etc., but you somehow need to address
> the failure case.

Everything breaks.  I don't want to live with just one of something. 
Just 2 of something would be good enough for most the things that will 
fit on one machine.  In some cases I wouldn't mind 2 independent data 
collectors but don't know how well riak would de-dup the data where both 
values are written at the same time as new items with no vector clock 

   Les Mikesell
    lesmikesell at

More information about the riak-users mailing list