lesmikesell at gmail.com
Wed Jan 11 12:56:08 EST 2012
On Wed, Jan 11, 2012 at 10:39 AM, Ian Plosker <ian at basho.com> wrote:
> In Riak, there is no single primary copy considered the canonical version.
> For each key, there will be N (3 by default) partitions responsible for
> storing the associated value. In effect, there are N primaries for any key.
> This is how Riak makes its availability guarantees, as well as why "absolute
> consistency" is difficult.
But who makes the decision that a partition needs to migrate and where
a key is at any time during that migration? That isn't independently
decided by each node, is it? And if you have an authority for that,
why can't a client ask that authority to control ordering of
operations for certain things where the client is willing to trade the
time it might take for atomicity.
More information about the riak-users