On Dec 20, 2013, at 2:19 PM, Jens Rantil <[email protected]> wrote:

> The current implementation only supports basic projections because reduce
> input order is undefined in Couch.

I _think_ that in a standard CouchDB (but not BigCouch) the rows are passed to 
the reduce function in ascending key order.

> I've heard of the plugin system. Would
> ordered reduce be something that could be implemented as a plugin?

Well, the ordering is left undefined because of clustered/sharded 
implementations like BigCouch, where the documents and index rows can be split 
across multiple machines. In a setup like that, I imagine ordered reduce 
wouldn’t scale well because it would require transferring index rows between 
the nodes so that a reduce function running on one node could work on a 
contiguous set of rows. So if you had this, you might not want it…

—Jens

Reply via email to