On Tue, Jan 13, 2009 at 8:29 PM, Dean Landolt <[email protected]> wrote: > On Tue, Jan 13, 2009 at 6:44 PM, Nicolas Fouché <[email protected]>wrote: > >> OK, that's what I was thinking. I currently have all my documents in >> full-text search indexes. I guess I'll use CouchDB for storage and >> some reduce functions to build dashboards, and not for my >> "complicated" queries. >> I don't want to load my indexes down with the document content. For >> more efficiency, they would only contain the reverse indexes. >> >> As far as I know, the _external process cannot return document ids to >> CouchDB, so CouchDB returns the full document to the caller. Would you >> recommend doing this on the cliend side ? Query the index, and then >> query CouchDB with a bunch of keys ? > > > This is probably a question for Paul D as he's done quite a bit of work in > this area (and is also working on a pure Erlang FTI as couch plugin). The > _external hook returns full docs, sure, but you control the scripts that > pump data into your external indices. >
Unless I'm very much mistaken on changes made to the _external code, there's nothing that requires you to return full docs whatsoever. The complete response can be specified by the OS process. _external is extremely shallow in terms of what it requires as a response. You can mimic views or do something entirely different with them. For all it cares, you could draw mandlebrot fractals in JSON and return that as a body. > But don't forget that couch is more than a db -- it's an app server. It > wouldn't hurt to poke around github for some of the pdavis work on this. You > may be surprised with what you could accomplish by using couch as a proxy to > and interface for other query mechanisms. Also, I'm pretty sure some of this > FI stuff is set for inclusion by 1.0, which is nice. >
