persistant key-value store for node with CouchDB-style views
Fast in-process in-memory key-value store for node with snapshot and AOF persistance and CouchDB-style map-reduce views. Just make sure your data fits in memory, currently I wouldn't recommend divan for a dataset with more than 500K docs.
npm install divan
// Make a divan with local compacted append-only and snapshot files,// namespace works in the same way it does for `dirty`:var divan = require 'divan'db = divancwd 'friends' ;// Save some docs, generating your own id-s:dbsave _id : 'don1' type : 'person' name : 'Don' gender : 'male' ;dbsave _id : 'samantha' type : 'person' name : 'Sam' gender : 'female' ;dbsave _id : 'i.v.a.n' type : 'person' name : 'Ivan' gender : 'male' ;// These will be flushed to the AOF and then later compacted as a db snapshot.// Now register a view:dbaddView'gender/count'divanmrif doctype === 'person'emit docgender 1 ;var i n = vlength sum = 0;for i = 0; i < n; i ++ sum += v i ;return sum;;// Query the view:dbview 'gender/count' group : truedatarowsforEachconsole.log rowkey rowvalue ;;;// Outputs:// female 1// male 2
You can add views via
or by parsing a directory of
design files via
The design-files can either be
.json files of couchdb-design-doc flavour,
or .js files that export objects
Note that when using .js docs,
map functions need to accept
emit function as the second parameter.
Instead of populating views immediately,
divan maps all documents for a view on first
This means that you can have as many designs as you want,
if you only use a few the rest won't eat up memory.
Also, reduce results are only computed and cached for the ranges of a view that you actually access. Once warmed up, the caches are invalidated and rebuilt very quickly on writes and deletes, because divan caches intermediate reduce results.
Brief, if you want to fully warm up a reduce view,
query it with
(depending on whether you'll ever use ungrouped results),
without specifying a key-range.
db.addView("view-name", ["source-view", "other-source-view"], viewObj)you can perform chained map/reduce, although its not optimized and is really slow right now.