divan

0.0.3 • Public • Published

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.

why? because!

  • some tasks and workloads just don't deserve their own Couch but can benefit from a similar data-model.
  • not a Couch - which among other things means it doesn't do MVCC, so delete and update as much as you want.
  • super fast - when fully warmed up serves thousands of queries per second, and there's no network latency.

usage

npm install divan

index.js

 
//  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 = divan.cwd ( 'friends' );
 
//  Save some docs, generating your own id-s:
 
db.save ({ _id : 'don1', type : 'person', name : 'Don', gender : 'male' });
db.save ({ _id : 'samantha', type : 'person', name : 'Sam', gender : 'female' });
db.save ({ _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:
 
db.addView
(
    'gender/count',
    divan.mr
    (
        function ( doc, emit )
        {
            if ( doc.type === 'person' )
                emit ( doc.gender, 1 );
        },
        function ( k, v )
        {
            var i, n = v.length, sum = 0;
            for ( i = 0; i < n; i ++ ) sum += v [ i ];
            return sum;
        }
    )
);
 
//  Query the view:
 
db.view ( 'gender/count', { group : true }, function ( err, data )
{
    data.rows.forEach ( function ( row )
    {
        console.log ( row.key, row.value );
    });
});
 
//  Outputs:
//  female 1
//  male 2
 

You can add views via db.addView or by parsing a directory of design files via db.design(path). The design-files can either be .json files of couchdb-design-doc flavour, or .js files that export objects with map and reduce methods. Note that when using .js docs, map functions need to accept the emit function as the second parameter.

CouchDB view API coverage

Everything but group_level and include_docs.

lazy views and reduce caching

Instead of populating views immediately, divan maps all documents for a view on first .query(). 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 group=true or group=false (depending on whether you'll ever use ungrouped results), without specifying a key-range.

what else

  • You can iterate your entire db with db.forEach(func)
  • If you look at the sources, you'll see that there's an option to have your snapshots on Amazon S3.
  • By using 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.

license

MIT

Readme

Keywords

none

Package Sidebar

Install

npm i divan

Weekly Downloads

5

Version

0.0.3

License

none

Last publish

Collaborators

  • hdachev