Awesome

MongoDB Search

Experiment with Full-Text-Search using Node.js and MongoDB

WHY

Finding relevance in an ocean of content.

Having spent the last couple of days wrestling with Meteor + Search trying to get it let me search through a collection I've decided to take a different approach to the probrem.

Hopefully by the end of this exercise we will be able to search through a collection of "posts" (tweets) and find relevant results.

WHAT

Full-text search without having to manage Solr or ElasticSearch. (keeping it simple with just one data store)

HOW

Ensure You have MongoDB Installed

see: https://www.mongodb.com/docs/manual/installation/

Going Native

In the past I've used Mongoose to interact with MongoDB. Mongoose offers many great abstractions when building applications specifically around pre-defining models and providing constructors to validate fields on insert/update. We don't need that here.

All we need is the ability to:

Connect to MongoDB (Local)
Insert records into a collection
Create a text index for the text field in the collection
Execute a text (search) query against the records in the collection
Return a list of all the record (_ids) that match the search criteria
Store the results in a "results" collection to speed up future searches

For these simple tasks the Node MongoDB Native client is perfect.

https://github.com/mongodb/node-mongodb-native

Install the node-mongodb-native node module using NPM:

npm install mongodb

Startup The Meteor Mongo DB

Use this command to start Mongo with textSearchEnabled:

mongod --bind_ip 127.0.0.1 --dbpath ~/code/meteor-search/.meteor/local/db --setParameter textSearchEnabled=true

Create the Index

MongoClient.connect('mongodb://127.0.0.1:27017/meteor', function(err, db) {
  if(err) throw err;
  var posts = db.collection('posts');
  posts.indexInformation(function(err, info) { // all indexes on posts collection
    console.dir(info);
  });
}) // end MongoClient

Output:

{ 
  _id_: [ [ '_id', 1 ] ],
  post_search_index: [ [ '_fts', 'text' ], [ '_ftsx', 1 ] ] 
}

ignore the id index that's a mongodb default. the one which interests us is the post_search_index we created earlier.

https://mongodb.github.io/node-mongodb-native/markdown-docs/indexes.html

Searching the Data

db.posts.find({}).sort({time:-1}).limit(100);

Node.js MongoDB Native does not have runCommand which is used in most full-text search examples. <br />So we cannot just do:

db.posts.runCommand( "text", { search: "justin" } )

MongoDB Native NO runCommand

But a bit of investigation yields:

// unintuitively the text field is actually the collection!
db.command({text:"posts" , search: "maths science" }, function(err, cb){ 
	console.log(cb.results);
});

Storing Search Results

The result of the above db.command search query has the format:

  { score: 2.142857142857143,
    obj: 
     { text: 'Math, science, history, unraveling the mystery it all started with a #BigBang💥',
       time: Sun Mar 30 2014 07:03:08 GMT+0100 (BST),
       avatar: 'https://pbs.twimg.com/profile_images/442935363095379968/CuEcmYsH_normal.jpeg',
       _id: 'Kxssadbi2e5X7ga5L' } },
 { score: 2.142857142857143,
    obj: 
     { text: 'I was just about to set my maths and science books on fire…#ihateschool',
       time: Sun Mar 30 2014 06:22:31 GMT+0100 (BST),
       avatar: 'https://pbs.twimg.com/profile_images/449715822454243329/cNN69E3A_normal.jpeg',
       _id: 'ghoi72BoEfswZgfws' } }

This returns the score (a float) and the entire record (all fields). We could return these results directly to the user and we are done. But going back to our original reason for diving into "naitve" node, we want to be able to "pipe" these results back into our Meteor app.

New Collection: Search Results

db.search_results.insert(
	{
		keywords:"science",
		posts : [ 
				{
					score: 2.142857142857143,
					_id: 'Kxssadbi2e5X7ga5L'
				},
				{
					score: 2.142857142857143,
					_id: 'ghoi72BoEfswZgfws' 
				}
			]
		last_updated: new Date()
	}
);