github twitter linkedin
MongoDB and Spark Input
Jul 14, 2014

Say your writing a Spark application and you want to pull in data from MongoDB. There are a couple of ways to accomplish this task.

Directly from MongoDB

To read directly from MongoDB, create a new org.apache.hadoop.conf.Configuration with (at least) the parameter mongo.job.input.format (set to MongoInputFormat). Then use your SparkContext to create a new RDD from a Hadoop-backed file using the newAPIHadoopFile(...) method:

Configuration inputDataConfig = new Configuration();
inputDataConfig.set("mongo.job.input.format", "MongoInputFormat.class");
JavaPairRDD<Object,BSONObject> inputData = 
        MongoInputFormat.class, Object.class, BSONObject.class, 

The first argument to newAPIHadoopFile is the path to the data to be read. This should be a valid MongoDB connection string including the database and collection name (e.g.

Reading BSON Files

This step assumes you have already used mongodump to dump the contents of your database as a series of BSON files, one per collection, and stored them somewhere accessible (S3, HDFS, etc.).

Like before, create a new org.apache.hadoop.conf.Configuration with mongo.job.input.format set to BSONFileInputFormat. You’ll use the same function newAPIHadoopFile(...) as before but this time the first argument should be the full path to your BSON file. You’ll need to read each one individually into it’s own RDD.

Configuration bsonDataConfig = new Configuration();
bsonDataConfig.set("mongo.job.input.format", "BSONFileInputFormat.class");
JavaPairRDD<Object,BSONObject> bsonData = 
        BSONFileInputFormat.class, Object.class, BSONObject.class, 

Update: here’s a link to my mongodb-spark-demo repo if you want to see an example in action.

Back to posts