Loading Data from MapR Database as an Apache Spark DataFrame

To load data from a MapR Database JSON table into an Apache Spark DataFrame, invoke the following API:
For loading as a DataFrame, apply the following method on a SparkSession object:
def loadFromMapRDB[T](tableName: String, 
              schema: StructType): DataFrame  
import com.mapr.db.spark.sql._

val df = sparkSession.loadFromMapRDB[T]("/tmp/user_profiles"): DataFrame  
For loading as a DataFrame (Datasets of Row), apply the following method on a MapRDBJavaSession object:
def loadFromMapRDB(tableName: String, schema: StructType, sampleSize: Double):DataFrame
import com.mapr.db.spark.sql.api.java.MapRDBJavaSession;
import org.apache.spark.sql.SparkSession;
MapRDBJavaSession maprSession = new MapRDBJavaSession(spark);
Note: Java supports only DataSets of Row (Dataset<Row>).
For loading as a DataFrame, apply the following method on a SparkSession object:
loadFromMapRDB(table_name, schema, sample_size)
from pyspark.sql import SparkSession
df = spark.loadFromMapRDB("/tmp/user_profiles")
Note: PySpark supports only DataFrames (Dataset<Row>).
Note: The only required parameter to the methods is tableName. All the others are optional.
This creates a DataFrame object corresponding to the MapR Database table specified by the tableName parameter.

Both DataFrames and MapR Database tables work with structured data. DataFrames need a fixed schema, whereas MapR Database allows for a flexible schema. When loading data into a DataFrame, you can map your data to a schema by specifying the schema parameter in the loadFromMapRDB call. You can also provide an application class as the type [T] parameter in the call. These two approaches are the preferred methods for loading data into DataFrames.

For data exploration use cases, you might not know the schema of your MapR Database table. For those situations, the MapR Database OJAI connector for Apache Spark can infer the schema by sampling data from the table.

Whenever possible, the MapR Database OJAI Connector for Apache Spark pushes projections and filters for better performance. This allows MapR Database to project and filter data before returning it to your client application.

The following subtopics describe these techniques.