A Spark session is a connection object. It's the application entry point and is used to create a dataset (ie therefore also a dataframe).
Previously, it was known as the SQLContext but it has been deprecated.