When we know precisely what query we should run to get the data we want from a SQL database, we don’t need to load multiple tables in PySpark, and emulate the joins and selects in the Python code. Instead of that, we can pass the SQL query as the source of the DataFrame while retrieving it from the database.

If my code to retrieve the data looks like this:

1
2
3
4
5
6
df = spark.read \
    .format("jdbc") \
    .option("url", "jdbc:mysql://localhost:port") \
    .option("dbtable", "schema.tablename") \
    ...
    .load()

I can replace the dbtable parameter with a SQL query and use the result as the table loaded by PySpark:

1
.option("dbtable", "(SELECT column_A, column_B FROM some_table) AS tbl")