You can also just read it as text file to memory RDD and than read that RDD as json. During reading as text file is easy to set partitioning. so it will be not one big object but for example 8:

rdd = sc.textFile(sourceFile, 8)
df = spark.read.json(rdd)

If there are problems with json text you can clean it using map:

rdd = sc.textFile(sourceFile, 8).map(your lambda function)
df = spark.read.json(rdd)