Update: Got it working by using the *_corrupt_record *field for the first case (record 4)
schema = schema.add("_corrupt_record", DataTypes.StringType); Dataset<Row> ds = spark.read().schema(schema).option("mode", "PERMISSIVE").json("path").collect(); ds = ds.filter(functions.col("_corrupt_record").isNull()).collect(); However, I haven't figured out on how to ignore record 5. Any help is appreciated. On Mon, 3 Jul 2023 at 19:24, Shashank Rao <shashank93...@gmail.com> wrote: > Hi all, > I'm trying to read around 1,000,000 JSONL files present in S3 using Spark. > Once read, I need to write them to BigQuery. > I have a schema that may not be an exact match with all the records. > How can I filter records where there isn't an exact schema match: > > Eg: if my records were: > {"x": 1, "y": 1} > {"x": 2, "y": 2} > {"x": 3, "y": 3} > {"x": 4, "y": "4"} > {"x": 5, "y": 5, "z": 5} > > and if my schema were: > root > |-- x: long (nullable = true) > |-- y: long (nullable = true) > > I need the records 4 and 5 to be filtered out. > Record 4 should be filtered out since y is a string instead of long. > Record 5 should be filtered out since z is not part of the schema. > > I tried applying my schema on read, but it does not work as needed: > > StructType schema = new StructType().add("a", DataTypes.LongType).add("b", > DataTypes.LongType); > Dataset<Row> ds = spark.read().schema(schema).json("path/to/file") > > This gives me a dataset that has record 4 with y=null and record 5 with x > and y. > > Any help is appreciated. > > -- > Thanks, > Shashank Rao > -- Regards, Shashank Rao