I have a delta table which has thrift data from kafka and I am using a UDF to deserialize it. I have no issues when I use regular UDF, but I get an error when I try to use Pandas UDF.
This runs fine i.e. ruglar UDF
def decoder(thrift_data): schema_file = thriftpy2.load("/dbfs/FileStore/schema_file.thrift") schema = schema_file.SchemaClass() decoded_payload = deserialize(schema, thrift_data, TCyBinaryProtocolFactory()) json_data = proto.struct_to_json(decoded_payload) return json.dumps(json_data) decoder_udf = udf(decoder, StringType()) data = spark.sql("""SELECT value FROM data_table""") data = data.withColumn('decoded_json', decoder_udf(data.value))
But when I use Pandas UDF
def decoder(thrift_data: pd.Series) -> pd.Series: schema_file = thriftpy2.load("/dbfs/FileStore/schema_file.thrift") schema = schema_file.SchemaClass() decoded_payload = deserialize(schema, thrift_data, TCyBinaryProtocolFactory()) json_data = proto.struct_to_json(decoded_payload) return json.dumps(json_data) decoder_udf = pandas_udf(decoder, returnType=StringType()) data = spark.sql("""SELECT value FROM data_table""") data = data.withColumn('decoded_json', decoder_udf(data.value))
I get an error PythonException: 'RuntimeError: Result vector from pandas_udf was not the required length: expected 5000, got 651'.
Advertisement
Answer
Figured out the solution, we have to return the output as a series