TRANSFORM

    Spark’s script transform supports two modes:

    1. Hive support disabled: Spark script transform can run with spark.sql.catalogImplementation=in-memory or without SparkSession.builder.enableHiveSupport(). In this case, now Spark only uses the script transform with ROW FORMAT DELIMITED and treats all values passed to the script as strings.
    2. Hive support enabled: When Spark is run with spark.sql.catalogImplementation=hive or Spark SQL is started with SparkSession.builder.enableHiveSupport(), Spark can use the script transform with both Hive SerDe and ROW FORMAT DELIMITED.

    Syntax

    • expression

      Specifies a combination of one or more values, operators and SQL functions that results in a value.

    • row_format

    • RECORDWRITER

      Specifies a fully-qualified class name of a custom RecordWriter. The default value is org.apache.hadoop.hive.ql.exec.TextRecordWriter.

    • RECORDREADER

      Specifies a fully-qualified class name of a custom RecordReader. The default value is org.apache.hadoop.hive.ql.exec.TextRecordReader.

    ROW FORMAT DELIMITED BEHAVIOR

    When Spark uses ROW FORMAT DELIMITED format:

    • Spark uses the character \u0001 as the default field delimiter and this delimiter can be overridden by FIELDS TERMINATED BY.
    • Spark uses the character \n as the default line delimiter and this delimiter can be overridden by LINES TERMINATED BY.
    • Spark uses a string \N as the default NULL value in order to differentiate NULL values from the literal string NULL. This delimiter can be overridden by NULL DEFINED AS.
    • Spark casts all columns to STRING and combines columns by tabs before feeding to the user script. For complex types such as ARRAY/MAP/STRUCT, Spark uses casts it to an input JSON string and uses from_json to convert the result output JSON string to ARRAY/MAP/STRUCT data.
    • COLLECTION ITEMS TERMINATED BY and MAP KEYS TERMINATED BY are delimiters to split complex data such as ARRAY/MAP/STRUCT, Spark uses to_json and from_json to handle complex data types with JSON format. So COLLECTION ITEMS TERMINATED BY and MAP KEYS TERMINATED BY won’t work in default row format.
    • The standard output of the user script is treated as tab-separated STRING columns. Any cell containing only a string \N is re-interpreted as a literal NULL value, and then the resulting STRING column will be cast to the data types specified in col_type.

    • If the actual number of output columns is less than the number of specified output columns, additional output columns will be filled with NULL. For example:

    • If the actual number of output columns is more than the number of specified output columns, the output columns only select the corresponding columns, and the remaining part will be discarded. For example, if the output has three tabs and there are only two output columns:

    When Hive support is enabled and Hive SerDe mode is used:

    • Spark uses the Hive SerDe org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe by default, so columns are cast to STRING and combined by tabs before feeding to the user script.
    • All literal NULL values are converted to a string \N in order to differentiate literal NULL values from the literal string NULL.
    • The standard output of the user script is treated as tab-separated STRING columns, any cell containing only a string \N is re-interpreted as a NULL value, and then the resulting STRING column will be cast to the data type specified in col_type.
    • If the actual number of output columns is less than the number of specified output columns, additional output columns will be filled with NULL.
    • If there is no AS clause after USING my_script, the output schema is key: STRING, value: STRING. The key column contains all the characters before the first tab and the value column contains the remaining characters after the first tab. If there is no tab, Spark returns the NULL value.
    • These defaults can be overridden with or ROW FORMAT DELIMITED.

    Examples