Machine Learning & Big Data Blog

Using Hive Advanced User Defined Functions with Generic and Complex Data Types

2 minute read
Walker Rowe

Previously we wrote how to write user defined functions that can be called from Hive. You can write these in Java or Scala. (Python does not work for UDFs per se. Instead you can use those with the Hive TRANSFORM operation.)

Programs that extend org.apache.hadoop.hive.ql.exec.UDF are for primitive data types, i.e., int, string. Etc. If you want to process complex types you need to use org.apache.hadoop.hive.ql.udf.generic.GenericUDF. Complex types are array, map, struct, and uniontype.

Generic functions extend org.apache.hadoop.hive.ql.udf.generic.GenericUDF and implement the 4 interfaces shown below.

class MapUpper extends GenericUDF {override def initialize(args: Array[ObjectInspector]): ObjectInspector = {
}override def getDisplayString(arg0: Array[String] ) : String = { return "silly me"; }override def evaluate(args: Array[DeferredObject]): Object = {}

This is the same as the simple UDF code, except there are two additional functions: initialize and getDisplay. Those set up an ObjectInspector and display a message if there is an error.

initialize looks at the value passed from Hive SQL to the function. There you check the argument count and type. Then it determines the type of argument that was passed to it. Then the evaluate function uses that typeless-argument, which is contained in org.apache.hadoop.hive.ql.udf.generic.GenericUDF.DeferredObject.

As you can see from the Scala code above, that returns an object of type Object, meaning there is no type definition and no ability for the compiler to find errors (Thus will show up at runtime.).

The initialize function returns the type of argument expected by the evaluate function.

Create Some Hive Map Data

We do not write a complete code example here. Instead we explain how you would set up to write a GenericUDF with a Map data type and give the general code outline above.

First, we create some data of Hive Map type. Run Hive and then execute:


This creates a table with one column: a map.

That will then let you parse a line of text line this:


Then you can load that into Hive like this:

load data inpath '/home/walker/Documents/hive/students.txt' into table students;

Which will then produce this output:

select * from students;

As you can see, each column is a (key->value) map.

Note that you can only load data into a Map column type using something like that. The Hive documentation makes clear that you cannot add values to a Map using SQL:

“Hive does not support literals for complex types (array, map, struct, union), so it is not possible to use them in INSERT INTO…VALUES clauses. This means that the user cannot insert data into a complex datatype column using the INSERT INTO…VALUES clause.”

Run Program in Hive

The way you run a program like this in Hive is to make these Hive jar file available to Hive by setting the classpath to:

export CLASSPATH=/usr/local/hive/apache-hive-2.3.0-bin/lib/hive-exec-2.3.0.jar:/usr/hadoop/hadoop-2.8.1/share/hadoop/mapreduce/hadoop-mapreduce-client-core-2.8.1.jar:/home/walker/Documents/bmc/hadoop-common-2.8.1.jar

All of those are contained in Hive and Hadoop lib folders, except for hadoop-common, which you download from Maven Central.

Add Jar to Hive

After you have written and compiled your program you put it in a jar file. Then in Hive you make it available using, where MapUpper is the name of the example we use here:

add jar /home/walker/Documents/bmc/udf/target/scala-2.12/mapupper_2.12-1.0.jar;create temporary function MapUpper as 'MapUpper';

Then you can run this command to execute the MapUpper function against the student column in the students table.

select MapUpper(student) from students;

This will run some operation on the keys or values and return a new map. Or it could return a primitive type if that is what you need.

Learn ML with our free downloadable guide

This e-book teaches machine learning in the simplest way possible. This book is for managers, programmers, directors – and anyone else who wants to learn machine learning. We start with very basic stats and algebra and build upon that.

These postings are my own and do not necessarily represent BMC's position, strategies, or opinion.

See an error or have a suggestion? Please let us know by emailing

Business, Faster than Humanly Possible

BMC works with 86% of the Forbes Global 50 and customers and partners around the world to create their future. With our history of innovation, industry-leading automation, operations, and service management solutions, combined with unmatched flexibility, we help organizations free up time and space to become an Autonomous Digital Enterprise that conquers the opportunities ahead.
Learn more about BMC ›

About the author

Walker Rowe

Walker Rowe is an American freelancer tech writer and programmer living in Cyprus. He writes tutorials on analytics and big data and specializes in documenting SDKs and APIs. He is the founder of the Hypatia Academy Cyprus, an online school to teach secondary school children programming. You can find Walker here and here.