How to flatten a struct in a Spark DataFrame?

How to convert DataFrame fields into separate columns.

Bartosz Mikulski 02 Oct 2020 – 1 min read

This article will show you how to extract the struct field and convert them into separate columns in a Spark DataFrame.

Let’s assume that I have the following DataFrame, and the to_be_flattened column contains a struct with two fields:

+-------------------+
|    to_be_flattened|
+-------------------+
|  [1183, Amsterdam]|
|    [06123, Ankara]|
| [08067, Barcelona]|
|       [3030, Bern]|
|     [75116, Paris]|
| [1149-014, Lisbon]|
|   [00-999, Warsaw]|
|      [00199, Rome]|
|[HR-10 040, Zagreb]|
+-------------------+

Extracting those fields into columns is trivial, and we need only this line of code to achieve it:

df.select(col('to_be_flattened.*'))

As a result, we get this DataFrame:

+-----------+---------+
|postal_code|     city|
+-----------+---------+
|       1183|Amsterdam|
|      06123|   Ankara|
|      08067|Barcelona|
|       3030|     Bern|
|      75116|    Paris|
|   1149-014|   Lisbon|
|     00-999|   Warsaw|
|      00199|     Rome|
|  HR-10 040|   Zagreb|
+-----------+---------+

We have lost the original column name. What if I wanted to prefix the extracted columns with its previous name, and instead of postal_code and city have columns to_be_flattened_postal_code and to_be_flattened_city?

We can do it by getting the field names from the struct schema, iterating over them, and adding the prefix to every field:

df.select(col('to_be_flattened.*')) \
    .select([col(c).alias('to_be_flattened_' + c) for c in struct_schema.fieldNames()])

+---------------------------+--------------------+
|to_be_flattened_postal_code|to_be_flattened_city|
+---------------------------+--------------------+
|                       1183|           Amsterdam|
|                      06123|              Ankara|
|                      08067|           Barcelona|
|                       3030|                Bern|
|                      75116|               Paris|
|                   1149-014|              Lisbon|
|                     00-999|              Warsaw|
|                      00199|                Rome|
|                  HR-10 040|              Zagreb|
+---------------------------+--------------------+

How to flatten a struct in a Spark DataFrame?

What is the difference between CUBE and ROLLUP and how to use it in Apache Spark?

When to cache an Apache Spark DataFrame?

How to flatten a struct in a Spark DataFrame?

What is the difference between CUBE and ROLLUP and how to use it in Apache Spark?

When to cache an Apache Spark DataFrame?

Related Posts

What is shuffling in Apache Spark, and when does it happen?

How to measure Spark performance and gather metrics about written data

Testing data products: BDD for data engineers