Introduction to Spark SQL functions

Spark SQL functions make it easy to perform DataFrame analyses.

This post will show you how to use the built-in Spark SQL functions and how to build your own SQL functions.

Make sure to read Writing Beautiful Spark Code for a detailed overview of how to use SQL functions in production applications.

Review of common functions

The Spark SQL functions are stored in the org.apache.spark.sql.functions object.

The documentation page lists all of the built-in SQL functions.

Let’s create a DataFrame with a number column and use the factorial function to append a number_factorial column.

import org.apache.spark.sql.functions._

val df = Seq(2, 3, 4).toDF("number")

df
  .withColumn("number_factorial", factorial(col("number")))
  .show()
+------+----------------+
|number|number_factorial|
+------+----------------+
|     2|               2|
|     3|               6|
|     4|              24|
+------+----------------+

The factorial() function takes a single Column argument. The col() function, also defined in the org.apache.spark.sql.functions object, returns a Column object based on the column name.

If Spark implicits are imported (i.e. import spark.implicits._), then you can also create a Column object with the $ operator. This code also works.

import org.apache.spark.sql.functions._
import spark.implicits._

val df = Seq(2, 3, 4).toDF("number")

df
  .withColumn("number_factorial", factorial($"number"))
  .show()

lit() function

The lit() function creates a Column object out of a literal value. Let’s create a DataFrame and use the lit() function to append a spanish_hi column to the DataFrame.

val df = Seq("sophia", "sol", "perro").toDF("word")

df
  .withColumn("spanish_hi", lit("hola"))
  .show()
+------+----------+
|  word|spanish_hi|
+------+----------+
|sophia|      hola|
|   sol|      hola|
| perro|      hola|
+------+----------+

The lit() function is especially useful when making boolean comparisons.

when() and otherwise() functions

The when() and otherwise() functions are used for control flow in Spark SQL, similar to if and else in other programming languages.

Let’s create a DataFrame of countries and use some when() statements to append a country column.

val df = Seq("china", "canada", "italy", "tralfamadore").toDF("word")

df
  .withColumn(
    "continent",
    when(col("word") === lit("china"), lit("asia"))
      .when(col("word") === lit("canada"), lit("north america"))
      .when(col("word") === lit("italy"), lit("europe"))
      .otherwise("not sure")
  )
  .show()
+------------+-------------+
|        word|    continent|
+------------+-------------+
|       china|         asia|
|      canada|north america|
|       italy|       europe|
|tralfamadore|     not sure|
+------------+-------------+

Spark lets you cut the lit() method calls sometimes and to express code compactly.

df
  .withColumn(
    "continent",
    when(col("word") === "china", "asia")
      .when(col("word") === "canada", "north america")
      .when(col("word") === "italy", "europe")
      .otherwise("not sure")
  )
  .show()

Here’s another example of using when() to manage control flow.

val df = Seq(10, 15, 25).toDF("age")

df
  .withColumn(
    "life_stage",
    when(col("age") < 13, "child")
      .when(col("age") >= 13 && col("age") <= 18, "teenager")
      .when(col("age") > 18, "adult")
  )
  .show()
+---+----------+
|age|life_stage|
+---+----------+
| 10|     child|
| 15|  teenager|
| 25|     adult|
+---+----------+

Writing your own SQL function

You can use the built-in Spark SQL functions to build your own SQL functions. Let’s create a lifeStage() function that takes an age as an argument and returns child, teenager or adult.

import org.apache.spark.sql.Column

def lifeStage(col: Column): Column = {
  when(col < 13, "child")
    .when(col >= 13 && col <= 18, "teenager")
    .when(col > 18, "adult")
}

Let’s use the lifeStage() function in a code snippet.

val df = Seq(10, 15, 25).toDF("age")

df
  .withColumn(
    "life_stage",
    lifeStage(col("age"))
  )
  .show()
+---+----------+
|age|life_stage|
+---+----------+
| 10|     child|
| 15|  teenager|
| 25|     adult|
+---+----------+

Let’s create a trimUpper() function that trims all whitespace and capitalizes all of the characters in a string.

import org.apache.spark.sql.Column

def trimUpper(col: Column): Column = {
  trim(upper(col))
}

Let’s run trimUpper() on a sample data set.

val df = Seq(
  "   some stuff",
  "like CHEESE     "
).toDF("weird")

df
  .withColumn(
    "cleaned",
    trimUpper(col("weird"))
  )
  .show()
+----------------+-----------+
|           weird|    cleaned|
+----------------+-----------+
|      some stuff| SOME STUFF|
|like CHEESE     |LIKE CHEESE|
+----------------+-----------+

Custom SQL functions can typically be used instead of UDFs. Avoiding UDFs is a great way to write better Spark code as described in this post.

Testing SQL functions

You can inspect the SQL that’s generated by a SQL function with the toString method.

lifeStage(lit("10")).toString
CASE
  WHEN (10 < 13) THEN child
  WHEN ((10 >= 13) AND (10 <= 18)) THEN teenager
  WHEN (10 > 18) THEN adult
END

In our test suite, we can make sure that the SQL string that’s generated equals what’s expected.

val expected = "CASE WHEN (10 < 13) THEN child WHEN ((10 >= 13) AND (10 <= 18)) THEN teenager WHEN (10 > 18) THEN adult END"
lifeStage(lit("10")).toString == expected

We can also create a DataFrame, append a column with the lifeStage() function, and use the spark-fast-tests library to compare DataFrame equality.

Next steps

Spark SQL functions are preferable to UDFs because they handle the null case gracefully (without a lot of code) and because they are not a black box.

Most Spark analyses can be run by leveraging the standard library and reverting to custom SQL functions when necessary. Avoid UDFs at all costs!

Registration

1 Comment


  1. i don’t understand map and reduce action in RDD.

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *