Version Matrix

spark-test-assertions

A single assertion to compare Spark DataFrames and Datasets (future work).

Basic usage

libraryDependencies += "com.github.ffmmjj" % "spark-test-assertions_2.11" % "<version>"

and then

import com.github.ffmmjj.spark.assertions.DataFrameAssertions._

and then, in your test,

actualDataframe shouldHaveSameContentsAs expectedDataframe

Goals

  1. Minimal set of dependencies (only Spark itself as a provided dependency);
  2. Readable error messages (clearly pointing differences instead of just dumping the expected and actual DataFrames to the output);
  3. Flexible and readable assertion constraints (allowing, for example, to ignore column order in the compared DataFrames);

Examples

it should "raise an exception detailing the missing fields if the expected dataframe has columns that don't exist in the actual dataframe" in {
  val actual = Seq("value1").toDF("field1")
  val expected = Seq(("value1", "value2", "value3")).toDF("field1", "field2", "field3")

  actual shouldHaveSameContentsAs expected
}

Will output an exception with message

assertion failed: Dataframe [field1: string] doesn't have column(s) [field2, field3]
it should "raise an exception detailing the extra fields if the actual dataframe has columns that dont't exist in the expected dataframe" in {
  val actual = Seq(("value1", "value2", "value3")).toDF("field1", "field2", "field3")
  val expected = Seq("value1").toDF("field1")

  actual shouldHaveSameContentsAs expected
}

Will output an exception with message

assertion failed: Dataframe [field1: string, field2: string ... 1 more field] contains extra columns [field2, field3]
it should "raise an exception if the number of rows in the actual dataframe is different than in the expected dataframe" in {
  val actual = Seq(
    ("value1", 7.68910)
  ).toDF("field1", "field2")
  val expected = Seq(
    ("value1", 7.68910),
    ("value2", 15.161718)
  ).toDF("field1", "field2")

  actual shouldHaveSameContentsAs expected

Will output an exception with message

The number of rows in the actual dataframe is different than in the expected dataframe.
Expected: 2
Actual: 1
it should "raise an exception if the columns in the actual and expected dataframes follow a different order" in {
  val actual = Seq(("value1", "value2")).toDF("field1", "field2")
  val expected = Seq(("value2", "value1")).toDF("field2", "field1")
    
  actual shouldHaveSameContentsAs expected
}

Will output an exception with message

assertion failed: DataFrame [field1: string, field2: string] has the same columns as [field2: string, field1: string], but in a different order - do you really care about column order in this test?

It's possible, however, to add the withAnyColumnOrdering qualifier to prevent column ordering from being checked like actual shouldHaveSameContentsAs (expected withAnyColumnOrdering) (parenthesis required) or as an explicit argument like actual.shouldHaveSameContentsAs(expected, withAnyColumnOrdering=true)

it should "raise an exception if the columns are the same but the values differ in some of the dataframe lines" in {
  val actual = Seq(
    ("value1", "value2", "value3"),
    ("value4", "value5", "value6")
  ).toDF("field1", "field2", "field3")
  val expected = Seq(
    ("value1", "value7", "value8"),
    ("value9", "value5", "value6")
  ).toDF("field1", "field2", "field3")

  actual shouldHaveSameContentsAs expected
}

Will output an exception with message

assertion failed: Different values found in some lines.
Mismatched values in actual DataFrame:
+----+------+------+------+
|line|field1|field2|field3|
+----+------+------+------+
|   0|  null|value2|value3|
|   1|value4|  null|  null|
+----+------+------+------+
Mismatched values in expected DataFrame:
+----+------+------+------+
|line|field1|field2|field3|
+----+------+------+------+
|   0|  null|value7|value8|
|   1|value9|  null|  null|
+----+------+------+------+