Spark scala 模拟 Spark.implicits 用于单元测试

当尝试使用 Spark 和 Scala 简化单元测试时，我使用 scala-test 和mockito-scala（以及mockito Sugar）。这只是让你做这样的事情：

val sparkSessionMock = mock[SparkSession]

然后你通常可以用“何时”和“验证”来完成所有的魔法。

但是如果你有一些需要导入的实现

import spark.implicits._

在其代码中，那么单元测试的简单性似乎消失了（或者至少我还没有找到解决这个问题的最合适的方法）。

我最终得到这个错误：

org.mockito.exceptions.verification.SmartNullPointerException: 
You have a NullPointerException here:
-> at ...
because this method call was *not* stubbed correctly:
-> at scala.Option.orElse(Option.scala:289)
sparkSession.implicits();

由于输入问题，简单地模拟 SparkSession 内“隐式”对象的调用不会有帮助：

val implicitsMock = mock[SQLImplicits]
when(sparkSessionMock.implicits).thenReturn(implicitsMock)

不会让你通过，因为它说它需要你的模拟中对象的类型：

require: sparkSessionMock.implicits.type
found: implicitsMock.type

请不要告诉我我应该做 SparkSession.builder.getOrCreate()...从那时起，这不再是单元测试，而是更重量级的集成测试。

（编辑）：这是一个完整的可重现示例：

import org.apache.spark.sql._
import org.mockito.Mockito.when
import org.scalatest.{ FlatSpec, Matchers }
import org.scalatestplus.mockito.MockitoSugar

case class MyData(key: String, value: String)

class ClassToTest()(implicit spark: SparkSession) {
    import spark.implicits._

    def read(path: String): Dataset[MyData] = 
         spark.read.parquet(path).as[MyData]
}

class SparkMock extends FlatSpec with Matchers with MockitoSugar {

     it should "be able to mock spark.implicits" in {
         implicit val sparkMock: SparkSession = mock[SparkSession]
         val implicitsMock = mock[SQLImplicits]
         when(sparkMock.implicits).thenReturn(implicitsMock)
         val readerMock = mock[DataFrameReader]
         when(sparkMock.read).thenReturn(readerMock)
         val dataFrameMock = mock[DataFrame]
         when(readerMock.parquet("/some/path")).thenReturn(dataFrameMock)
         val dataSetMock = mock[Dataset[MyData]]
         implicit val testEncoder: Encoder[MyData] = Encoders.product[MyData]
         when(dataFrameMock.as[MyData]).thenReturn(dataSetMock)

         new ClassToTest().read("/some/path/") shouldBe dataSetMock
    }
 }

你不能嘲笑隐式。隐式在编译时解决，而模拟在运行时发生（运行时反射，通过字节码操作）字节好友 https://bytebuddy.net）。您不能在编译时导入仅在运行时才会被模拟的隐式。您必须手动解析隐式（原则上，如果您在运行时再次启动编译器，则可以在运行时解析隐式，但这会困难得多1 https://stackoverflow.com/questions/64308467/is-there-anyway-in-scala-to-get-the-singleton-type-of-something-from-the-more 2 https://stackoverflow.com/questions/64045433/load-dataset-from-dynamically-generated-case-class 3 https://stackoverflow.com/questions/59348301/in-scala-2-or-3-is-it-possible-to-debug-implicit-resolution-process-in-runtime 4 https://stackoverflow.com/questions/64111895/dynamically-parse-json-in-flink-map/).

Try

class ClassToTest()(implicit spark: SparkSession, encoder: Encoder[MyData]) {
  def read(path: String): Dataset[MyData] = 
    spark.read.parquet(path).as[MyData]
}

class SparkMock extends AnyFlatSpec with Matchers with MockitoSugar {

  it should "be able to mock spark.implicits" in {
    implicit val sparkMock: SparkSession = mock[SparkSession]
    val readerMock = mock[DataFrameReader]
    when(sparkMock.read).thenReturn(readerMock)
    val dataFrameMock = mock[DataFrame]
    when(readerMock.parquet("/some/path")).thenReturn(dataFrameMock)
    val dataSetMock = mock[Dataset[MyData]]
    implicit val testEncoder: Encoder[MyData] = Encoders.product[MyData]
    when(dataFrameMock.as[MyData]).thenReturn(dataSetMock)

    new ClassToTest().read("/some/path") shouldBe dataSetMock
  }
}

//[info] SparkMock:
//[info] - should be able to mock spark.implicits
//[info] Run completed in 2 seconds, 727 milliseconds.
//[info] Total number of tests run: 1
//[info] Suites: completed 1, aborted 0
//[info] Tests: succeeded 1, failed 0, canceled 0, ignored 0, pending 0
//[info] All tests passed.

请注意"/some/path"两个地方应该是一样的。在您的代码片段中，两个字符串不同。