Test Your Pipeline

Testing your pipeline is a particularly important step in developing an effective data processing solution. The indirect nature of the Beam model, in which your user code constructs a pipeline graph to be executed remotely, can make debugging failed runs a non-trivial task. Often it is faster and simpler to perform local unit testing on your pipeline code than to debug a pipeline’s remote execution.

Before running your pipeline on the runner of your choice, unit testing your pipeline code locally is often the best way to identify and fix bugs in your pipeline code. Unit testing your pipeline locally also allows you to use your familiar/favorite local debugging tools.

You can use DirectRunner, or PrismRunner. Both are local runners helpful for testing and local development.

After you test your pipeline locally, you can use the runner of your choice to test on a small scale. For example, use the Flink runner with a local or remote Flink cluster.

The Beam SDKs provide a number of ways to unit test your pipeline code, from the lowest to the highest levels. From the lowest to the highest level, these are:

To support unit testing, the Beam SDK for Java provides a number of test classes in the testing package. You can use these tests as references and guides.

Testing Transforms

To test a transform you’ve created, you can use the following pattern:

TestPipeline

TestPipeline is a class included in the Beam Java SDK specifically for testing transforms.

TestPipeline is a class included in the Beam Python SDK specifically for testing transforms.

For tests, use TestPipeline in place of Pipeline when you create the pipeline object. Unlike Pipeline.create, TestPipeline.create handles setting PipelineOptions internally.

You create a TestPipeline as follows:

Pipeline p = TestPipeline.create();
with TestPipeline as p:
    ...
import "github.com/apache/beam/sdks/v2/go/pkg/beam/testing/ptest"

// Override TestMain with ptest.Main,
// once per package.
func TestMain(m *testing.M) {
	ptest.Main(m)
}

func TestPipeline(t *testing.T) {
	...
	// The Go SDK doesn't use a TestPipeline concept,
	// and recommends using the ptest harness
	// to wrap pipeline construction.
	pr := ptest.BuildAndRun(t, func(s beam.Scope) {
		...
	})
	...
}

Note: Read about testing unbounded pipelines in Beam in this blog post.

Using the Create Transform

You can use the Create transform to create a PCollection out of a standard in-memory collection class, such as Java or Python List. See Creating a PCollection for more information.

PAssert

PAssert is a class included in the Beam Java SDK that is an assertion on the contents of a PCollection. You can use PAssertto verify that a PCollection contains a specific set of expected elements.

For a given PCollection, you can use PAssert to verify the contents as follows:

PCollection<String> output = ...;

// Check whether a PCollection contains some elements in any order.
PAssert.that(output)
.containsInAnyOrder(
  "elem1",
  "elem3",
  "elem2");
from apache_beam.testing.util import assert_that
from apache_beam.testing.util import equal_to

output = ...

# Check whether a PCollection contains some elements in any order.
assert_that(
    output,
    equal_to(["elem1", "elem3", "elem2"]))
import "github.com/apache/beam/sdks/v2/go/pkg/beam/testing/passert"

output := ... // beam.PCollection

// Check whether a PCollection contains some elements in any order.
passert.EqualsList(s, output, ["elem1", "elem3", "elem2"])

Any Java code that uses PAssert must link in JUnit and Hamcrest. If you’re using Maven, you can link in Hamcrest by adding the following dependency to your project’s pom.xml file:

<dependency>
    <groupId>org.hamcrest</groupId>
    <artifactId>hamcrest</artifactId>
    <version>2.2</version>
    <scope>test</scope>
</dependency>

For more information on how these classes work, see the org.apache.beam.sdk.testing package documentation.

An Example Test for a Composite Transform

The following code shows a complete test for a composite transform. The test applies the Count transform to an input PCollection of String elements. The test uses the Create transform to create the input PCollection from a List<String>.

public class CountTest {

  // Our static input data, which will make up the initial PCollection.
  static final String[] WORDS_ARRAY = new String[] {
  "hi", "there", "hi", "hi", "sue", "bob",
  "hi", "sue", "", "", "ZOW", "bob", ""};

  static final List<String> WORDS = Arrays.asList(WORDS_ARRAY);

  public void testCount() {
    // Create a test pipeline.
    Pipeline p = TestPipeline.create();

    // Create an input PCollection.
    PCollection<String> input = p.apply(Create.of(WORDS));

    // Apply the Count transform under test.
    PCollection<KV<String, Long>> output =
      input.apply(Count.<String>perElement());

    // Assert on the results.
    PAssert.that(output)
      .containsInAnyOrder(
          KV.of("hi", 4L),
          KV.of("there", 1L),
          KV.of("sue", 2L),
          KV.of("bob", 2L),
          KV.of("", 3L),
          KV.of("ZOW", 1L));

    // Run the pipeline.
    p.run();
  }
}
import unittest
import apache_beam as beam
from apache_beam.testing.test_pipeline import TestPipeline
from apache_beam.testing.util import assert_that
from apache_beam.testing.util import equal_to

class CountTest(unittest.TestCase):

  def test_count(self):
    # Our static input data, which will make up the initial PCollection.
    WORDS = [
      "hi", "there", "hi", "hi", "sue", "bob",
      "hi", "sue", "", "", "ZOW", "bob", ""
    ]
    # Create a test pipeline.
    with TestPipeline() as p:

      # Create an input PCollection.
      input = p | beam.Create(WORDS)

      # Apply the Count transform under test.
      output = input | beam.combiners.Count.PerElement()

      # Assert on the results.
      assert_that(
        output,
        equal_to([
            ("hi", 4),
            ("there", 1),
            ("sue", 2),
            ("bob", 2),
            ("", 3),
            ("ZOW", 1)]))

      # The pipeline will run and verify the results.
import (
	"testing"

	"github.com/apache/beam/sdks/v2/go/pkg/beam"
	"github.com/apache/beam/sdks/v2/go/pkg/beam/register"
	"github.com/apache/beam/sdks/v2/go/pkg/beam/testing/passert"
	"github.com/apache/beam/sdks/v2/go/pkg/beam/testing/ptest"
	"github.com/apache/beam/sdks/v2/go/pkg/beam/transforms/stats"
)

// formatFn takes a key value pair and puts them
// into a single string for comparison.
func formatFn(w string, c int) string {
	return fmt.Sprintf("%s: %d", w, c)
}

// Register the functional DoFn to ensure execution on workers.
func init() {
	register.Function2x1(formatFn)
}

func TestCountWords(t *testing.T) {
	// The pipeline will run and verify the results.
	ptest.BuildAndRun(t, func(s beam.Scope) {
		words := []string{"hi", "there", "hi", "hi", "sue", "bob",
			"hi", "sue", "", "", "ZOW", "bob", ""}

		wantCounts := []string{"hi: 5", "there: 1", "sue: 2", "bob: 2"}

		// Create a PCollection from the words static input data.
		input := beam.CreateList(s, words)

		// Apply the Count transform under test.
		output := stats.Count(s, col)
		formatted := beam.ParDo(s, formatFn, output)

		// Assert that the output PCollection matches the wantCounts data.
		passert.Equals(s, formatted, wantCounts...)
	})
}

Testing a Pipeline End-to-End

You can use the test classes in the Beam SDKs (such as TestPipeline and PAssert in the Beam SDK for Java) to test an entire pipeline end-to-end. Typically, to test an entire pipeline, you do the following:

Testing the WordCount Pipeline

The following example code shows how one might test the WordCount example pipeline. WordCount usually reads lines from a text file for input data; instead, the test creates a List<String> containing some text lines and uses a Create transform to create an initial PCollection.

WordCount’s final transform (from the composite transform CountWords) produces a PCollection<String> of formatted word counts suitable for printing. Rather than write that PCollection to an output text file, our test pipeline uses PAssert to verify that the elements of the PCollection match those of a static String array containing our expected output data.

public class WordCountTest {

    @Rule public final transient TestPipeline p = TestPipeline.create();

    // Our static input data, which will comprise the initial PCollection.
    static final String[] WORDS_ARRAY = new String[] {
      "hi there", "hi", "hi sue bob",
      "hi sue", "", "bob hi"};

    static final List<String> WORDS = Arrays.asList(WORDS_ARRAY);

    // Our static output data, which is the expected data that the final PCollection must match.
    static final String[] COUNTS_ARRAY = new String[] {
        "hi: 5", "there: 1", "sue: 2", "bob: 2"};

    // Example test that tests the pipeline's transforms.

    public void testCountWords() throws Exception {
      // Create a PCollection from the WORDS static input data.
      PCollection<String> input = p.apply(Create.of(WORDS));

      // Run ALL the pipeline's transforms (in this case, the CountWords composite transform).
      PCollection<String> output = input.apply(new CountWords());

      // Assert that the output PCollection matches the COUNTS_ARRAY known static output data.
      PAssert.that(output).containsInAnyOrder(COUNTS_ARRAY);

      // Run the pipeline.
      p.run();
    }
}
import unittest
import apache_beam as beam
from apache_beam.testing.test_pipeline import TestPipeline
from apache_beam.testing.util import assert_that
from apache_beam.testing.util import equal_to

class CountWords(beam.PTransform):
    # CountWords transform omitted for conciseness.
    # Full transform can be found here - https://github.com/apache/beam/blob/master/sdks/python/apache_beam/examples/wordcount_debugging.py

class WordCountTest(unittest.TestCase):

  # Our input data, which will make up the initial PCollection.
  WORDS = [
      "hi", "there", "hi", "hi", "sue", "bob",
      "hi", "sue", "", "", "ZOW", "bob", ""
  ]

  # Our output data, which is the expected data that the final PCollection must match.
  EXPECTED_COUNTS = ["hi: 5", "there: 1", "sue: 2", "bob: 2"]

  # Example test that tests the pipeline's transforms.

  def test_count_words(self):
    with TestPipeline() as p:

      # Create a PCollection from the WORDS static input data.
      input = p | beam.Create(WORDS)

      # Run ALL the pipeline's transforms (in this case, the CountWords composite transform).
      output = input | CountWords()

      # Assert that the output PCollection matches the EXPECTED_COUNTS data.
      assert_that(output, equal_to(EXPECTED_COUNTS), label='CheckOutput')

    # The pipeline will run and verify the results.
package wordcount

import (
	"testing"

	"github.com/apache/beam/sdks/v2/go/pkg/beam"
	"github.com/apache/beam/sdks/v2/go/pkg/beam/testing/passert"
	"github.com/apache/beam/sdks/v2/go/pkg/beam/testing/ptest"
)

// CountWords and formatFn are omitted for conciseness.
// Code for the Full transforms can be found here:
// https://github.com/apache/beam/blob/master/sdks/go/examples/debugging_wordcount/debugging_wordcount.go

func CountWords(s beam.Scope, lines beam.PCollection) beam.PCollection { ... }

func formatFn(w string, c int) string { ... }

func TestCountWords(t *testing.T) {
	// The pipeline will run and verify the results.
	ptest.BuildAndRun(t, func(s beam.Scope) {
		words := []string{"hi", "there", "hi", "hi", "sue", "bob",
			"hi", "sue", "", "", "ZOW", "bob", ""}

		wantCounts := []string{"hi: 5", "there: 1", "sue: 2", "bob: 2"}

		// Create a PCollection from the words static input data.
		input := beam.CreateList(s, words)

		// Run ALL the pipeline's transforms
		// (in this case, the CountWords composite transform).
		output := CountWords(s, input)
		formatted := beam.ParDo(s, formatFn, output)

		// Assert that the output PCollection matches
		// the wantCounts data.
		passert.Equals(s, formatted, wantCounts...)
	})
}