pytest-spark 0.8.0

pytest plugin to run the tests with support of pyspark.

pytest plugin to run the tests with support of pyspark (Apache Spark).

This plugin will allow to specify SPARK_HOME directory in pytest.ini and thus to make “pyspark” importable in your tests which are executed by pytest.

You can also define “spark_options” in pytest.ini to customize pyspark, including “spark.jars.packages” option which allows to load external libraries (e.g. “com.databricks:spark-xml”).

pytest-spark provides session scope fixtures spark_context and spark_session which can be used in your tests.

Note: no need to define SPARK_HOME if you’ve installed pyspark using pip (e.g. pip install pyspark) - it should be already importable. In this case just don’t define SPARK_HOME neither in pytest (pytest.ini / --spark_home) nor as environment variable.

Install

$ pip install pytest-spark

Usage

Set Spark location

To run tests with required spark_home location you need to define it by using one of the following methods:

1. Specify command line option “–spark_home”:

   $ pytest --spark_home=/opt/spark

2. Add “spark_home” value to pytest.ini in your project directory:

   [pytest]
   spark_home = /opt/spark

3. Set the “SPARK_HOME” environment variable.

pytest-spark will try to import pyspark from provided location.

Note

“spark_home” will be read in the specified order. i.e. you can override pytest.ini value by command line option.

Customize spark_options

Just define “spark_options” in your pytest.ini, e.g.:

[pytest]
spark_home = /opt/spark
spark_options =
    spark.app.name: my-pytest-spark-tests
    spark.executor.instances: 1
    spark.jars.packages: com.databricks:spark-xml_2.12:0.5.0

Using the spark_context fixture

Use fixture spark_context in your tests as a regular pyspark fixture. SparkContext instance will be created once and reused for the whole test session.

Example:

def test_my_case(spark_context):
    test_rdd = spark_context.parallelize([1, 2, 3, 4])
    # ...

Warning: spark_context isn’t supported with Spark Connect functionality!

Using the spark_session fixture (Spark 2.0 and above)

Use fixture spark_session in your tests as a regular pyspark fixture. A SparkSession instance with Hive support enabled will be created once and reused for the whole test session.

Example:

def test_spark_session_dataframe(spark_session):
    test_df = spark_session.createDataFrame([[1,3],[2,4]], "a: int, b: int")
    # ...

Overriding default parameters of the spark_session fixture

By default spark_session will be loaded with the following configurations :

Example:

{
    'spark.app.name': 'pytest-spark',
    'spark.default.parallelism': 1,
    'spark.dynamicAllocation.enabled': 'false',
    'spark.executor.cores': 1,
    'spark.executor.instances': 1,
    'spark.io.compression.codec': 'lz4',
    'spark.rdd.compress': 'false',
    'spark.sql.shuffle.partitions': 1,
    'spark.shuffle.compress': 'false',
    'spark.sql.catalogImplementation': 'hive',
}

You can override some of these parameters in your pytest.ini. For example, removing Hive Support for the spark session :

Example:

[pytest]
spark_home = /opt/spark
spark_options =
    spark.sql.catalogImplementation: in-memory

Using spark_session fixture with Spark Connect

pytest-spark also works with Spark Connect that allows to execute code on the remote servers. You need Spark 3.4+ with pyspark installed with the connect extension (pyspark[connect] for PySpark 3.4+), or install pyspark-connect package (for PySpark 4.x).

It could be enabled with one of the following options:

• by setting SPARK_REMOTE environment variable to the URL of Spark Connect server.

• specifying URL of Spark Connect server as spark_connect_url option in pytest.ini.

• with --spark_connect_url command-line argument.

Note: in this mode, some of the Spark configurations will be ignored, such as, spark.executor.cores, spark.executor.instances, etc. that doesn’t have an effect on the existing Spark Session.

Development

Tests

Run tests locally:

$ docker-compose up --build