pytest-spark 0.8.0

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.

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.