pytest-mock-unity-catalog 0.0.4

Unity Catalog pyspark fixtures

pytest-mock-unity-catalog

Pytest plugin that provides PySpark fixtures for testing code that reads and writes Unity Catalog tables — without a live Databricks cluster. Table operations are redirected to a local Delta directory so tests run fully offline.

Installation

pip install pytest-mock-unity-catalog

Pytest discovers the plugin automatically via its entry point. No imports or conftest.py changes are needed in the consuming project.

Fixtures

spark

A session-scoped SparkSession configured for local testing with Delta Lake enabled.

def test_something(spark):
    df = spark.createDataFrame([(1, "a")], ["id", "value"])
    assert df.count() == 1

By default uses delta-spark_4.1_2.13:4.1.0 (PySpark 4.1, Scala 2.13). Override via the SPARK_VERSION environment variable for other versions:

# PySpark 3.5 / Scala 2.12
SPARK_VERSION=2.12:3.2.1 pytest

# PySpark 4.0 / Scala 2.13
SPARK_VERSION=4.0_2.13:4.0.0 pytest

mock_save_as_table

Patches DataFrame.write.saveAsTable to write a Delta table to a local temp directory instead of Unity Catalog. The Unity Catalog-style three-part name (catalog.schema.table) is mapped to a directory path.

def test_write(spark, mock_save_as_table):
    df = spark.createDataFrame([(1, "a")], ["id", "value"])
    df.write.saveAsTable("my_catalog.my_schema.my_table")  # writes locally

mock_read_table

Patches both spark.read.table and spark.table to read from the same local Delta path that mock_save_as_table writes to. Use both fixtures together to round-trip through a table.

def test_read(spark, mock_read_table):
    df = spark.read.table("my_catalog.my_schema.my_table")
    assert df.count() == 2

    df2 = spark.table("my_catalog.my_schema.my_table")
    assert df2.count() == 2

local_table_base_path

The Path to the session-scoped temp directory used as the root for all table storage. Useful for asserting on the filesystem directly or for sharing the path in custom fixtures.

def test_path(local_table_base_path):
    assert (local_table_base_path / "my_catalog" / "my_schema" / "my_table").exists()

mock_volume

Redirects all /Volumes/... filesystem access to a local temp directory for the duration of the test. The fixture yields the local base Path so tests can seed files before exercising the code under test.

Intercepted access patterns:

Pattern	Mechanism
open("/Volumes/...")	patches builtins.open
open(Path("/Volumes/..."))	patches builtins.open via PathLike
Path("/Volumes/...").read_text()	patches Path.__fspath__
Path("/Volumes/...").write_text(...)	patches Path.__fspath__
Path("/Volumes/...").exists() / .stat() / .mkdir()	patches Path.__fspath__
pd.read_csv("/Volumes/...")	pandas delegates to open()
pd.DataFrame.to_csv("/Volumes/...")	pandas delegates to open()

Limitation: binary/columnar readers that bypass Python's open() — e.g. pandas.read_parquet backed by pyarrow — are not intercepted.

Parent directories under the temp root are created automatically, so no explicit mkdir is needed before writing.

def test_read_volume(mock_volume):
    # Seed a file at the equivalent of /Volumes/cat/schema/vol/data.csv
    seed = mock_volume / "cat" / "schema" / "vol" / "data.csv"
    seed.parent.mkdir(parents=True, exist_ok=True)
    seed.write_text("id,value\n1,a\n2,b\n")

    # Code under test uses the real /Volumes path — it is transparently redirected
    import pandas as pd
    df = pd.read_csv("/Volumes/cat/schema/vol/data.csv")
    assert len(df) == 2

Works with pathlib.Path too:

def test_write_volume(mock_volume):
    from pathlib import Path

    Path("/Volumes/cat/schema/vol/out.txt").write_text("hello")

    result = Path("/Volumes/cat/schema/vol/out.txt").read_text()
    assert result == "hello"

volume_base_path

The session-scoped Path used as the root for all volume storage. Injected automatically into mock_volume; only needed directly when building custom fixtures on top of the volume base.

Example: full round-trip

def test_round_trip(spark, mock_save_as_table, mock_read_table):
    df = spark.createDataFrame([(1, "a"), (2, "b")], ["id", "value"])
    df.write.saveAsTable("my_catalog.my_schema.my_table")

    result = spark.read.table("my_catalog.my_schema.my_table")
    assert result.count() == 2

Databricks / on-cluster usage

When tests run inside a Databricks notebook or job (i.e. DATABRICKS_RUNTIME_VERSION is set), the plugin detects this automatically:

• spark returns the active SparkSession instead of creating a local one.
• mock_read_table is a no-op — spark.read.table hits Unity Catalog as normal.
• mock_save_as_table is a no-op — df.write.saveAsTable writes to Unity Catalog as normal. The table is dropped with DROP TABLE IF EXISTS in teardown.
• mock_volume is a no-op — /Volumes/... paths reach the real Unity Catalog volume.

No code changes are needed; the same tests run locally (mocked) and on Databricks (real).

How it works

mock_save_as_table and mock_read_table patch PySpark's DataFrameWriter.saveAsTable and DataFrameReader.table for the duration of the test. The Unity Catalog table name is converted to a filesystem path by replacing . separators with /:

my_catalog.my_schema.my_table  →  <tmp>/my_catalog/my_schema/my_table

The temp directory is managed by pytest (tmp_path_factory) and lives under the OS temp space (e.g. /var/folders/.../pytest-of-<user>/pytest-<N>/). Pytest retains the last three runs before pruning.