---
title: "How to use Virtualenv to prepare a separate environment for Python function running in Airflow"
description: "How to use the PythonVirtualenvOperator in Airflow"
author: "Bartosz Mikulski"
author_bio: "Principal AI Engineer & MLOps Architect. I bridge the gap between \"it works in a notebook\" and \"it works for 200 million users.\""
author_url: https://mikulskibartosz.name
author_linkedin: https://www.linkedin.com/in/mikulskibartosz/
author_github: https://github.com/mikulskibartosz
canonical_url: https://mikulskibartosz.name/use-virtualenv-to-run-python-functions-in-airflow
---

It is not difficult to turn your Python environment into a mess. Soon, the libraries become incompatible with one another, start producing weird results or suddenly crash in the middle of a computation.

Fortunately, we can create separate environments using Virtualenv or Conda. This feature is also available in Airflow, but in this case, we have access only to Virtualenv (unless you add a custom operator).

First, we have to define a Python function we want to run. Note that we must define ALL imports inside the function, and it cannot reference anything defined outside. Even if it is a global variable. We must pass all such variables as arguments of the `PythonVirtualenvOperator`.

```python
def some_python_function():
    import pandas as pd

    # do something with Pandas

    return "some value"
```

The returned value is available in the Airflow XCOM, and we can reference it in the subsequent tasks.

There is one issue concerning returned values (and input parameters). If the Python version used in the Virtualenv environment differs from the Python version used by Airflow, we cannot pass parameters and return values. In this case, we can use only the `string_args` parameter.

## Use PythonVirtualenvOperator

Now, I can configure the Airflow operator. I pass the required libraries as the `requirements` parameter. It supports the same syntax as the `requirements.txt` file, so I can also define a version:

```python
virtualenv_task = PythonVirtualenvOperator(
    task_id="virtualenv_pandas",
    python_callable=some_python_function,
    requirements=["pandas"],
    system_site_packages=False,
    dag=dag,
)
```