Deprecation of official Docker releases in favor of Python packages

[omri374]

Hello Presidio community,

We’re exploring a change to how we distribute Presidio and would love your input before making any moves. Our aim is to streamline releases and focus on our core Python packages while still giving container users everything they need to build and run Presidio on their own.

Why We’re Suggesting This Change

1. Sharper Focus on Core Functionality
   Maintaining both pre-built Docker images and Python packages adds CI/CD complexity and slows down our ability to deliver new features, improvements, and fixes. By concentrating on PyPI releases, we can iterate faster on the core product, which is the SDK.
2. Higher-Quality Python API & Libraries
   Less overhead means more engineering time for performance optimizations, richer documentation, and enhanced test coverage of the Python libraries - Presidio’s primary interface.
3. Customization Is at Our Core
   We know that many advanced use cases involve tweaking the Dockerfile (adjusting dependencies and configurations). Instead of publishing opaque images, we’d surface the Dockerfiles so you can build exactly the image you need for the platform you need.

What This Could Mean for You

• Docker images would remain available “as-is” on Docker Hub, but we would not issue any further updates including feature releases, bug fixes, or security patches.
• If you depend on updated containers, you can build your own from our maintained Dockerfiles and instructions in the repo.
• All new features, fixes, and security updates will be delivered via the Python packages on PyPI.

We’d Love Your Feedback

Before we move forward, please let us know:

1. How you currently use our Docker images:
   • Do you pull official tags directly, or build from source?
   • Which versions or base platforms matter most?
2. What you rely on in the existing images:
   • Specific entrypoint scripts, environment-variable setups, or bundled tools.
3. Any concerns or potential roadblocks:
   • Scenarios where updates (especially security patches) are critical and couldn't easily be handled via custom builds.

How to Share Feedback

Please let us know what you think by:

• Replying directly to this GitHub Discussion, or
• Emailing us at presidio@microsoft.com with details on how you’re using Docker today and any feedback on the proposed change.

Tentative Timeline

• Feedback period: Now through August 1st, 2025
• Decision on next steps: By August 14, 2025

We’re committed to supporting our community and want to make this transition, if it happens, as smooth and informed as possible.

Thanks for your continued support and partnership!

The Presidio Maintenance Team

[StefH]

A question:

1. Are the current Docker images created using the latest source-code in this repository?
2. Or are the Docker images using the official released and versioned Python packages?

[omri374]

Docker images are released to Dockerhub at the same time Python packages are released to PyPI

[StefH]

I mean: are the docker images build using the 'raw' source code from this repository. Or is the main app python file referring to python packages generated?

[omri374]

The Docker images are built using the raw source code from the repository, but the Python package and Docker image are always released together, so the code in both is identical.
We don't do continuous deployment, so there could be a gap between the source code in main and the code in a prebuilt Docker image.

[fosron]

I feel like having a Docker image is an easier starting point for developers from non-Python backgrounds. And with the nature of PII, the companies that would use it, tend to keep their tech stacks tight, so f.e for a PHP company creating a custom Python code to use a PII provider might be not feasible.

[elementhrobbie]

As a non-python developer, I do not want to dive into using and running python packages, where it is no problem to run another docker container. I feel docker containers are much more operations and multi programming language friendly than the very specific single language packages.

Try to imagine you wanted to use something like this yourself and the owner would like to switch from Docker to something language specific in a language you do not want to dive into and not want to learn how to operate.

Please keep distributing the docker version 🙏🙂

[bfisseler]

I am developing in R mainly. That said, Presidio could be a valuable tool in the context of data pseudonymization for research. And as a few other people have mentioned, having a Docker image makes things easier. I am currently trying to customize the Docker image for multi-language support but still without success. So having an official Docker image is of valuable help.

[SharonHart]

Thanks for the feedback everyone, we will skip the deprecation for now