Skip to content

docs: lead installation guide with Docker/Apptainer over from-source#908

Open
sevmag wants to merge 1 commit into
graphnet-team:mainfrom
sevmag:docs/docker-first
Open

docs: lead installation guide with Docker/Apptainer over from-source#908
sevmag wants to merge 1 commit into
graphnet-team:mainfrom
sevmag:docs/docker-first

Conversation

@sevmag

@sevmag sevmag commented Jul 6, 2026

Copy link
Copy Markdown
Collaborator

Closes #889.

Reorganises docs/source/installation/install.rst so the recommended path — the pre-built Docker/Apptainer images — comes first, with from-source install framed as the option for writable/editable installs or systems without a container runtime. No technical content changes: just ordering and framing.

What changed

  • Quick Start now opens with a short pointer to the Docker/Apptainer images as the recommended path, with a cross-reference to Installing From Source_ for writable/editable installs.
  • Docker & Apptainer Images moved to the top, framed as the default. Added a note that Apptainer is the standard path on HPC clusters.
  • Installing From Source follows, with a lead-in explaining when to use it (development/editable installs, or no container runtime).
  • The IceTray subsection now points up to the IceTray+GraphNeT image as the preferred route and explains why (bundles a known-good IceTray + GraphNeT combination, avoids version-pinning issues); the CVMFS recipe is retained as a fully-supported alternative.
  • Minor whitespace cleanup (trailing spaces, missing final newline).

Motivation

HPC users often can't install from source on a bare cluster — e.g. RHEL/Rocky 8 nodes ship glibc 2.28, while pyg_lib's libpyg.so in the PyPI wheels requires GLIBC_2.29, so pip install -e .[torch-XX] outside CVMFS/conda-forge tends to fail at import time. The pre-built images sidestep this entirely and are autobuilt per release for both CPU and GPU. Leading with them better matches what's actively maintained.

🤖 Generated with Claude Code

@sevmag

sevmag commented Jul 8, 2026

Copy link
Copy Markdown
Collaborator Author

Feel free to add suggestions that change the wording or text now that we are already reworking this section @Aske-Rosted @christianlocatelli

which opens an image with a CPU-installation of GraphNeT 1.8.0 + PyTorch v2.6.0 + IceTray v1.13.0 installed and ready to use. You can replace the image path with the one you want to use from the DockerHub.


Installing From Source

@christianlocatelli christianlocatelli Jul 10, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe here you could emphasize that this option is the recommended one for contributors.
"Installing From Source (for development, or if Docker/Apptainer isn't available)"

===========
Here we provide a quick start guide for getting you started with |graphnet|\ GraphNeT.

The easiest way to get started is with our pre-built Docker images — see `Docker & Apptainer Images`_ below. They ship GraphNeT, PyTorch and (optionally) IceTray ready to use, with both CPU and GPU variants, and can be run without sudo via Apptainer. If you instead need a writable install or an environment without a container runtime, see `Installing From Source`_.

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would adapt this section a bit. Instead of "easiest way to get started", I would suggest to instead use something like: "The most reliable way for GraphNeT Users to get a working environment...".

I would also already mention in that section, that the Docker option is not available for KM3NeT so far. Otherwise KM3NeT Users reading it from top to bottom could be a bit mislead. Could we add a one-line qualifier in the intro, e.g. "Docker images are currently available for IceTray-based experiments (IceCube & P-ONE); KM3NeT users should install from source for now"?


The easiest way to get started is with our pre-built Docker images — see `Docker & Apptainer Images`_ below. They ship GraphNeT, PyTorch and (optionally) IceTray ready to use, with both CPU and GPU variants, and can be run without sudo via Apptainer. If you instead need a writable install or an environment without a container runtime, see `Installing From Source`_.

Docker & Apptainer Images

@christianlocatelli christianlocatelli Jul 10, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't know how familiar people in the community are with Docker/Apptainer. I personally never used containers before and would therefore have to look some things up before installing it.
In the Source installation section there is a link leading people to the Anaconda installation, so I thought it could be useful to provide at least a some links to the get started pages of Docker and Apptainer somewhere in this section. This is just a minor change but it leads ppl who are unfamiliar immediately to the most relevant pages.

Smth like:
(New to containers? See the get-started guides for Docker and Apptainer.)

@christianlocatelli christianlocatelli left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi Severin, I left some comments on your PR. Let me know what you think :)

christianlocatelli

This comment was marked as duplicate.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

docs: lead installation guide with Docker/Apptainer over from-source

2 participants