[4/6 Deps Update] Update docker images use uv #399

svij-sc · 2025-11-26T05:14:53Z

Background Context: #405
Some guidance on stack of PRs: #405 (comment)

Scope of PR

Generally updating docker images to use UV but ran into some issues while migrating to UV and addressed the issues as well:
- Since we are not using conda anymore migrated to using plain ubuntu images where possible
- New docker client version is not compatible w/ the docker api version the GCP cloud runner's docker deamon uses - so I locked the docker client version for new builder images
- Using pytorch image for GPU images as these are pre-optimized and have all drivers correctly preconfigured - issues if we try to set this up ourselves / makes our image build code complicated. Subsequently we make use of host python - i.e. python interpreter already installed on the image which uv can leverage.
- The older version of building Dataflow base image does not work anymore i.e. COPY --from=apache/beam_python3.9_sdk:2.53.0 /opt/apache/beam /opt/apache/beam with the updated python and sdk version creates runtime failures when running dataflow jobs. Thus, we are now just using apache/beam_python3.11_sdk:2.56.0 as base image instead of trying to copy from it.

containers/Dockerfile.builder

kmontemayor2-sc · 2025-11-26T17:33:52Z

containers/Dockerfile.cpu.base

    wget \
    cmake \
    iputils-ping \
+    curl \


qq why do we need curl here?

Used for downloading uv installer

GiGL/requirements/install_py_deps.sh

Line 62 in dbc161f

curl -LsSf -o uv_installer.sh https://astral.sh/uv/0.9.5/install.sh # Matches the version in .github/actions/setup-python-tools/action.yml

hmmm, this also means any dev boxes we have also need curl right? I think they have it by default but maybe they don't? Should the curl instatllation be elsewhere?

Should the curl instatllation be elsewhere?

Like in install_deps?

I feel like its a pretty common package id rather leave it where it is for now and not increate. the scope even more trying to centralize ubuntu package installations to some shared script.

I do believe it comes preinstalled in the dev boxes, and even if it doesnt I can argue for something like that maybe we should not be installing it for the user, similar to how we dont install docker, etc.

yliu2-sc · 2025-12-03T19:55:56Z

containers/Dockerfile.builder


-
+# Dec 1, 2025 (svij-sc):
+# GCP Cloud build agents run an older version of docker deamon


will we need to update docker version when gcp packages update? As we don't fix the gcp package versions and they auto resolve

this issue is diff than GCP packages.
This is more an issue w. docker deamon running on the GCP cloud build agent VMs* that run our CI/CD.

yliu2-sc · 2025-12-03T19:57:27Z

containers/Dockerfile.builder

+ENV PATH="${UV_PROJECT_ENVIRONMENT}/bin:${PATH}"
+# We also need to make UV detectable by the system
+ENV PATH="/root/.local/bin:${PATH}"
+RUN uv tool install pip==25.3


Do we need to fix pip version?

More so, why don't we lock the pip version we install with uv?

Good catch, I locked it in pyproject.toml; let me remove this from here and see what happens

kmontemayor2-sc · 2025-12-03T21:39:02Z

containers/Dockerfile.builder

+ENV PATH="${UV_PROJECT_ENVIRONMENT}/bin:${PATH}"
+# We also need to make UV detectable by the system
+ENV PATH="/root/.local/bin:${PATH}"
+RUN uv tool install pip==25.3


More so, why don't we lock the pip version we install with uv?

kmontemayor2-sc · 2025-12-03T21:39:48Z

containers/Dockerfile.cpu.base

    wget \
    cmake \
    iputils-ping \
+    curl \


hmmm, this also means any dev boxes we have also need curl right? I think they have it by default but maybe they don't? Should the curl instatllation be elsewhere?

kmontemayor2-sc · 2025-12-03T21:41:25Z

containers/Dockerfile.cuda.base

+# Already has python 3.11 installed - no need to install it again.
+# We use system python since it has packages pre-installed for us.
+ENV UV_SYSTEM_PYTHON=true
+ENV UV_PROJECT_ENVIRONMENT=/opt/conda/


Again I'm not sure we should be using system python (and we generally want to avoid conda?)

I assume we set this to use /opt/conda because the default torch images use that one?

I assume we set this to use /opt/conda because the default torch images use that one?

This is exactly it. Their python installation is using conda-forge

Re:

Again I'm not sure we should be using system python

Lets continue our conversation in the other thread https://github.com/Snapchat/GiGL/pull/398/files#r2565901566

Unrelated but do you think it's a bit confusing for us to have different python install dirs for different images?

note I also don't think we should be using /opt/conda if we are in fact no using conda - I guess it's fine for us to have the different install paths.

This is a prod docker image not a dev container.
If we dont use the default system python we need to then reinstall everything into some other environment and we lose on the benifits of having cuda libs/drivers/pytorch pre-installed installed for us.

Especially because we are not using conda to manage the rest of the deps.

Secondly, this also causes issues listed here: #398 (comment)

kmontemayor2-sc · 2025-12-05T18:06:18Z

containers/Dockerfile.cuda.base

+# Already has python 3.11 installed - no need to install it again.
+# We use system python since it has packages pre-installed for us.
+ENV UV_SYSTEM_PYTHON=true
+ENV UV_PROJECT_ENVIRONMENT=/opt/conda/


Unrelated but do you think it's a bit confusing for us to have different python install dirs for different images?

kmontemayor2-sc · 2025-12-05T18:08:13Z

containers/Dockerfile.cuda.base

+# Already has python 3.11 installed - no need to install it again.
+# We use system python since it has packages pre-installed for us.
+ENV UV_SYSTEM_PYTHON=true
+ENV UV_PROJECT_ENVIRONMENT=/opt/conda/


note I also don't think we should be using /opt/conda if we are in fact no using conda - I guess it's fine for us to have the different install paths.

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

update docker images

485638a

svij-sc changed the base branch from main to svij/install-uv November 26, 2025 05:16

svij-sc changed the title ~~Update docker images use uv~~ [4/6 Deps Update] Update docker images use uv Nov 26, 2025

svij-sc added 2 commits November 26, 2025 06:00

Merge branch 'svij/install-uv' into svij/update-docker-images-use-uv

a083f41

Merge branch 'svij/install-uv' into svij/update-docker-images-use-uv

b771633

kmontemayor2-sc reviewed Nov 26, 2025

View reviewed changes

svij-sc mentioned this pull request Dec 2, 2025

Update deps / python to latest supported version - use UV for dep management #405

Open

svij-sc added 4 commits December 2, 2025 23:22

Merge branch 'svij/install-uv' into svij/update-docker-images-use-uv

60a790c

update docker images

1a29c95

pr comments

65e2e81

Merge branch 'svij/install-uv' into svij/update-docker-images-use-uv

27f9abe

yliu2-sc approved these changes Dec 3, 2025

View reviewed changes

kmontemayor2-sc reviewed Dec 3, 2025

View reviewed changes

svij-sc added 2 commits December 5, 2025 02:49

pr comments

8115e26

Merge branch 'svij/install-uv' into svij/update-docker-images-use-uv

5d4984e

kmontemayor2-sc approved these changes Dec 5, 2025

View reviewed changes

This was referenced Dec 8, 2025

[3/6 Deps Update] Install UV + update dev usage to UV #398

Merged

[5/6 Deps Update] Update all workflows and actions to use UV #400

Merged

kmontemayor2-sc approved these changes Dec 8, 2025

View reviewed changes

[5/6 Deps Update] Update all workflows and actions to use UV (#400)

5952343

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

svij-sc marked this pull request as ready for review December 12, 2025 18:30

svij-sc requested review from mkolodner-sc, nshah-sc, xgao4-sc and zfan3-sc as code owners December 12, 2025 18:30

svij-sc merged commit d6a3dc4 into svij/install-uv Dec 12, 2025
6 checks passed

svij-sc deleted the svij/update-docker-images-use-uv branch December 12, 2025 18:31

svij-sc mentioned this pull request Dec 13, 2025

Update deps / python to latest supported version - use UV for dep management #414

Merged

10 tasks



		# Dec 1, 2025 (svij-sc):
		# GCP Cloud build agents run an older version of docker deamon

[4/6 Deps Update] Update docker images use uv #399

[4/6 Deps Update] Update docker images use uv #399

Uh oh!

Conversation

svij-sc commented Nov 26, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

svij-sc commented Nov 26, 2025 •

edited

Loading