ComfyUI container fails to start: pip missing from venv prevents requirements.txt installation

[Daemonrat]

Environment

• Harbor version: Latest (as of November 2025)
• Host OS: Arch Linux
• Service: ComfyUI
• GPU: Tesla P40 (24GB VRAM)

Problem Summary

The ComfyUI container enters a crash loop on startup with ModuleNotFoundError: No module named 'av'. The issue is NOT that the dependency is missing from requirements.txt, but that pip is not available in the Python virtual environment at /opt/environments/python/comfyui, preventing automatic installation of requirements during container startup.

Error Logs

harbor.comfyui  | [ComfyUI-Manager] Neither python -m pip nor uv are available. Cannot proceed with package operations.
harbor.comfyui  | Failed to execute startup-script: /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/prestartup_script.py / Neither pip nor uv are available for package management
harbor.comfyui  | 
harbor.comfyui  | Traceback (most recent call last):
harbor.comfyui  |   File "/workspace/ComfyUI/comfy_api/latest/_input/video_types.py", line 5, in <module>
harbor.comfyui  |     import av
harbor.comfyui  | ModuleNotFoundError: No module named 'av'

Reproduction Steps

1. harbor up comfyui
2. Container crashes immediately
3. harbor logs comfyui shows the above errors
4. Container enters FATAL state after multiple restart attempts

Investigation & Findings

Verified the requirements.txt includes av

av>=14.1.0

is present in /workspace/ComfyUI/requirements.txt (confirmed by inspecting the file inside the container).

Manual installation works

When I manually fix the issue inside the container:

harbor exec comfyui bash
/opt/environments/python/comfyui/bin/python -m ensurepip --upgrade
/opt/environments/python/comfyui/bin/pip install -r /workspace/ComfyUI/requirements.txt
cd /workspace/ComfyUI
/opt/environments/python/comfyui/bin/python main.py

ComfyUI starts successfully without errors.

The issue does NOT persist

After exiting and running harbor restart comfyui, the container crashes again with the same error. This confirms that pip installations are not persisting because pip itself is missing from the venv during container initialization.

Root Cause

The virtual environment at /opt/environments/python/comfyui was created without pip installed (likely using python -m venv --without-pip or a minimal venv tool like uv). This causes:

1. ComfyUI's requirements.txt installation to fail silently or not run at all
2. ComfyUI-Manager's prestartup script to fail with the "Neither python -m pip nor uv are available" error
3. The av module (and potentially other dependencies) to never get installed

Proposed Solution

The Harbor ComfyUI container image needs to ensure pip is installed in the venv during image build or container initialization. Add to the container's entrypoint or Dockerfile:

/opt/environments/python/comfyui/bin/python -m ensurepip --upgrade

This should run before ComfyUI's main.py is executed, ensuring that:

1. pip is available for requirements.txt installation
2. ComfyUI-Manager's prestartup scripts can function properly
3. All dependencies (including the recently-added av>=14.1.0 for video support) are installed automatically

Workaround

Users can temporarily fix this by execing into the container and running:

/opt/environments/python/comfyui/bin/python -m ensurepip --upgrade
/opt/environments/python/comfyui/bin/pip install -r /workspace/ComfyUI/requirements.txt

However, this must be repeated after every container restart.

Additional Context

This issue likely affects recent ComfyUI updates that added av>=14.1.0 as a core dependency for video processing functionality (see Comfy-Org/ComfyUI#9999 and #7869). Older ComfyUI versions may not have triggered this issue if they had fewer dependencies.

[Daemonrat]

One of the pre-requesites here is that you have to update to the latest version of Comfyui from the "Manager Menu"

[av]

Hi, thank you for using Harbor and for the report,

I just reproduced the issue - when updating via Manager, ComfyUI initially works as expected, but after restart fails with the error you mentioned

[av]

Ok, after some tests, It's actually an older problem with how comfyui updates are setup in the ai-dock containers. To correctly persist them, they must be synced to a volume/host location after changes in the packages in the container. In Harbor, that's implemented via:

harbor comfyui workspace sync

I tested that on a fresh install, after update and sync - restart restores to the same state, without import errors.

I've emphasised this procedure in the Harbor's ComfyUI Wiki a bit more. Ideally, we might want to migrate to some other ComfyUI pre-built container, but I'm not aware of any good candidates since I've been out of touch with its ecosystem since the initial implementation in Harbor (please hit me up if you know any good ones).

[realnikolaj]

Ok, after some tests, It's actually an older problem with how comfyui updates are setup in the ai-dock containers. To correctly persist them, they must be synced to a volume/host location after changes in the packages in the container. In Harbor, that's implemented via:

harbor comfyui workspace sync

I tested that on a fresh install, after update and sync - restart restores to the same state, without import errors.

I've emphasised this procedure in the Harbor's ComfyUI Wiki a bit more. Ideally, we might want to migrate to some other ComfyUI pre-built container, but I'm not aware of any good candidates since I've been out of touch with its ecosystem since the initial implementation in Harbor (please hit me up if you know any good ones).

I've replaced the included ConfyUI with mmartial/ComfyUI-Nvidia-Docker and now that the latest release supports UV with USE_UV env variable I cannot recommend the project enough.

It's really well documented, proposes A ton of options and is a near 1-1 compose migration. I can include my current harbor config and compose if relevant since I've spent many hours testing things and now that uv makes the whole thing more reliable, and, much faster startup I finally feel comfortable recommending it for this project.