Running roop-unleashed in a docker container

[Rooper-ai]

Describe the bug
Trying to install roop-unleashed in a docker container with some success but with a few weird errors along the way

To Reproduce
I'm running docker on linux and using an Ubuntu 22.04 based container from [ghcr.io/linuxserver/webtop:ubuntu-kde] and passing in the GPU with:

--gpus=all
--device /dev/dri:/dev/dri

Then install all the nvidia dependencies as per:

Install nvidia repos

sudo apt update; sudo apt -y upgrade
sudo apt install -y --no-install-recommends gnupg2 curl ca-certificates
curl -fsSLO https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb

Update repos and in case keyring is out of date

sudo apt update; sudo apt -y upgrade

Install cuda dependencias

sudo apt install cuda-toolkit-11-8 cudnn9-cuda-11

Add envs to bash

nano ~/.bashrc

cut and paste into .bashrc

export PATH=$PATH:~/.local/bin
export PATH=$PATH:/usr/local/nvidia/bin:/usr/local/cuda/bin
export LD_LIBRARY_PATH=/usr/local/cuda-11/lib64:/usr/local/cuda-11.8/lib64

Restart container

Install roop

git clone https://github.com/C0untFloyd/roop-unleashed
cd roop-unleashed
pip install -r requirements.txt

Install correct runtime

pip uninstall onnxruntime onnxruntime-gpu
pip install onnxruntime-gpu==1.15.1

I can then run up roop with python3 run.py. I get a weird error which ends with [Aborted (core dumped)] but in fact the roop service IS running in the background:

*** ERROR ***
QObject::moveToThread: Current thread (0x570d7f3ab330) is not the object's thread (0x570d7f3dd610).
Cannot move to target thread (0x570d7f3ab330)

qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "/config/.local/lib/python3.10/site-packages/cv2/qt/plugins" even though it was found.
This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem.

Available platform plugins are: xcb, eglfs, linuxfb, minimal, minimalegl, offscreen, vnc.

Aborted (core dumped)
*** ERROR ***

I also get the following warning when I use roop, although doesn't appear to affect anything (??)

*** ERROR ***
(run.py:812): dbind-WARNING **: 16:08:19.741: AT-SPI: Error retrieving accessibility bus address: org.freedesktop.DBus.Error.ServiceUnknown: The name org.a11y.Bus was not provided by any .service files
*** ERROR ***

On researching the qt plugin error I discovered the following suggestion:

#fix xcb qt plugin issue (doesn't work)
pip uninstall opencv-python
pip install opencv-python-headless

However installing both packages does better:

#fix xcb qt plugin issue (works)
pip uninstall opencv-python
pip uninstall opencv-python-headless
pip install opencv-python
pip install opencv-python-headless

I then get a permissions issue on a dot file:

*** ERROR ***
QStandardPaths: wrong permissions on runtime directory /config/.XDG, 0755 instead of 0700
*** ERROR ***

Which is easily fixed with:

chmod 700 ~/.XDG

Hopefully the above info is useful for anyone else planning to run roop in a container, but welcome to any suggestions if I'm doing something daft, and why I'm getting the weird errors...

Thanks

Details
What OS are you using?

• [YES] Linux [host=ubuntu 22.04 container=ubuntu 22.04]
• Linux in WSL
• Windows
• Mac

Are you using a GPU?

• No. CPU FTW
• [Yes] NVIDIA [Host=RTX3060,Driver=535.171,Cuda=12.2 container=cuda=11.8
• AMD
• Intel
• Mac

Which version of roop unleashed are you using?

3.6.7

Screenshots
If applicable, add screenshots to help explain your problem.

[C0untFloyd]

Thanks for your work on a docker image, I'd be happy to include a docker install script once this is successful. I'm pretty sure, the app doesn't use Qt anywhere but opencv headless sounds like a logical choice anyway. I believe you can get a lot of help and inspiration out of this repo because it's a also a fork of roop.

[OAtulA]

Hey did you get it done? Any success or updates?

[vidgen13]

Here is my dockerfile and it works. Only problem is I cannot bind the output volume because it shows that the device is busy.

FROM nvidia/cuda:12.1.0-cudnn8-runtime-ubuntu22.04
RUN apt-get update && \
         apt-get install -y \
           ffmpeg git pip python3 python-is-python3
RUN git clone https://github.com/C0untFloyd/roop-unleashed
WORKDIR /roop-unleashed
RUN pip install -r requirements.txt && \
         pip uninstall -y onnxruntime-gpu && \
         pip install onnxruntime-gpu==1.15.1
ENV GRADIO_SERVER_NAME="0.0.0.0"
EXPOSE 7860
ENTRYPOINT ["python", "run.py"]"

[C0untFloyd]

docker build -t roop-unleashed .

https://github.com/C0untFloyd/roop-unleashed/blob/main/Dockerfile

[vidgen13]

i did see that but I think I had problems with the cuda driver.

…

On Tue, Aug 20, 2024, 16:58 C0untFloyd ***@***.***> wrote: docker build -t roop-unleashed . https://github.com/C0untFloyd/roop-unleashed/blob/main/Dockerfile — Reply to this email directly, view it on GitHub <#626 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/APQXP6GKRVTISXFDOYVD5E3ZSLSJLAVCNFSM6AAAAABKGAGLGOVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTAMZZGE4TQMI> . You are receiving this because you commented.Message ID: ***@***.*** com>

[vidgen13]

Ok, I used that but I had to add the nvidia toolkit so it uses my GPU, the first one used nvidia as the base and just added python. Thanks. Can you tell me why I can't bind the output directory?

FROM python:3.11
RUN curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey |\
    gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg &&\
    curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list |\
    sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' |\
    tee /etc/apt/sources.list.d/nvidia-container-toolkit.list
RUN apt-get update &&\
    apt-get install -y \
      ffmpeg \
      git \
      nvidia-container-toolkit
RUN git clone https://github.com/C0untFloyd/roop-unleashed
WORKDIR /roop-unleashed
RUN pip install -r requirements.txt &&\
    pip uninstall -y onnxruntime-gpu &&\
    pip install onnxruntime-gpu==1.15.1
ENV GRADIO_SERVER_NAME="0.0.0.0"
ENV NVIDIA_VISIBLE_DEVICES=all
ENV NVIDIA_DRIVER_CAPABILITIES=all
EXPOSE 7860
ENTRYPOINT ["python", "run.py"]

[C0untFloyd]

Sorry I'm not using Docker myself, the code was donated by pull request. Anyway, what exactly is your issue? The files can't be written into the output folder or did I misunderstand you?

[pyromate]

Can you also share (or do you not use) the Compose file?
like:
` ports:

• 7880:7860
  deploy:
  resources:
  reservations:
  devices:
  • driver: nvidia
    capabilities: [ gpu, video ]`

Because for me your container doesn't work , still get the CPU. (also when in settings I select CUDA as provider:
Using provider ['CPUExecutionProvider'] - Device:cpu Running on local URL: http://0.0.0.0:7860

[vidgen13]

when I map the volume like so: volumes: - somefolder:/output it throws off an error that the device is busy. not a big deal because I can download anyway instead of opening the folder. thanks for the reply.

…

On Thu, Aug 22, 2024, 18:40 C0untFloyd ***@***.***> wrote: Sorry I'm not using Docker myself, the code was donated by pull request. Anyway, what exactly is your issue? The files can't be written into the output folder or did I misunderstand you? — Reply to this email directly, view it on GitHub <#626 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/APQXP6E5HNRCEOQFZIGXXGTZSWPWRAVCNFSM6AAAAABKGAGLGOVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTANBRGY3DGOI> . You are receiving this because you commented.Message ID: ***@***.*** com>

[pyromate]

Have the same issue, but i assume the folder gets deleted upon a new job. But that can't be done because it's mounted by docker. have you tried to mount the root instead of /output ?