sleap-webRTC

Description

This repo contains three DockerFiles for lightweight containers (~8 GB) corresponding to the Worker, SLEAP Worker, and Signaling Server. The container repository is located at https://hub.docker.com/repository/docker/eberrigan/sleap-cuda/general.

The base image used is python-3.9-slim for the Worker. For the SLEAP Worker, the base image is ghcr.io/talmolab/sleap-cuda:linux-amd64-2a8cefd7d0291d2ce07998fa3c54c9d0d618d31b

• The Worker Dockerfile is located at ./webRTC_worker_container/Dockerfile.
• The SLEAP Worker Dockerfile is located at ./webRTC_worker_sleap_container/Dockerfile
• The Signaling Server Dockerfile is located at ./webRTC_external/Dockerfile.
• The repo has CI set up in .github/workflows for building and pushing the image when making changes.
  • The workflow uses the linux/amd64 & linux/arm64 platform to build.
• .devcontainer/devcontainer.json is convenient for developing inside a container made with the DockerFile using Visual Studio Code.

Currently, the Signaling Server is run on a single AWS EC2 node inside a container such that both the Worker/SLEAP Worker (inside its own separate container) and the Client (outside both containers) can register with this Signaling Server and discover each other to initate simple text messaging and file sharing. The SLEAP Worker for the webRTC MVP Demo is also capable of using the received files for training, as it is hosted on the Salk GPU cluster.

Installation

Make sure to have Docker Daemon running first

You can pull the image if you don't have it built locally, or need to update the latest, with

docker pull eberrigan/sleap-cuda:latest

Usage

Then, to run the image with gpus interactively:

docker run --gpus all -it eberrigan/sleap-cuda:latest bash

and test with

python -c "import sleap; sleap.versions()" && nvidia-smi

In general, use the syntax

docker run -v /path/on/host:/path/in/container [other options] image_name [command]

Note that host paths are absolute.

Use this syntax to give host permissions to mounted volumes

docker run -u $(id -u):$(id -g) -v /your/host/directory:/container/directory [options] your-image-name [command]

docker run -u $(id -u):$(id -g) -v ./tests/data:/tests/data --gpus all -it eberrigan/sleap-cuda:latest bash

Test:

 python3 -c "import sleap; print('SLEAP version:', sleap.__version__)"
 nvidia-smi # Check that the GPUs are discoverable
 sleap-train "tests/data/initial_config.json" "tests/data/dance.mp4.labels.slp" --video-paths "tests/data/dance.mp4"

Notes:

• The eberrigan/sleap-cuda is the Docker registry where the images are pulled from. This is only used when pulling images from the cloud, and not necesary when building/running locally.
• -it ensures that you get an interactive terminal. The i stands for interactive, and t allocates a pseudo-TTY, which is what allows you to interact with the bash shell inside the container.
• The -v or --volume option mounts the specified directory with the same level of access as the directory has on the host.
• bash is the command that gets executed inside the container, which in this case is to start the bash shell.
• Order of operations is 1. Pull (if needed): Get a pre-built image from a registry. 2. Run: Start a container from an image.

Contributing

• Use the devcontainer.json to open the repo in a dev container using VS Code for the Worker and Signaling Server separately.

  • There is some test data in the tests directory that will be automatically mounted for use since the working directory is the workspace.
  • Rebuild the container when you make changes using Dev Container: Rebuild Container.

• Please make a new branch, starting with your name, with any changes, and request a reviewer before merging with the main branch since this image will be used by others.

• Please document using the same conventions (docstrings for each function and class, typing-hints, informative comments).

• Tests are written in the pytest framework. Data used in the tests are defined as fixtures in tests/fixtures/data.py (https://docs.pytest.org/en/6.2.x/reference.html#fixtures-api).

Build

To build and push via automated CI, just push changes to a branch.

• Pushes to main result in an image with the tag latest.
• Pushes to other branches have tags with -test appended.
• See .github/workflows for testing and production workflows.

To test test images locally use after pushing the test images via CI:

docker pull eberrigan/sleap-cuda:linux-amd64-test

then

docker run -v ./tests/data:/tests/data --gpus all -it eberrigan/sleap-cuda:linux-amd64-test bash

To build locally for testing you can use the command (from the root of the repo):

docker build --platform linux/amd64 ./sleap_cuda

Support

contact Elizabeth at eberrigan@salk.edu

Temp

1. start the server
   • mamba env create -f env.yaml to make conda env for the first time
   • mamba activate sleap-webrtc to activate conda env every time
   • python3 sleap_webRTC\webRTC_external\server.py
2. start the worker in the container
   • make sure to have docker desktop running on computer and signed in
   • visit https://github.com/talmolab/sleap-cuda-container/pkgs/container/sleap-webrtc for latest docker image to pull the test image
   • docker pull ghcr.io/talmolab/sleap-webrtc:linux-amd64-99025482bab7ecd1ac7859f8eff1773738c5a17e-test (git hash changes each time we modify the image)
   • docker run -it ghcr.io/talmolab/sleap-webrtc:linux-amd64-99025482bab7ecd1ac7859f8eff1773738c5a17e-test bash to run interactive container with bash
   • python3 /app/worker.py
3. start the client
   • new terminal
   • mamba activate sleap-webrtc to activate conda env every time
   • python3 sleap_webRTC\webRTC_external\client.py
4. send messages between client and worker
   • "quit" to exit