Deployment Overview of CogVideo on Server¶

Prerequisites and Basic Requirements¶

The server must meet the following operating system and hardware requirements to successfully deploy the application: - Operating System: Ubuntu 22.04 LTS. - Privileges: Root access or sudo privileges are required for installation and configuration. - Hardware: NVIDIA GPU with compatible drivers is mandatory for CUDA acceleration. - Network: The server must have internet access to download dependencies and certificates. - Ports: Port 7860 is used internally by the application; external access is managed via the proxy server.

File and Directory Structure¶

The application and its supporting components are organized in the following directory structure: - /opt/CogVideo: The main installation directory containing the application source code cloned from the repository. - /opt/CogVideo/env: The Python virtual environment directory containing installed dependencies. - /root/nginx: The directory containing the Docker Compose configuration for the proxy server. - /data/nginx/user_conf.d: The directory storing custom Nginx configuration files for specific host keys. - /data/nginx/nginx-certbot.env: The environment file containing configuration for the Nginx-Certbot service. - /etc/systemd/system/cogvideo.service: The systemd unit file defining the application service.

Application Installation Process¶

The application is installed by cloning the source repository and setting up a dedicated Python environment with specific versions of PyTorch and other dependencies.

1. System Preparation: The system is updated, and the linux-generic-hwe-22.04 kernel headers are installed if running on Ubuntu 22.04. The GCC compiler is installed to support CUDA installation.
2. CUDA and Driver Setup:
   • The CUDA keyring is downloaded and installed.
   • The cuda-toolkit and nvidia-cuda-toolkit packages are installed via apt.
   • Environment variables for PATH and LD_LIBRARY_PATH are appended to ~/.bashrc to include /usr/local/cuda/bin and /usr/local/cuda/lib64.
   • The nouveau kernel module is removed, and NVIDIA drivers are initialized.
3. Docker and NVIDIA Container Toolkit:
   • Docker is installed if not present.
   • The nvidia-docker2 package is installed to enable GPU support within containers.
   • The Docker service is restarted to apply changes.
4. Application Source and Environment:
   • The CogVideo repository is cloned to /opt/CogVideo.
   • Python 3.10 and python3.10-venv are installed.
   • A virtual environment is created at /opt/CogVideo/env.
   • pip is upgraded within the virtual environment.
5. Dependency Installation:
   • Core requirements are installed from /opt/CogVideo/requirements.txt.
   • PyTorch version 2.0.1+cu118, torchvision version 0.15.2+cu118, and torchaudio version 2.0.2+cu118 are installed with CUDA 118 support.
   • The moviepy package version 1.0.3 is installed.

Docker Containers and Their Deployment¶

The proxy server is deployed using Docker Compose to handle SSL termination and routing.

• Image: The container uses the jonasal/nginx-certbot:latest image.
• Configuration: The deployment is managed via a docker compose file located at /root/nginx/compose.yml.
• Volumes:
  • nginx_secrets: An external volume mounted to /etc/letsencrypt for storing SSL certificates.
  • /data/nginx/user_conf.d: Mounted to /etc/nginx/user_conf.d to provide custom Nginx configurations.
• Network: The container runs in host network mode.
• Environment: The service uses an environment file located at /data/nginx/nginx-certbot.env and sets the CERTBOT_EMAIL to [email protected].
• Execution: The container is started using the command docker compose up -d from the /root/nginx directory.

Proxy Servers¶

Nginx is configured as a reverse proxy to route traffic to the application running on the local host.

• Routing Configuration: A custom configuration file is generated at /data/nginx/user_conf.d/{{ prefix }}{{ server_id }}.hostkey.in.conf.
• Proxy Pass: The Nginx configuration includes a location / block that forwards requests to http://127.0.0.1:7860.
• SSL Management: The nginx-certbot container automatically manages SSL certificates using Let's Encrypt.
• Custom Domains: The proxy setup supports custom domains defined in the user configuration files within the /data/nginx/user_conf.d directory.

Starting, Stopping, and Updating¶

The CogVideo application is managed as a systemd service, allowing for standard service control commands.

• Service Name: cogvideo
• Start the Service:

  sudo systemctl start cogvideo
  

• Stop the Service:

  sudo systemctl stop cogvideo
  

• Restart the Service:

  sudo systemctl restart cogvideo
  

• Enable on Boot:

  sudo systemctl enable cogvideo
  

• Check Status:

  sudo systemctl status cogvideo
  

The service is configured to restart automatically (Restart=always) and runs as the root user from the working directory /opt/CogVideo. The entry point for the application is /opt/CogVideo/env/bin/python inference/gradio_composite_demo/cogstudio.py.

Permission Settings¶

The following permissions are applied to ensure the system operates correctly:

• /root/nginx: Directory permissions are set to 0755 with root as the owner and group.
• /root/nginx/compose.yml: File permissions are set to 0644 with root as the owner and group.
• /tmp/install_script.sh: The installation script is created with executable permissions 0755.
• /etc/systemd/system/cogvideo.service: The service file is created with standard system file permissions.