A webapp for monitoring GPU machines. The app aggregates the output from gpustat across all machines and displays them on a single page.
The frontend is written in Vue.js (+ Vuetify) and backend is written in Flask.
The backend uses paramiko to connect to the servers via ssh in order to query the information.
Each server is queried once per minute (can be changed).
Multithreading is used to send out the requests to avoid long waiting times.
Install Node.js and npm (example given for Ubuntu):
sudo apt install nodejs
sudo apt install npm
Clone the repository:
git clone https://github.com/matthias-wright/server-dashboard.git
cd into the frontend directory:
Install node packages:
cd into the backend directory:
Install python packages:
pip install -r requirements.txt
Add your server addresses to server-dashboard/backend/config/servers.yaml:
gpustat needs to be installed on every server you add to the list.
Add your credentials to server-dashboard/backend/config/config.yaml:
The username will be used for the ssh requests and path_to_ssh_key is the path to the corresponding ssh key. If you currently use a password to login to the servers, you have to generate a ssh key pair and copy your public key to the servers (see here for more instructions).
Don’t use ~ for path_to_ssh_key, write out the full path.
process_timeout is the time in seconds to wait for a server to respond and port is the port for the Flask app. You don’t have to change these.
After installing and configurating the app, you can start it as follows:
cd into the server-dashboard directory:
Run the run.py script:
The app should be available at: http://localhost:8080/
I have been using the app for a while now and it is working well for me but it has not been extensively tested. If you encounter any problems while setting the app up, don’t hesitate to open an issue!