Hello, I would like to start a discussion about clearml-session .

Clearml-session uses ssh tunneling by default. Although ssh tunneling has advantages such as security and convenience in accessing remote servers from local machines, it also has disadvantages such as the need to authenticate with clearml server for each local machine, the requirement to always have ssh tunneling connection, and slow data transfer speeds.

For these reasons, I modified the code for clearml-agent and clearml-session to enable port opening for ssh, vs code, and jupyter, and developed a Kubernetes service to enable connections to each of these tools. With this modification, developers do not need to authenticate on their local machines to use clearml-session or clearml SDK. After cloning a task and modifying the Docker image and selecting queue developers can immediately connect to ssh, vs code, or JupyterLab using the external IP listed in the configuration.

I am interested in hearing your thoughts on this approach.