Exposing Custom Ports#

HPE Machine Learning Development Environment allows you to expose a custom network port in a task container, and access it using a local tunnel.

For multi-container tasks, such as distributed training experiments, only the ports on the chief container (rank=0) will be exposed.

Configuration#

First, specify the ports in the environments -> proxy_ports section of the experiment or task config, for example:

environment:
  proxy_ports:
    - proxy_port: 8265
      proxy_tcp: true

Launch your task or experiment normally. Then, use the det CLI to start a tunnel. Running the following command will setup a tunnel proxying localhost:8265 to port 8265 in the task container.

python -m determined.cli.tunnel --listener 8265 --auth $DET_MASTER $TASK_ID:8265

where $DET_MASTER is your HPE Machine Learning Development Environment master address, and $TASK_ID is the task id of the launched task or experiment. You can look up the task id using CLI command det task list.

Alternatively, you can use a shortcut which allows to launch the experiment, follow its logs, and run the tunnel all at once:

det e create config_file.yaml model_def -f -p 8265

Unauthenticated Mode#

Optionally, you can run a tunnel with determined authentication turned off. This mode may be useful when the proxied app is handling security by itself, such as a web app protected by username and password. To use it,

  1. Add unauthenticated: true option in the task config.

  2. Omit --auth option from the tunnel CLI.

environment:
  proxy_ports:
    - proxy_port: 8265
      proxy_tcp: true
      unauthenticated: true

python -m determined.cli.tunnel --listener 8265 $DET_MASTER $TASK_ID:8265