couler-proj / couler

Unified Interface for Constructing and Managing Workflows on different workflow engines, such as Argo Workflows, Tekton Pipelines, and Apache Airflow.

Home Page:https://couler-proj.github.io/couler/index.html

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Unexpected env "NVIDIA_**" in YAML

Yancey1989 opened this issue · comments

After running a simple Couler program, the generated Argo YAML always contains the NVIDIA_* envs just like:

        env:
          - name: NVIDIA_VISIBLE_DEVICES
            value: ""
          - name: NVIDIA_DRIVER_CAPABILITIES
            value: ""

Why these variables and how can I remove the unexpected environment variables in specs?

These two variables are needed to set with empty value to avoid unplanned mounting GPU devices to containers. If the image contains setting these two variables, the container will try to load gpu device even you don't request GPU from the scheduler.

Closing this for now since this change is no longer necessary for the integration with SQLFlow. This is an issue with docker in general and keeping this around would avoid the unexpected case mentioned above. Feel free to re-open if you ever come across any issues with this.