proposal: use args not command for kubernetes

Hi,

thanks for this great project 😃

I noticed that in a few places (helm chart, dagster-celery-k8s) pipeline pods are creating setting the command to dagster.

There is a different meaning between command in k8s and in docker, because in kubernetes what is specified in command overrides the image entrypoint (see: kubernetes docs). This might be a problem for docker images where the entrypoint is set (e.g. to provide a conda environment).

Would it be possible to not set command and use args instead so that the entrypoint is preserved? In alternative, is it possible to let the user specify the command and args?

This is just a proposal, but if you are happy with it I can try opening a PR.

Thanks again 🙂

[abkfenris]

The dagster kubernetes chart still uses the “command” in the user template deployment. Is this on purpose or would the same logic apply there as well?

I’m running in to the same problem there trying to use dagster that’s installed in a conda environment

https://github.com/dagster-io/dagster/blob/a790cd3736eecccea0e63777e0853213b2eba507/helm/dagster/charts/dagster-user-deployments/templates/deployment-user.yaml#L44

I’m running into this with user code deployments too, specifically using a micromamba base image with conda environments.

[daviskirk]

The dagster kubernetes chart still uses the “command” in the user template deployment. Is this on purpose or would the same logic apply there as well?

I’m running in to the same problem there trying to use dagster that’s installed in a conda environment

https://github.com/dagster-io/dagster/blob/a790cd3736eecccea0e63777e0853213b2eba507/helm/dagster/charts/dagster-user-deployments/templates/deployment-user.yaml#L44