Got it—with the current Helm chart and celery-k8s executor you’ll need two images:
1. An image w/ user code which you’ll rebuild and push to a private repo and use this for Dagit and for pipeline run execution.
2. An image for the Celery workers; this needs Dagster libraries but no user code
In the near future, (1) will be relaxed so that the hosted Dagit does not need user code, and only the pipeline run image needs it. We just haven’t updated the Helm/K8s integration to take advantage of the new user code separation that we’ve built in 0.8.0 yet