Anyone know if there's a way for the flyte deployment to handle verifying if images exist in a registry and if not to build them using a buildkit server deployed in the same namespace as flyte, all without having the pyflyte client to talk to the remote buildkit server? Right now, pyflyte has to talk to buildkit server to orchestrate image verification and building, but it would be a lot easier from a client setup perspective to just have something like flyte propeller handle this when buildkit is deployed in same namespace

ImageSpec support pluggable image builders. You can implement one that does just about anything 🙂 @ union.ai we built one that does what you describe (a bit fancier too) it leverages multi-tier cache for pip packages, can optimize and cache images... etc.

This is what my team is using to build images: Kubernetes driver | Docker Docs

@high-park-82026 You got it right! Ideally, I'd like a flyte deployment where a user just gets the flyte config file needed for pyflyte to work, and your done, everything else is handled by the cluster. This way, the only thing the user needs to configure pyflyte/flytectl is the config.yaml file and a requierments.txt file which installs pyflyte

That's a good point, configuring docker is a pain