Ray runtime env from docker image or bazel target

Hi! We are interested in using Ray, but our python code is packaged using bazel rather than pip or conda. In our other similar setups (e.g. spark) we create a docker image that contains our code and its dependencies and run the workers on that. Is this possible to create a ray cluster with our code already installed onto the workes in this way, or otherwise provide our bazel dependencies to the ray cluster?


That should be doable with these docker configs: Cluster YAML Configuration Options — Ray v1.10.0

Alternatively, you can install dependencies with worker_setup_commands, but going with the docker image will likely be better in your case.

1 Like