What I want to accomplish is to have a single image with the dependencies I need, and use that image to execute multiple flows

What I want to avoid is having to create an image for every flow, because it takes a while to build and transmit each image to the docker registry during our devops process

1. When executed, the flow will be run from inside of the container, so the custom modules will be available assuming they were added and the python path was adjusted. Am I interpreting these lines correctly? https://github.com/anna-geller/packaging-prefect-flows/blob/master/flows_no_build/docker_script_kubernetes_run_custom_ecr_image.py#L17-L18. It seems to me that the custom modules must also be available at registration time, because the flow's python module must be executed to register the flow

2. This pattern assumes that the image already exists and is in a repository. When does this image get built? I've only ever done

flow.register()

which builds the image in what seems like a prefect context, because there are several Docker build steps relate to prefect that we didn't define in our Dockerfile.

flow.register()

also couples the registration and image building, and this demo app shows how to do registration--but how is the image built?

I found the answer to #2! At https://docs.prefect.io/orchestration/execution/overview.html#storage

Is my thinking accurate on #1?

on #2 - in order to use the same docker image for multiple flows, you really don't use the "typical" Docker storage registration which pickles your flow at registration, builds the image, and pushes it to the registry. Instead, you build the image yourself. I explained that in the docstring here:

To use flow script files in Docker, we need the following arguments to be set properly:
- `stored_as_script` = True
- `path` must point to the path in the image
- if this path is different from/opt/prefect/flows/YOUR_FLOW.py,
  then you also need to change the default `prefect_directory` to your custom directory
- you need to add the flow to the storage explicitly before registering, e.g.: docker_storage.add_flow(flow)
- finally, registering the flow must happen from Python, since the `prefect register` CLI doesn't have the option
  to pass build=False, and this is critical to include to prevent pickling flow and rebuilding the image on the registration
- the parameter `ignore_healthchecks` on the Docker storage is optional and doesn't affect this process at all

example build commands are here https://github.com/anna-geller/packaging-prefect-flows/blob/master/aws_docker_build_commands.bash LMK if you have any more questions about it