Hi. I am having some trouble making

prefect register --module

work. My project looks like this:

flows
  + __init__.py
  + my_flow.py
  + shared_tasks
    + __init__.py
    + util.py

When I run

prefect register --project 'Prefect Testing' -m '<http://flows.my|flows.my>_flow'

, I get

No module named 'flows'

. What am I missing?

Hi @Florian Kühnlenz, is your module named

flows

or

flow

?

flows

, but sadly the typo was only in slack 😞. I can also import the module just fine in python.

What is your Prefect version?

Nope, I have not yet seen this. Prefect version is 0.14.15

I haven’t had time yet to try this myself but I’ll try to get back to you later

Mmm. If that is the case it is very confusing that

prefect register

would need the module installed via pip while the normal

flow.register

method would not. Is this mentioned somewhere in the documentation?

This is not because of the CLI. This is specific to Module storage because it loops through the flows of the module and registers them one at a time. Or do you mean

flow.register

with the

Module

storage? This is not documented. I’m considering adding it

Well, I have my flow setup with module storage, because file storage brought some problems with debugging. But even given the example you linked from @Jim Crist-Harif I don’t see how I should discover that the module needs to be installed via pip. But maybe I just missed this in the context. I think installing does not make to much sense for my case either. For context, we want to have the flows stored in docker as well as docker being used as run environment. We managed a setup with docker as run config and local storage for the flows. But it had some downsides for debugging and was cumbersome around the edges to register the flows and managing the imports (manipulate the sys path etc).

I understand the pip install was not intuitive. I will add it to the Module storage docs. About the cumbersome imports, i’ve had issues too at previous jobs with sys.paths and we went to modules and absolute paths. Putting your flows as a module and then installing that wheel in the Docker container actually may be an approach to consider. What was cumbersome when registering the flows?

I will play around with that tomorrow. We use poetry for package management anyway. So creating and installing a module might be fairly easy. About the registering: we used a separate script to register each flow. Which made it necessary to have a

if __name__ == “__main__”:
    flow.register()

At the bottom of each flow. Which one would need to change for debugging. With the new cli I did not manage to make it work, maybe because there is a fix upstream about the python context (https://github.com/PrefectHQ/prefect/pull/4332)

Got it. I think the module is a cleaner way to debug because you wouldn’t register it in the flow definition file. You can also import run a low individually to debug