I might be missing something but after trying many approaches I find flow registration to be the weak spot of Prefect. On one hand the architecture is supposed to be decoupled with the flows using

Storage

, on the other hand you can register them directly from the

flow

object, thus de facto coupling the registration to your main repo (where your flow code lives). You need to specify a runner in which all dependencies can be resolved (makes sense ofc), but then

Storage

also needs all dependencies to be set up (at least when using

Module

). When trying to set it up “externally” - from the CLI - it won’t handle it because the flows are run as scripts, so if they are in a package they won’t be able to resolve any dependencies. Long story short: for now I didn’t find a straight forward way of deploying. But don’t get me wrong, still love Prefect 🙂

The flow's dependencies need to be resolved because we need to run the flow context to construct the DAG to send on registration

You can put the imports in a

try:
  ...
except ModuleNotFoundError:
   print("Skipping module imports for registration...")