We're trying to load our repositories from a python package.  Our package structure is like so:

my_package:
  workspace.yaml
  subfolder:
     repository.py

The

workspace.yaml

file under

my_package

looks like this:

load from:
    - relative_path: subfolder/repository.py

We start dagit from a folder with another

workspace.yaml

in it that contains:

load_from:
  - python_package: my_package

We get this error:

dagster.core.errors.DagsterUserCodeProcessError: (DagsterInvariantViolationError) - dagster.core.errors.DagsterInvariantViolationError: No pipelines or repositories found in "my_package".

I've changed some names and paths to simplify this.  Also, this loads properly when we do it from our local environments where

my_package

is not packaged (i.e. we set our

PYTHONPATH

to the

my_package

parent folder and our

workspace.yaml

in the folder where dagit starts points at relative path

my_package/subfolder/repository.py

So, is this meant to work?  We'd like to be able to have a

workspace.yaml

file in our deployment target that has pointers to packages as they're developed and these packages will have repositories inside them.  In that way we can allow different teams to develop their own dagster packages and deploy them to a central dagit server.

are the

@respository

objects in the top level

__init__.py

of

my_package

? We don’t crawl the whole package, just the entry point. For example, i am guessing if you did

python_pacakge: my_package.subfolder.repository

it would work

I thought I'd already moved the module with the

@repository

object up to the top level of the package (same folder as

__init__.py

without luck. Tried your idea and still no good, BUT, I must not have tested the first idea properly. Package structure is now:

my_package:
  workspace.yaml
  repository.yaml
  subfolder:
     ...

The workspace.yaml file looks like:

load_from:
  - python_file:
      relative_path: repository.py

And everything works! Thanks

just to be clear, i meant that the @repository object needs to be imported in the

__init__.py

file itself

OK, I'll give that a try as well @alex