The cloud-native orchestrator for the whole development lifecycle, with integrated lineage and observability.

dagster

Hello!

My team is trying to set up a dynamically loaded `repository` with the `RepositoryData` class. <https://docs.dagster.io/_apidocs/repositories#dagster.repository|I adapted the example from the Repository documentation> (I'll post code in the comments).

*The problem*: Despite configuring the dynamic-repo's `working_directory` in `workspace.yaml` Dagit has trouble finding the modules due to working-directory conflicts.

More specifically, if a breakpoint is inserted inside `DynamicRepo.get_all_pipelines()`,  the debugger tells me the working-directory hasn't changed from.

Have I misunderstood the documentation? <https://docs.dagster.io/concepts/repositories-workspaces/workspaces#loading-relative-imports>

My code:
`workspace.yaml`
```load_from:
- python_file:
    relative_path: repo.py
    working_directory: /some/far/away/directory```
`repo.py` , located in a distant folder
```import importlib
from pathlib import Path
from dagster import repository, RepositoryData

class DynamicRepo(RepositoryData):
    def get_all_pipelines(self):
        return [
            self._construct_job_def_from_filename("dynamically_generated_job1.py"),
            self._construct_job_def_from_filename("dynamically_generated_job2.py")
        ]

    def _construct_job_def_from_filename(self, filename):
        # ... construct jobs by importing them as modules
        # If we break here and call Path.cwd(), we recieve the DAGIT_HOME as the working directory
        # I expected Path.cwd() to return "/some/far/away/directory"
        module_name = Path(filename).name
        module = importlib.import_module(module_name)
        return module.__dict__['main_job']```
