I am having some trouble executing a workflow with FlyteRemote which is a list[DataClasses] The input type of the workflow is a list[InputDocument] where input documents is as follows

@dataclass_json
@dataclass
class InputDocument:
    document_id: str
    s3_path: str

The workflow looks like this and goes through multiple tasks which returned intermediary dataclass types

def my_workflow(inputs: list[InputDocument]) -> list[ProcessedDocument]:
    ...

When executing the workflow via

FlyteRemote.execute

I get the following error

File "flytekit/core/type_engine.py", line 321, in assert_type
    for f in dataclasses.fields(type(v)):  # type: ignore
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/Cellar/python@3.11/3.11.5/Frameworks/Python.framework/Versions/3.11/lib/python3.11/dataclasses.py", line 1244, in fields
    raise TypeError('must be called with a dataclass type or instance') from None
TypeError: must be called with a dataclass type or instance

With the data

{
  "inputs": [
    {
      "document_id": "test1",
      "s3_path": "<s3://inputs/123.txt>"
    }
  ]
}

Essentially FlyteRemote thinks that the

list

should be a dataclass. Any idea what is going on here? It seems like flytkit can’t handle a list (native) of dataclasses. I can get the workflow to run by adding an additional

type_hints={"inputs" : list[dict[str,str]]}

, but I would really rather not constrain the workflow to this. I want to be able to arbitrarily call multiple different workflows with different inputs.

We have an extremely similar class and function to yours that works -- our

InputDocument

class extends

DataClassJsonMixin

(from

dataclasses_json

), can you see if that resolves your issue?

Thanks @Joe Kelly - will try!

could you try

typing.List

instead of

list

@Kevin Su @Joe Kelly Thanks for your comments. Unfortunately neither of these have worked and we are seeing the same behaviour. A issue has been raised here in aid of progressing this https://github.com/flyteorg/flyte/issues/4098