Hello, I have defined a Flyte workflow which calls a few flyte Tasks. All but one tasks return a string, and one task,

ingest_reports

, returns a

List[Dict[str, str]]

. Furthermore in my workflow, I have

>>

operator defining the sequence of task execution. Now when I test the workflow, I have a couple of questions: 1. When I test the workflow, I want to mock the task calls. Now,

>>

operator defines the sequence of task execution, and it expects the tasks to return a Promise. Since a Promise is expected, so the mocked task should be able to return a Promise, otherwise this operation fails with an error like:

except Exception as exc:
>                   raise type(exc)(f"Error encountered while executing '{fn_name}':\n  {exc}") from exc
E                   AttributeError: Error encountered while executing 'phishing_reports_workflow':
E                     'list' object has no attribute 'ref'

2. To do so I defined the mock as below:

def mock_promise(value: any):
        magic_mock_promise = MagicMock(spec=Promise)
        mock_output = MagicMock()
        mock_output.value = value
        magic_mock_promise.output = mock_output
        return magic_mock_promise

3. This mocking of tasks to return a FlytePromise works fine as long as task return type is a String. For the task,

ingest_reports

, return type is

List[Dict[str, str]]

. When I try to mock the response of

ingest_reports

using

mock_promise

above, by passing in a list of dict of strings, it fails with exception below:

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = Typed List Transforms (<class 'list'>) to Flyte native
ctx = FlyteContext(file_access=<flytekit.core.data_persistence.FileAccessProvider object at 0x103772730>, level=1, flyte_cli...tackframe=<FrameSummary file /Library/Python/3.9/site-packages/flytekit/core/context_manager.py, line 877 in <module>>)
python_val = <MagicMock name='ingest_phishing_reports()' spec='Promise' id='5383586432'>
python_type = typing.List[typing.Dict[str, str]]
expected = <FlyteLiteral collection_type { map_value_type { simple: STRING } }>

    def to_literal(self, ctx: FlyteContext, python_val: T, python_type: Type[T], expected: LiteralType) -> Literal:
        if type(python_val) != list:
>           raise TypeTransformerFailedError("Expected a list")
E           flytekit.core.type_engine.TypeTransformerFailedError: Expected a list

/Library/Python/3.9/site-packages/flytekit/core/type_engine.py:1042: TypeTransformerFailedError

Any help on how to solve this?

can you copy/paste in a minimal reproducible example please?

Workflow:

@task
def task1() -> List[Dict[str, str]]:
  return [{"messageId": "12345"}]

@task
def task2() -> str:
  return "task2"

@task
def task3() -> str:
  return "task3"

@workflow
def reports_workflow(styx_parameter: datetime.datetime) -> List[Dict[str, str]]:
  dateStr = str(styx_parameter)
  output_list: List[Dict[str, str]] = task1()
  output_str_mocked = task2()
  output_str = task3()

  # Change the order of below two lines and the errors will change depending on if mock promise is compared to list or promise is compared to list. Which is why it makes sense to be able to mock output_list response.
  output_list >> output_str_mocked
  output_list >> output_str
  return output_list

lp_phishing_reports = LaunchPlan.create(
  "ReportsWorkflow",
  reports_workflow,
)

Test:

@patch("phishing_reports_and_training_pipelines.report_workflow.task1")
@patch("phishing_reports_and_training_pipelines.report_workflow.task2")
def test_reports_workflow(
    task2_mock_ref: MagicMock, # Notice the reversed order of args
    task1_mock_ref: MagicMock,
) -> None:
    reports: List[Dict[str, str]] = [{"messageId": "12345"}]

    def mock_promise(value: Any) -> MagicMock:
        magic_mock_promise = MagicMock(spec=Promise)
        mock_output = MagicMock()
        mock_output.value = value
        magic_mock_promise.output = mock_output
        return magic_mock_promise

    # When you uncomment this line, it fails for flytekit.core.type_engine.TypeTransformerFailedError: Expected a list
    # task1_mock_ref.return_value = mock_promise(reports)

    # When you send reports as a list instead of a FlytePromise mock, then it fails for >> comparison
    task1_mock_ref.return_value = reports
    task2_mock_ref.return_value = mock_promise("uri")

    from phishing_reports_and_training_pipelines.report_workflow import (
        reports_workflow,
    )
    result = reports_workflow(
        styx_parameter=datetime.datetime(2020, 12, 18),
    )
    assert result == reports

haven’t looked yet

but confused about the order… are you saying that that’s correct or not correct?

The reverse order, as shown in my test, works!

I've tested your code and it's working for me.

Strange coz if you see the conversation above, @thankful-minister-83577 was able to reproduce the issue. Can you please share the workflow and test you are trying?

Workflow:

from typing import Dict, List

from flytekit import task, workflow


@task
def task1() -> List[Dict[str, str]]:
    return [{"messageId": "12345"}]


@task
def task2() -> str:
    return "task2"


@task
def task3() -> str:
    return "task3"


@workflow
def reports_workflow() -> List[Dict[str, str]]:
    output_list = task1()
    output_str_mocked = task2()
    output_str = task3()

    # Change the order of below two lines and the errors will change depending on if mock promise is compared to list
    # or promise is compared to list. Which is why it makes sense to be able to mock output_list response.
    output_list >> output_str_mocked
    output_list >> output_str
    return output_list

Test:

from typing import Any, Dict, List
from flytekit.testing import patch
from .flyte_remote import reports_workflow, task1, task2


def test_reports_workflow1() -> None:
    x = reports_workflow()
    print(x)


@patch(task1)
@patch(task2)
def test_reports_workflow(task2_mock, task1_mock) -> None:
    reports: List[Dict[str, str]] = [{"messageId": "54321"}]

    # When you send reports as a list instead of a FlytePromise mock, then it fails for >> comparison
    task1_mock.return_value = reports
    task2_mock.return_value = "mocked task 2"

    result = reports_workflow()
    print(result)
    assert result == reports

Interesting!

Can you try with 1.8.2?

Yep

i’m not able to repro the issue

I am not sure then why am still getting TypeError thrown from /flytekit/exceptions/scopes.py:203 when i run the test

E                     Error encountered while executing 'reports_workflow':
E                     unsupported operand type(s) for >>: 'list' and 'str'

.tox/py39-test/lib/python3.9/site-packages/flytekit/exceptions/scopes.py:203: TypeError

this is something we’ll address in the coming months, but yeah, for now local execution will not respect the

>>

order. only tasks with an explicit data dependency will be respected.

I am not sure how to put it better that I did try your files and I get the error I am mentioning when I run the test!

@flaky-jackal-54421, as we mentioned, the code snippet Yee shared is working for us. Can you install a fresh virtual environment and try running Yee's code? It has to work.

Hello again, so I think I found what we were doing differently! My test method looked something like

@patch("spotify_flytekit.core.remote_task_reference.hades.SdkRemoteHadesTaskReference")
@patch("training_pipelines.tasks.ingest_reports")
@patch("training_pipelines.tasks.log_data")
@patch("training_pipelines.tasks.publish_to_gcs_as_json")
def test_reports_workflow(
    gcs_mock_ref: MagicMock,
    log_data_mock_ref: MagicMock,
    ingest_reports_mock_ref: MagicMock,
    mock_reference: MagicMock,
)

I was able to make things work like @thankful-minister-83577 suggested, when I remove the MagicMock instantiation in the method and in patch I just use the method name directly by importing it. Basically I was using python's unittest.mock.Patch, and you suggested to use flytekit.testing.patch But there is one issue I am facing. How to mock the staticmethod using flytekit.testing.patch? So

SdkRemoteHadesTaskReference.generate_hades_partition_uri_v0

and

SdkRemoteHadesTaskReference.publish_hades_partition_v0

are two static method calls which I need to mock.

Are you seeing any error when you use

flytekit.testing.patch

?

Yes, since it is not a flyte task, flytekit.testing.patch raises an exception, "Exception: Can only use mocks on tasks/workflows declared in Python."

Why do you want to use a non-Flyte task?

Good question! There is a library from which we use a static method which we use to publish dataset to a system. Do you suggest we wrap this call inside a flyte method and then use it?

If you're using it in a workflow, then yes, it needs to be called from within a Flyte task.

So we have a spotify flytekit library built on top of your flytekit library. This staticmethod is inside our spotify flytekit library and is used to publish dataset to our Hades system. The recommended way was to call this static method directly from workflow. And it does not look like a flyte task. Let me understand internally why we call it directly from workflow instead of calling as a flyte task!

You mean a flyte task?

I mean a flyte remote task https://docs.flyte.org/projects/flytekit/en/latest/generated/flytekit.remote.remote.FlyteRemote.html

hi sorry are you asking how to mock a remote.fetch’ed entity?