josh-fell opened a new issue #19538:
URL: https://github.com/apache/airflow/issues/19538


   ### Apache Airflow version
   
   2.2.0
   
   ### Operating System
   
   Debian GNU/Linux 11 (bullseye)
   
   ### Versions of Apache Airflow Providers
   
   ```shell
   apache-airflow-providers-amazon          1!2.2.0
   apache-airflow-providers-cncf-kubernetes 1!2.0.3
   apache-airflow-providers-elasticsearch   1!2.0.3
   apache-airflow-providers-ftp             1!2.0.1
   apache-airflow-providers-google          1!6.0.0
   apache-airflow-providers-http            1!2.0.1
   apache-airflow-providers-imap            1!2.0.1
   apache-airflow-providers-microsoft-azure 1!3.2.0
   apache-airflow-providers-mysql           1!2.1.1
   apache-airflow-providers-postgres        1!2.3.0
   apache-airflow-providers-redis           1!2.0.1
   apache-airflow-providers-slack           1!4.1.0
   apache-airflow-providers-sqlite          1!2.0.1
   apache-airflow-providers-ssh             1!2.2.0
   ```
   
   ### Deployment
   
   Astronomer
   
   ### Deployment details
   
   Local deployment using the Astronomer CLI.
   
   ### What happened
   
   Creating a TaskFlow function with a return type annotation of `dict` does 
not yield `XComs` for each key within the returned dict.  Additionally, the 
inference does not work for both `dict` and `Dict` (without arg annotation) 
types in Python 3.6.
   
   ### What you expected to happen
   
   When creating a TaskFlow function and not explicitly setting 
`multiple_outputs=True`, the unfurling of `XComs` into separate keys is 
inferred by the return type annotation (as noted 
[here](https://airflow.apache.org/docs/apache-airflow/stable/tutorial_taskflow_api.html#multiple-outputs-inference)).
 When using a return type annotation of `dict`, separate `XComs` should be 
created. There is an explicit check for this type as well:
   
   
https://github.com/apache/airflow/blob/7622f5e08261afe5ab50a08a6ca0804af8c7c7fe/airflow/decorators/base.py#L207
   
   Additionally, on Python 3.6, the inference should handle generating multiple 
`XComs` for both `dict` and `typing.Dict` return type annotations as expected 
on other Python versions.
   
   ### How to reproduce
   
   This DAG can be used to demonstrate the different results of dict typing:
   ```python
   from datetime import datetime
   from typing import Dict
   
   from airflow.decorators import dag, task
   from airflow.models.baseoperator import chain
   from airflow.models import XCom
   
   
   @dag(
       start_date=datetime(2021, 11, 11),
       schedule_interval=None,
   )
   def __test__():
       @task
       def func_no_return_anno():
           return {"key1": "value1", "key2": "value2"}
   
       @task
       def func_with_dict() -> dict:
           return {"key1": "value1", "key2": "value2"}
   
       @task
       def func_with_typing_dict() -> Dict:
           return {"key1": "value1", "key2": "value2"}
   
       @task
       def func_with_typing_dict_explicit() -> Dict[str, str]:
           return {"key1": "value1", "key2": "value2"}
   
       @task
       def get_xcoms(run_id=None):
           xcoms = XCom.get_many(
               dag_ids="__test__",
               task_ids=[
                   "func_no",
                   "func_with_dict",
                   "func_with_typing_dict",
                   "func_with_typing_dict_explicit",
               ],
               run_id=run_id,
           ).all()
   
           for xcom in xcoms:
               print(f"Task ID: {xcom.task_id} \n", f"Key: {xcom.key} \n", 
f"Value: {xcom.value}")
   
       chain(
           [
               func_no_return_anno(),
               func_with_dict(),
               func_with_typing_dict(),
               func_with_typing_dict_explicit(),
           ],
           get_xcoms(),
       )
   
   
   dag = __test__()
   
   ```
   
   **Expected `XCom` keys**
   - func_no_return_anno
     - `return_value`
   - func_with_dict
     - `return_value`, `key1`, and `key2`
   - func_with_typing_dict
     - `return_value`, `key1`, and `key2`
   - func_with_typing_dict_explicit
     - `return_value`, `key1`, and `key2`
   
   Here is the output from the `get_xcoms` task which is gathering all of the 
`XComs` generated for the run:
   
![image](https://user-images.githubusercontent.com/48934154/141336206-259bd78b-8ef3-4edb-81a6-b161d783f39f.png)
   
   The `func_with_dict` task does not yield `XComs` for `key1` and `key2`.
   
   ### Anything else
   
   The inference also doesn't function as intended on Python 3.6 when using 
simple `dict` or `Dict` return types.
   
   For example, isolating the existing 
`TestAirflowTaskDecorator::test_infer_multiple_outputs_using_typing` unit test 
and adding some parameterization on Python 3.6:
   ```python
   @parameterized.expand(
           [
               ("dict", dict),
               ("Dict", Dict),
               ("Dict[str, int]", Dict[str, int]),
           ]
       )
       def test_infer_multiple_outputs_using_typing(self, _, 
test_return_annotation):
           @task_decorator
           def identity_dict(x: int, y: int) -> test_return_annotation:
               return {"x": x, "y": y}
   
           assert identity_dict(5, 5).operator.multiple_outputs is True
   ```
   **Results**
   
![image](https://user-images.githubusercontent.com/48934154/141338408-5d7f2877-6465-4c81-857f-5ca6d1b612ee.png)
   
   
   However, since Python 3.6 will reach EOL on 2021-12-23, this _may_ not be an 
aspect that needs to be fixed.
   
   ### Are you willing to submit PR?
   
   - [X] Yes I am willing to submit a PR!
   
   ### Code of Conduct
   
   - [X] I agree to follow this project's [Code of 
Conduct](https://github.com/apache/airflow/blob/main/CODE_OF_CONDUCT.md)
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to