Use enum for task_states/dag_states

[mik-laj]

Hello,

We use Python 3+, so we should consider using an enumerated type for objects that are buckets for contents This will give us better type hints.
https://github.com/apache/airflow/blob/master/airflow/utils/state.py

Best regards,
Kamil Breguła

[mik-laj]

CC: @turbaszek

[muyajil]

Hi,
I'd like to help out with this issue.

[turbaszek]

@muyajil should I assign you?

[mik-laj]

@muyajil I assigned you to this ticket. I look forward to your contributions 😻

@turbaszek I think we can assign people without confirmation. The assignment of people is important because it is visible on the ticket list. This makes finding free tickets easier.

[chutcheson]

Is this still an open issue?

[muyajil]

@chutcheson Yes it is, unfortunately I never got the chance to really get started

[andrewgodwin]

I'll take this up, if someone wants to assign me.

[andrewgodwin]

I would like to ask, though - do we just want the type hinting here (in which case we can just declare a nice Literal type for the state tuples), or take on the problems that changing a tuple to an enum will cause? There's a few places in Airflow, and presumably in some third-party code, that presume they can do operations on State.task_states as if it were a sliceable/iterable.

(and also SQLAlchemy will not like the Enum values, resulting in a lot of casting back to string)

[uranusjr]

I briefly read related source, and it seems to me the main advantage would be to make sure the right values are passed the right value to SQLAlchemy, so Literal feels like a better solution. Enum may even be a bit awkward in some places, e.g.

airflow/airflow/models/dagrun.py

Line 490 in 44a6648

tis = list(self.get_task_instances(session=session, state=State.task_states + (State.SHUTDOWN,)))