Default start_date and schedule_interval to None

[BasPH]

Description

I suggest changing the bare-minimum code required for instantiating a DAG:

# Proposed

from airflow import DAG
from airflow.operators.bash import BashOperator

with DAG(dag_id="no_schedule_dag") as dag:
    BashOperator(task_id="foobar", bash_command="echo hello")

This is currently not possible because start_date is a required argument:

I regularly see people using externally triggered DAGs and you're currently forced to set a placeholder start_date because Airflow requires it:

# Current situation

import datetime

from airflow import DAG
from airflow.operators.bash import BashOperator

with DAG(dag_id="no_schedule_dag", start_date=datetime.datetime(2022, 1, 1), schedule_interval=None) as dag:
    BashOperator(task_id="foobar", bash_command="echo hello")

When people want to set a schedule_interval, they intentionally think of it, so defaulting to None feels like a better UX to me.

Use case/motivation

To improve the Airflow user experience, I suggest removing the start_date requirement on DAG:

DAG(dag_id="my_dag")

This DAG could be triggered externally.

Related issues

No response

Are you willing to submit a PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[ashb]

The "default" schedule_interval is currently daily. I'm all for this so long as we can do it without breaking back-compat.

[BasPH]

Yea that won't be possible. The default daily schedule interval has been there since the very first commit, and always felt like undocumented magic to me, and goes against Python's "explicit is better than implicit".

[ashb]

Then it's unlikely we can do it before 3.0

[ashb]

I think we might be able to do this by changing the function default to NOT_SET (a special singleton object) and then changing the value at runtime after checking other parameters.

(Changing the code-level default is possible so long as the behaviour to the user remains the same)

[potiuk]

Yep. This looks like a good way to do it. Technically this is I think Sentinel. I used similar approach in Breeze where I wanted to distinguish "defaut" value from "last used" one

airflow/dev/breeze/src/airflow_breeze/utils/custom_param_types.py

Line 109 in 011fbae

class CacheableDefault:

[malthe]

See also discussion here https://lists.apache.org/thread/hyctljm0o1odn1bntp4fcgok0d0433ov.

[eladkal]

See also discussion here https://lists.apache.org/thread/hyctljm0o1odn1bntp4fcgok0d0433ov.

As of Airflow 2.8 start_date is not required when schedule=None
#35199

I think this solves this issue to the best we can do without causing breaking changes.

[uranusjr]

With #41321 we should be set to do this.