DAGs can run before start_date time

[dimberman]

Apache Airflow version:

Kubernetes version (if you are using kubernetes) (use kubectl version):

Environment:

• Cloud provider or hardware configuration:
• OS (e.g. from /etc/os-release):
• Kernel (e.g. uname -a):
• Install tools:
• Others:
  What happened:

If you turn off a DAG in the UI, there seemingly is no way to prevent "missed" runs to schedule after the DAG is turned back on. I thought the workaround for this, since it is not a parameterized option to prevent, would be to update the start_date in the DAG code before turning the DAG back on. This does not work, and therefore the scheduler is running dag_runs before the listed start_date.

To reproduce:

Create a DAG with a schedule_interval
Let the DAG run at least once
Turn off the DAG in the UI
Allow the schedule_interval to pass at least twice
Update the start_date in the DAG to be be after the two interval time
(I then removed the compiled python file and restarted airflow/scheduler just to make sure)
Turn DAG back on in UI

Result: All dag_runs that were "missed" while the DAG was turned off run, despite the start_date being later.

Ideally the start_date would always be honored. And also there would be a parameter to just not run any "missed" dag_runs.

What you expected to happen:

How to reproduce it:

Anything else we need to know:

Moved here from https://issues.apache.org/jira/browse/AIRFLOW-372

[mik-laj]

Duplicate: #7931