sla_miss.execution_date is incorrect for daily dagruns

Apache Airflow version

2.2.4 (latest released)

What happened

When inspecting metadata in airflow tables about sla stuff found that

• in case of daily DAG’s schedule execution_date attribute for sla_miss is filled with unexpected value. Seems like data_interval_end
• in case of hourly schedule sla_miss.execution_date are as expected (equals to DAGRun’s data_interval_start/logical_date/execution_date)

this prevents to create any custom SLA analisys dashboards/metrics because unable to join sla_miss and dag_run/task_instance in direct and clear way

clicking on daily task’s link on SLA Miss dashboard forwarding to task instance which was never executed (because it is in “future”) and have no runtime metrics in Task instance details

All screenshots are taken @ 2022-03-25 17:xx:xx Z

What you think should happen instead

sla_miss.execution_date should be same as dag_run.execution_date in all cases

How to reproduce

setup a dag with any daily schedule setup an SLA on task wait sla miss to trigger observe in SLA Miss dashboard - incorrect incorrect execution date

Operating System

amazon linux

Deployment

Official Apache Airflow Helm Chart

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

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

[uranusjr]

We should advertise 2.4 as “SLA now actually works!”

[potiuk]

Actully - If you are much into it and REALLY want to dig into it and see if you can incrementally improve it - I think that might be a super-valuable contribution to the community 😃. So if you have the will, I am happy to help with reviews and comments 😃. Just ping me in the PRs.