Hi everyone,

I’m currently working on Issue #44003, which aims to expand the reference
points for the new Deadline Alerts feature in Airflow 3.1.

In my initial discussions with the issue maintainer (Ferruzzi), we noted
that there has been significant refactoring around Serialized DAGs
recently. Ferruzzi raised a valid point regarding whether
SerializedDagLastUpdatedDeadline is still a relevant or useful reference
point in the current architecture. I’d love to hear from the community: is
"DAG metadata freshness" still a practical trigger you'd use, or should we
deprioritize this in favor of more high-demand areas?

Following that thought, Ferruzzi also suggested that there is a growing
need for Task-level deadlines. To ensure we address the gaps left by the
legacy SLA system, I’m interested in hearing about your specific use cases
for TaskInstance monitoring.

Are there specific lifecycle events—such as time spent in a queued state or
delays in a task's start_date—that you would like to see as standardized
reference points? Your feedback will help shape a potential follow-up
proposal for a more exhaustive list of Task-level deadlines.

Thanks & Regards,

Ashish Rana

Reply via email to