P0 through P3 aren't just labels. They're a contract with your stakeholders about response time and escalation. Most teams skip classification entirely and go straight to debugging. That's how P1s turn into P0s.
The scheduler crashed. Or the DAG is stuck. Or the executor ran out of memory. Here's the ordered checklist for diagnosing Airflow failures fast — drawn from a decade of late-night incidents.
Before you start debugging, you need to know what's downstream. Most engineers jump straight to root cause. That's why they often fix the pipeline but miss the backfill that three dashboards needed.
Your upstream team renamed a column. Your pipeline doesn't know yet. This is the story of how schema drift accounts for 15% of all data incidents and how to build a runbook that handles it gracefully when it happens.
If your new hire can't resolve a P0 incident using only the documentation in their first month, that's not a problem with the hire. It's a problem with the documentation. Here's how to fix it.
A runbook that was accurate eight months ago but references a deprecated tool and an engineer who left is worse than no runbook. It gives false confidence. Three practices that keep runbooks from going stale.