The Upgrade Nobody Wanted to DoPipeline · hard · ~15 minYour Airflow runs a version that hits end-of-support in 10 weeks. The upgrade crosses a major boundary: several operators are removed, the scheduler behavior changed, and 200 DAGs — some untouched for two years, their authors gone — must survive it. Leadership's stance swings between 'why risk touching what works' and 'just upgrade it over a weekend'.Both extremes lose. Design the migration that doesn't bet the platform.Evidence

Assessment

Constraints

Checkpoints

Why is 'don't touch what works' the riskier position here?

The migration architecture that fits 200 mixed-ownership DAGs?

What do you do about the 12 ownerless DAGs and the 34 on removed operators — and how do you keep leadership committed for the full 10 weeks? Interview-style.