Who asked: Dana Whitlock, head of finance at Marigold Grocers (a fictional grocery chain).
Why: Store revenue reaches finance by spreadsheet, a day late and sometimes twice. Dana wants one nightly pipeline that loads each day's sales into the warehouse, reports the total, survives the supplier's flaky price API, and tells the data team when something breaks or runs late.
The data: marigold.py (daily store sales and the warehouse), fernway.py (the supplier's price API, which times out on the first calls of the day) and chat.py (the team channel). The warehouse connection marigold_warehouse is set as an environment variable.
Done looks like: marigold_nightly_sales runs at midnight UTC over the previous day; fetches prices with retries; replaces the day in daily_sales so reruns are safe; returns a one-line revenue report; posts to #data-alerts when a run fails and to #oncall when a run is late; and history from 1 to 7 November is backfilled.
In this project you turn a one-task file into the nightly pipeline Marigold's finance team relies on, one step at a time. Each step starts from your previous answer. The workspace clock is 06:00 UTC on 9 November; each check switches the Dag on, so the scheduler runs the midnight that has just passed.