Decide who runs an agent-written production backfill
Your task
An agent has written a backfill that fills a new users.region column for 80 million rows, from a geo-lookup on each user's last known IP address. The code looks good. The agent can run commands in your terminal, where you are signed in to everything you need for on-call this week. It offers to run the backfill now.
What do you let the agent run, and what do you run yourself? Decide where the lines are and what must be true before the real run.
It covers
- Risky change
- Checkpoints
- Data migration