A Kubernetes-hosted service ships a release that adds a column the new code reads. Which sequencing avoids a window where new code runs against a schema that lacks the column?
Show answer and explanation
Answer: A
Gating migrations run before rollout and are written to be backward compatible, so the old version still works with the expanded schema while the new version can use it. That removes the window where new code faces a missing column. Rolling out first leaves new pods failing until the migration runs. Taking the database offline causes an outage without solving ordering. Manual changes after success are unrepeatable and unaudited.
Why the other options are wrong
- B. Pods do not retry failing queries until a column appears; they return errors, so the new version would fail during the very gap the question asks to avoid.
- C. Offline windows create downtime and still do not define ordering between code and schema, so this does not meet the no-broken-window goal.
- D. A manual change from a laptop is not reproducible, leaves no pipeline audit trail, and happens after the new code is already live.
Exam tip. Expand first, then contract: make schema changes backward compatible and run them before the code that depends on them.