The Bike Shed 477: Change Management
Oct 7, 2025
Joël and Aji dive into the complexities of change management and data migrations. They explore the risks involved with data changes versus code changes, highlighting the permanent nature of corrupted data. Incremental migrations and the Strangler Fig pattern offer strategies for smooth transitions. The importance of planning zero-downtime changeovers is emphasized, along with handling in-flight requests during migrations. Joël shares insights from past migration experiences, while discussing innovative solutions like ID exchange endpoints and canary deploys.
AI Snips
Chapters
Transcript
Episode notes
Plan For Zero-Downtime Or Controlled Failure
- Decide early whether a change needs zero downtime or can tolerate brief failures.
- Design the cutover behavior and user-visible fallback before you start the technical work.
Idempotency Means Safely Retriable
- Idempotency means safe to retry, not necessarily transactional rollback.
- Make migrations and background jobs rerunnable so partial runs don't corrupt state.
Treat Data Migrations As High Risk
- Treat data changes as higher risk than code changes and plan accordingly.
- Avoid irrevocable data writes during transition windows unless you can recover or roll forward safely.