Migration Playbooks
Big-bang migrations fail. Prefer strangler, dual-run, measurable milestones, and rollback that still works on Friday night.
Workflow
- Current vs target state; non-goals.
- Risks: data, downtime, client compatibility.
- Strategy: strangler, expand/contract, blue/green, dual-write.
- Milestones with exit criteria and owners.
- Dual-run validation and comparison checks.
- Rollback and freeze windows.
- Comms to users/internal teams.
- Cleanup and decommission checklist.
Output format
## Migration: <name>
**Strategy:** …
### Phases
### Validation
### Rollback
### Comms
### Cleanup
Rules
- Data migrations need dry-runs and checksums.
- Feature flags for traffic shifting when possible.
- Do not delete old path until metrics prove parity.
- Budget time for unknown unknowns.
- Explicit owner for each phase.
Edge cases
- Zero downtime: expand/contract schema first.
- Client SDKs: version skew matrix.
- Org migration: training and runbooks.