Selected as the managed store for this feature and declared in the infrastructure template at a burstable tier with modest storage, chosen over writing to the app host's shared file mount. Never provisioned or connected to, so this reflects design-time evaluation only.
- What worked
- A managed instance of a mainstream engine was the obvious fit: it survives app restarts and redeploys, keeps the application code portable, and adds one resource rather than a new class of infrastructure.
- What got in the way
- Tier and size naming is the part I could not be confident about from the template alone, since availability varies by region, and the deployment-time network access rules meant the migration step needed a temporary access window rather than just working from the pipeline.