Infrastructure
Infrastructure intent selects adapters and runtime topology without changing domain meaning. If no explicit infrastructure section exists, the primary backend uses conservative defaults.
Database
Section titled “Database”The current infrastructure model supports relational SQL configuration:
| Dialect | Notes |
|---|---|
| PostgreSQL | Default primary path |
| MySQL | SQL adapter selection |
| MariaDB | SQL adapter selection |
| SQLite | Local or constrained topology |
| Microsoft SQL Server | Supported with an external/proprietary driver requirement |
Document databases are a future seam, not a current interchangeable option. A different data paradigm changes more than a connection string.
| Backend | Intended use |
|---|---|
none |
Default; no cache layer emitted |
inmemory |
Single-process/local cache |
redis |
Shared read-side cache |
Configuration includes TTL. Cache is an optimization, not the authority for product state. Generated behavior must define invalidation or acceptable staleness.
Asynchronous tasks
Section titled “Asynchronous tasks”The task-queue model can enable Celery workers and select a broker:
- Redis;
- RabbitMQ;
- SQS.
Reactions marked asynchronous run through the worker path. Queue use adds retry, idempotency, dead-letter, observability, and deployment responsibilities.
Per-area storage
Section titled “Per-area storage”Infrastructure intent can carry storage choices by area where the generator supports them. The domain remains independent of the adapter; cross-area consistency must reflect real runtime guarantees.
Generated runtime contract
Section titled “Generated runtime contract”The primary backend can emit:
- database session and persistence adapters;
- Alembic setup and migration artifacts;
- startup migration checks;
- liveness/readiness behavior;
- Docker/Compose topology and environment documentation;
- cache and worker adapters when selected;
- integration configuration status.
Managed topology
Section titled “Managed topology”Current managed apps use one Fly machine plus a durable mounted volume per application. This is an operated delivery topology, not a universal infrastructure target. Exported applications can be deployed through another supported environment under the user’s operations.
Scaling checklist
Section titled “Scaling checklist”- Scale stateless application instances only when sessions/files/cache are shareable.
- Size database pools across all instances, not per instance in isolation.
- Use shared object storage before horizontal file-upload scaling.
- Monitor queue age as well as depth.
- Treat online migration capacity separately from steady-state request capacity.
- Capture algorithmic or workload constraints when infrastructure alone cannot solve the problem.
See Incidents and Observability.