Skip to content

App lifecycle

A project directory contains one backend.fast source file. The CLI sends its source over HTTPS for a bounded remote build.

Terminal window
bfast deploy --name my-counter --tier small ./my-counter
bfast apps list --json
bfast inspect --json my-counter

The platform assigns your account as owner. Creation returns a stable app ID and a hostname such as amber-apex-a1b2c3.tr1.edge.backend.fast. The display name stays separate. Creation waits for public health by default.

Terminal window
bfast deploy --wait=false --name another-counter --tier medium ./my-counter
bfast operations --json '<request ID>'
bfast operations --wait '<request ID>'

Mutations print an operation ID. The default wait bound is fifteen minutes. A command interrupted after admission can still complete. Repeat the same command to reuse its pending request ID. Inspect that operation before using --new-request. A changed body with the same request ID fails.

The service bounds build admission and worker capacity. Overload can reject admission explicitly. A failed build does not silently select another runtime.

Terminal window
bfast update --app my-counter ./my-counter
bfast inspect --json my-counter

Normal update preserves the app ID, URL, tier, retained volume and acknowledged data. Keep the original DSL app name and compatible model/private schemas. Unsupported schema changes fail explicitly; this is not a general migration tool. Validate an authenticated read after the update.

Terminal window
bfast delete my-counter

Completed deletion removes the app from the active catalog and closes its URL with HTTP 410. The result reports retained_data: true. The data volume remains. Saved R2 objects remain charged. Deleted deployment names are not reusable. There is no public permanent purge or automatic restore workflow yet.

The current runtime keeps one writable pod and retained volume per app. K3s owns pod recovery. The platform preserves stop/delete intent and does not create extra writable replicas. High availability and replacement-node recovery remain separate work. A Ready pod alone does not prove public routing; the CLI also checks public HTTPS health.