Deploy Strategy

⛔ Dockerfile ≠ Container Apps. A Dockerfile that only serves static files (nginx, httpd, COPY . /usr/share/nginx/html) is NOT a backend app. Route static-only Dockerfiles as static sites per service-mapping.md § Static Dockerfile sites, not Pattern C.

Additional routing:

Pattern A: Oryx Auto-Build (Default)

Languages: Node.js, Python, .NET, Go, Java, PHP, Ruby

Oryx detects the stack from project manifests, installs dependencies, and builds automatically during zip deploy.

Still write deployStrategy to prepare-plan.json — even though Oryx auto-detects, the startup command and app settings MUST be in Bicep at scaffold time (not generated at deploy time). This eliminates imperative CLI commands during deploy.

Read prereq-output.json.entryPoint for the app’s start file — do NOT re-read manifests. Build the startup command from package.json start script (Node.js) or framework convention (Python gunicorn, .NET/Go/Java Oryx-native).

⛔ Do NOT set a custom appCommandLine for Pattern A. Let Oryx use package.json start script or framework defaults natively. A custom appCommandLine (cd /home/site/wwwroot && node {entryPoint}) replaces the Oryx launcher entirely — the launcher handles node_modules.tar.gz decompression, and bypassing it causes MODULE_NOT_FOUND crashes. Only set appCommandLine when initCommands[] has required: true entries (migrations).

⛔ TypeScript projects: Verify typescript + @types/* are in dependencies (not devDependencies) — Oryx production mode skips devDeps, causing tsc build failures.

⛔ When initCommands[] has required: true entries: Set startupCommand to prepend migrations: "cd /home/site/wwwroot && {initCommand} && {framework-default-start}". Migrations are idempotent — safe on every cold start. Otherwise, omit startupCommand entirely (let Oryx handle it).

Scaffold encodes startupCommand → Bicep appCommandLine, and requiredAppSettings → Bicep siteConfig.appSettings. Deploy only does: wait → zip → health check.

Pattern B: Startup-Install (Native Modules)

When native modules are detected, Oryx may fail to compile them. The startup-install pattern provides a two-layer safety net.

Two-Layer Strategy

1. Primary — Oryx zip build: SCM_DO_BUILD_DURING_DEPLOYMENT=true + ENABLE_ORYX_BUILD=true tells Oryx to run dependency installation during the Kudu-side zip deploy. The Kudu build environment on App Service Linux has gcc, make, and build tools available, so native compilation CAN succeed here.

2. Fallback — startup-install command: appCommandLine runs dependency installation on first container boot IF the dependency directory doesn’t exist. The existence guard ensures it only runs when needed — subsequent restarts skip it because /home is persistent storage.

Both layers are set in Bicep at scaffold time. The startup command is insurance — not the primary mechanism.

Why two layers? az webapp deploy --type zip uses the OneDeploy API, which may not trigger Oryx even with SCM_DO_BUILD_DURING_DEPLOYMENT=true. The startup command catches this case. If Oryx DID build successfully, the guard skips the redundant install.

Deploy Strategy Schema

Write to prepare-plan.json.deployStrategy:

Replace startupCommand and reason with language-specific values from the entry point table below.

Entry Point & Startup Commands

Same as Pattern A, but Node.js adds a dependency guard: if [ ! -d node_modules ]; then npm install --production; fi before the start command.

⛔ Inline commands only. Never generate a .sh startup script file — CRLF causes bash exit code 2. ⛔ Python: do NOT use venv in startup commands.

SKU Implications

When f1Viable: false (any of: native modules, TypeScript build, large deps, WSGI/ASGI server), F1 is not viable — use B1 (~$13/mo) minimum. Surface at approval gate: “⚠️ {f1BlockReason}. B1 minimum required.”

Container Timeout

WEBSITES_CONTAINER_START_TIME_LIMIT controls how long Azure waits for the container to start responding.

Always set to 1800 when codeDeployPattern == "startup-install".

Pattern C: Container-Only

When the component has a Dockerfile with backend logic, route to Container Apps. The Dockerfile IS the deploy strategy — no deployStrategy needed in prepare-plan.json.

The deploy phase handles: ACR build → image push → Bicep redeploy with real image. See code-deployment-container-apps.md.

For Java apps using Jib (build.gradle + com.google.cloud.tools.jib), the build produces a container image without a Dockerfile — push to ACR via jib task.