ตัวให้บริการตามขอบเขตคำขอและข้อแลกเปลี่ยน

By default every NestJS provider is a singleton: one instance is created at bootstrap and shared across the whole application lifetime. This is fast and memory-efficient, and it is the right choice for the overwhelming majority of services.

But some scenarios need per-request state. In a multi-tenant backend you might want a provider that already knows which tenant the current request belongs to, so you do not pass the tenant id through every method call. NestJS solves this with injection scopes.

• Scope.DEFAULT — singleton (the default)
• Scope.REQUEST — a new instance per incoming request
• Scope.TRANSIENT — a new instance for each consumer that injects it

You opt into request scope by passing { scope: Scope.REQUEST } to the @Injectable() decorator. NestJS will then instantiate a fresh copy of this provider for every HTTP request (or message, in microservices/WebSocket transports).

Because a new instance exists per request, it is safe to store request-specific mutable state on it — concurrent requests never share the same object.

A request-scoped provider can inject the underlying request object using the REQUEST token. This is the canonical way to read headers, the resolved user, or a subdomain to determine the tenant.

Only providers that are themselves request-scoped (or transient) may inject REQUEST. Trying to inject it into a singleton is a design error — the request does not exist yet when a singleton is built.

This is the single most important consequence to internalize: scope bubbles up the injection chain.

If a controller or service injects a request-scoped provider, that consumer also becomes request-scoped — even if you never marked it that way. The effect cascades transitively up every provider that depends on it.

• A request-scoped TenantContext injected into OrdersService makes OrdersService request-scoped.
• If OrdersController injects OrdersService, the controller is instantiated per request too.

One small request-scoped leaf can quietly turn a large subtree of your app non-singleton, which has real performance implications.

Here neither OrdersService nor the controller declares a scope, yet both become request-scoped purely because of the transitive dependency on TenantContext. NestJS resolves the effective scope by taking the narrowest scope in the chain.

Keep this graph shallow: the deeper a request-scoped provider sits, the more of your tree it drags into per-request instantiation.

Request scope is not free. For every request, Nest must:

• Walk the dependency subgraph and build a fresh instance of each request-scoped (and inheriting) provider.
• Run their constructors and any lifecycle hooks on every request.
• Garbage-collect those instances after the request ends.

Under high throughput this adds measurable latency and GC pressure. The official guidance is blunt: use request scope sparingly. The more providers turn request-scoped via bubbling, the larger the per-request allocation cost grows.

A subtle gotcha: for request-scoped providers, lifecycle hooks like onModuleInit are not available, but request-scoped instances do go through construction on each request. Heavy setup in a constructor therefore runs on every single request.

If your provider opens a connection, parses a token, or hydrates config in its constructor, multiply that work by your request rate. Move expensive, request-independent setup into a singleton and inject it.

For performance-critical request-scoped providers, Nest lets you avoid inheriting the full Request object payload. Passing { scope: Scope.REQUEST } still injects a lightweight context.

In some transports (e.g. GraphQL, microservices) the REQUEST token actually resolves to the execution context, not an HTTP request. Write your provider to read only what it needs, and guard for the shape differences across transports rather than assuming Express.

For multi-tenant apps where you would otherwise make many providers request-scoped, Nest offers durable providers. The idea: instead of one instance per request, keep one instance per tenant and reuse it across that tenant's requests.

You define a ContextIdStrategy that maps a request to a stable sub-tree id (e.g. the tenant id). Nest then caches durable sub-trees keyed by that id, dramatically cutting instantiation cost while preserving per-tenant isolation.

Outside the normal injection flow — for example inside a singleton that occasionally needs a request-scoped provider — you cannot just inject it (that would bubble the singleton). Instead resolve it manually against the current context id using ModuleRef.

You must pass the request's ContextId so Nest returns the instance bound to that specific request, not a brand-new orphan instance.

Reach for request scope only when per-request state is genuinely needed and cannot be passed as a parameter. Common safer alternatives:

• AsyncLocalStorage (Node's node:async_hooks) to carry request context without making providers request-scoped — the whole graph stays singleton.
• Method parameters — just pass the tenant id explicitly.
• Durable providers — when you need per-tenant isolation at scale.

This snippet shows the AsyncLocalStorage pattern, which keeps services as fast singletons while still exposing per-request data.

Key takeaways for using request scope safely:

• Default to singletons. Request scope exists for genuine per-request state like tenant context.
• Scope bubbles up. One request-scoped leaf turns every dependent provider request-scoped, transitively.
• It costs performance. Per-request construction, lifecycle work, and GC pressure scale with throughput.
• Keep request-scoped providers shallow and cheap. Push heavy, request-independent setup into singletons.
• Prefer alternatives when you can: AsyncLocalStorage for context, explicit parameters, and durable providers for per-tenant reuse at scale.
• Resolve manually via ModuleRef.resolve with the request's ContextId when a singleton must reach a scoped instance.