The NestJS Request Context Problem: Request-Scoped DI vs AsyncLocalStorage (ALS) — a Practical Production Blueprint (2026)

This is not a “theory post”. It’s a practical guide you can apply in a real NestJS codebase — with verified sources and a clear decision framework.

“Request context” is everything you want to access deep in services without threading it through every method: current user, tenant, permissions, request ID, trace IDs, transaction handle, locale, feature flags, and more.

In NestJS, you usually end up with one of these outcomes:

- You pass context explicitly as parameters everywhere (reliable, but noisy and hard to maintain).

- You use REQUEST-scoped providers (simple API, but it can turn a big chunk of your app into per-request instantiation). [1]

- You use AsyncLocalStorage-based context (clean call-sites, keeps providers singleton, but needs correct setup and awareness of edge cases). [2][3]

NestJS clearly warns about it: request-scoped providers add overhead because instances are created per request, and — more importantly — scope “spreads” through dependencies. [1]

Two quotes that matter when you’re designing architecture:

> “Any provider that relies on a request-scoped provider automatically adopts a request scope, and this behavior cannot be altered.” [1]

> “Using request-scoped providers will have an impact on application performance... it will still have to create an instance of your class on each request.” [1]

This is how real apps get into trouble: a single “request-scoped datasource / logger / context” becomes a root dependency, and suddenly controllers + services + repositories are recreated per request. NestJS even calls out multi-tenant apps as a common scenario where this happens. [1]

Yes, Nest mentions the impact might be “~5% latency-wise” in a properly designed app — but the real cost is usually architectural: it’s harder to reason about lifetime, caching, and the “shape” of your DI tree. [1]

Node.js AsyncLocalStorage is essentially “thread-local storage for async code”: it lets you associate state with an execution chain (promises/callbacks) and read it later without passing parameters. [3]

Node’s docs recommend AsyncLocalStorage as the preferred approach over rolling your own async_hooks-based implementation, describing it as performant and memory-safe. [3]

But it’s not magic. Context loss can still happen in “rare situations”, especially around callback-based APIs or custom thenables — Node recommends promisifying or using AsyncResource to bind execution context correctly. [3]

So the question becomes: can we reliably wrap the Nest request lifecycle so everything downstream sees the same store? NestJS answers: yes — via middleware / lifecycle entry point. [2]

NestJS explicitly positions ALS as a way to propagate state without passing parameters and as an alternative to REQUEST-scoped providers “and some of their limitations”. [2]

Core idea: wrap the request early (middleware) with als.run(store, () => next()). Then any provider can read from ALS store later.

Minimal NestJS-style example (simplified from the official recipe): [2]

This is the “clean core” pattern. Real production setups add: safe header parsing, response header echo, trace correlation, queue propagation, and testing.

In production systems, request context is not only about “current user”. The most common drivers are observability and incident response: you need logs, traces, and metrics to correlate reliably across services.

That’s exactly what OpenTelemetry context propagation is about: moving trace/span context across process and network boundaries (defaulting to W3C Trace Context headers like traceparent). [6][7]

Two practical takeaways from enterprise reality:

- You can keep “in-process” context in ALS, but across service boundaries you must propagate it explicitly (HTTP headers, message metadata, etc.). [6][7]

- Don’t put sensitive data into cross-service context propagation. OpenTelemetry explicitly warns against putting secrets/PII into baggage-like propagated key-values. [6]

A 2026 operational note you should not ignore if you rely on ALS heavily (directly or via APM/tracing): Node.js shipped a mitigation for an async_hooks/ALS-related denial-of-service edge case in January 2026 and recommends updating patched versions. Major APM and OTel teams published mitigation guidance and context around it. [4][5]

Below is a high-signal comparison of packages that real teams use to solve NestJS request context. The goal is not “who has more stars”, but: correctness, ergonomics, integrations, and how well it avoids REQUEST scope.

Quick comparison table

Now the details with sources:

- nestjs-cls positions itself as a continuation-local storage module for NestJS built on AsyncLocalStorage, listing use cases like request ID tracking, multi-tenancy, and propagating transactions without explicit parameter passing. [8]

- nestjs-pino documents a key argument many maintainers repeat: avoiding REQUEST-scoped providers due to performance drops, using AsyncLocalStorage for request-bound loggers instead. [11]

- Nest’s official recipe acknowledges ALS as a viable alternative to REQUEST scope — but also states NestJS does not provide a built-in abstraction itself, meaning you either implement it or use a library. [2]

If your context needs are broad (multi-tenancy + transactions + proxy providers + “request context where REQUEST scope is not supported”), nestjs-cls is often the most complete toolbox. Its docs emphasize a wide range of use cases and explicitly mentions avoiding clunky REQUEST-scoped approaches. [8]

A frequent reason teams pick it: they want transaction propagation across services without passing a transaction object everywhere, plus a supported ecosystem of plugins around this pattern. [8]

If you’re already deeply invested in that ecosystem, switching away later is usually not worth it — but if you only need 2–3 fields (requestId, userId, tenantId) you may want something smaller.

@pas7/nestjs-request-context is directly aligned with this problem: providing request context via AsyncLocalStorage while keeping your DI tree singleton, with a strong focus on type-safety via typed keys and clean ergonomics for NestJS. [10]

What makes it particularly relevant to this article:

- Typed ContextKey<T> for context values (less stringly-typed “magic keys”). [10]

- NestJS-oriented ergonomics: decorators (e.g. parameter decorators) so you can read context without plumbing. [10]

- HTTP adapters for Express and Fastify in NestJS environments. [10]

- It explicitly positions itself against the REQUEST-scope approach from a performance/architecture standpoint. [10]

Caveat (honesty check): the repo notes that using the Fastify adapter outside NestJS is limited due to Fastify + AsyncLocalStorage incompatibilities, so you should treat it as NestJS-first. [10]

This section is the “do it right” checklist: where to initialize context, what to store, how to propagate to queues/microservices, and how to test it so you don’t ship context leaks.

Recommended context fields (practical default)

- requestId: stable correlation ID; echo back as x-request-id in response.

- userId: internal user identifier (avoid email/phone).

- tenantId: if multi-tenant.

- traceId / spanId: if you run tracing (or derive from OTel context).

- authLevel / role: if you need it for authorization decisions (prefer explicit checks in critical flows).

Pattern A: Minimal DIY ALS (good if you want full control)

Use the Nest recipe’s middleware approach and implement a small wrapper (with tests). [2]

Pattern B: Use nestjs-cls when you need the ecosystem

If you need transaction propagation plugins and proxy provider patterns, nestjs-cls is a strong choice. [8]

Pattern C: Use @pas7/nestjs-request-context when you want strict typed keys + lean surface area

If your need is a clean, typed request context (requestId/userId/tenantId + a few more), and you value a NestJS-first adapter layer, PAS7’s library is a sensible fit. [10]

Below are illustrative patterns based on the library’s approach (typed ContextKey<T>, Nest-first ergonomics, adapters). Always verify exact API names against the repo before locking in production usage. [10]

1) Define typed keys (one module/shared package)

2) Initialize in middleware (HTTP entry point)

3) Read context in services (no param threading)

The core win: your services stay singleton and clean, while still accessing per-request context. [10]

ALS does not magically cross process boundaries. If you publish a job to a queue, or call another service, you must propagate context explicitly (like requestId/trace context). [6][7]

A good default is to propagate:

- x-request-id (your correlation ID)

- traceparent (W3C Trace Context) if you have distributed tracing. [6][7]

HTTP outbound example (pseudo-pattern)

Queue job example (pseudo-pattern)

When consuming the job, initialize a fresh ALS context and restore the propagated fields before executing business logic.

If you choose request-scoped consumers for jobs, Nest notes that they create a new instance per job — similar trade-offs to HTTP request scope. [13]

Context bugs are notorious because they look fine locally but break under concurrency.

Your minimum testing strategy should include:

- Parallel request tests: two simultaneous requests must not see each other’s context.

- “Async boundary” tests: delayed tasks (setTimeout, message handlers) must retain context (or intentionally not retain it, if you design it that way). [3]

- Queue propagation tests: publish → consume restores context explicitly. [6][7]

PAS7’s library includes a testing story in-repo (and positions itself with a testkit approach); if you go DIY, you must build equivalent guardrails. [10]

If you want a fast decision without rereading everything:

Request context is one of those “small” architectural decisions that determines whether your codebase stays clean at 30 endpoints or collapses at 300.

If you’re building a NestJS product and want help with backend architecture, observability, or automation — PAS7 Studio can help design and implement production-grade systems.

Read more: https://pas7.com.ua/blog

All links below are directly relevant to the topic and were used to ground the factual claims and comparisons above.

We publish practical, citeable articles on web engineering, automation, security, and product development — focused on decisions that actually matter in production.