Skip to main content
nestrs is designed to bring NestJS’s modular architecture to Rust without sacrificing Rust’s performance or Axum’s speed. Every controller and route decorator compiles directly into native Axum router registrations and Tower service layers at startup. There are no per-request controller heap allocations, no runtime reflection, and no dynamic dispatch in the request-handling hot path.

​
The Overhead Philosophy

When evaluating web framework performance in Rust, comparing different frameworks with mismatched middleware, HTTP runtimes, and serialization layers often measures setup differences rather than the framework itself. The metric that matters most is the overhead of nestrs on top of bare Axum:
  • What is the latency delta between a handler registered directly on an axum::Router versus through nestrs?
  • How much overhead does the default production middleware stack introduce?
To answer this objectively, nestrs includes dedicated overhead benchmarks in nestrs/benches/axum_overhead.rs that benchmark the exact same handler (GET /ping -> "ok") across three configurations within the same process.

​
The Three Measurement Tiers

All measurements use in-process Router::oneshot invocations to eliminate TCP socket overhead and OS network jitter:

​
Benchmark Results

On modern x86_64 and Apple Silicon hardware, the dispatch delta between bare Axum and default nestrs routing is typically under 50 nanoseconds:
The difference between bare Axum and nestrs_get_ping_stack_off represents the essential safety layers: panic catching and payload guards. The routing dispatch itself adds effectively zero measurable latency.

​
Microbenchmarks Tracked in CI

In addition to the Axum overhead suite, every change to nestrs is tested against strict regression thresholds across four hot paths:

​
1. Router Hot Path

Tests URI resolution and handler invocation with route parameter extraction (GET /v1/bench/ping/:id).
  • Benchmark: router_get_v1_bench_ping
  • Allowed regression threshold: ≤ 3.0%

​
2. Middleware Pipeline

Measures execution through the complete guard, pipe, and interceptor sequence.
  • Benchmark: router_get_v1_bench_work_middleware_stack
  • Allowed regression threshold: ≤ 3.0%

​
3. Dependency Injection Resolution

Measures cached singleton service resolution from the ProviderRegistry. Singleton resolution uses a lock-free OnceLock path.
  • Benchmark: provider_registry_get_cached_singleton
  • Allowed regression threshold: ≤ 5.0%

​
4. Validated JSON Deserialization

Measures deserialization and validation of incoming JSON request payloads through ValidatedBody<T>.
  • Benchmark: router_post_v1_bench_signup_validated_json
  • Allowed regression threshold: ≤ 5.0%

​
Running Benchmarks Locally

You can run Criterion benchmarks locally with Cargo:
To compare against baselines and export structured JSON reports, use the CI load script: