One rust rate limiter for in-memory and distributed workloads.
Trypema gives you one API for in-process, Redis-backed, and hybrid sliding-window rate limiting. Start with a single service, then scale to distributed systems without switching libraries or changing mental models.
[dependencies] trypema = "1"
use trypema::{RateLimit, RateLimitDecision, RateLimiter}; let rl = RateLimiter::builder().build().unwrap(); let rate = RateLimit::try_from(10.0).unwrap(); match rl.local().absolute().inc("user_123", &rate, 1) { RateLimitDecision::Allowed => { /* proceed */ } RateLimitDecision::Rejected { retry_after_ms, .. } => { eprintln!("retry in {retry_after_ms}ms"); } RateLimitDecision::Suppressed { .. } => unreachable!(), }
Three providers, one model
Pick the deployment shape you need today. The provider API stays familiar as you move from one machine to many.
Local
In-process state with the lowest latency and the least operational overhead. Best for single services, workers, CLIs, and tests.
Redis
Best-effort distributed limiting with one Redis round-trip per call. Use it when multiple instances must share limits directly.
Hybrid
Local fast-path plus periodic Redis sync. This is the highest-throughput distributed option when per-request Redis I/O is too expensive.
Two strategies
Choose the admission style that matches your failure mode.
Absolute
Deterministic sliding-window limiting. Requests are either allowed or rejected, with best-effort retry hints on rejection.
Suppressed
Probabilistic shedding near or above capacity. Instead of an abrupt cliff, suppression increases as pressure rises.
What Trypema is built for
Fractional rates
Rate limits are f64, so 0.5 req/s and 5.5 req/s are first-class inputs.
Sliding windows
Smooth request accounting without fixed-window resets. Bucket coalescing lets you tune precision versus overhead.
Background cleanup
Cleanup can start automatically from build() or manually through run_cleanup_loop() when you construct from explicit options.
Runtime-aware Redis support
Enable exactly one of redis-tokio or redis-smol. Redis-backed providers require Redis 7.2+.
Start here