resilixLoad limiting for JavaScript

A dependency that has slowed to a crawl is still "up". resilix measures latency, works out how much concurrency your upstream can actually absorb, and sheds the rest — before the errors start. Zero dependencies, no I/O.

Get started

Why it exists

GitHub

Trips on slow, not just broken

An upstream that degrades from p50 0.35s to p50 10.4s with a flat error rate is invisible to a failure-rate breaker. resilix trips on slow-call rate — a dimension no other JavaScript breaker has.

Three trip conditions

Adaptive concurrency limiting

Works out how many concurrent calls your upstream can absorb, from latency alone, and sheds the excess proportionally. Vegas queue estimation over an O(1) P² quantile estimator.

How it adapts

Verdicts, not booleans

"Did the promise reject?" cannot express that a 404 is healthy and a 429 is backpressure. Six verdicts, one settled call, read differently by each policy.

The verdict model

Hedging and retry budgets

Both previously locked inside @grpc/grpc-js and the AWS SDK. Hedge at the measured p95, cap retry amplification at 1.1x, shed low-criticality work first.

Hedging and criticality

One line from opossum

The compatibility shim passes 362 of 362 of opossum's own test suite, run unmodified. Default behaviour stays opossum's, so swapping the import changes nothing on day one.

Migration guide

Runs everywhere, proven in CI

Node 18/20/22/24, Bun, Deno and Cloudflare Workers — verified every push by importing the built artifact and driving a real breaker, not asserted in a README.

Design commitments