FA-73772 / Rate limiter algorithms / Member archive

EWMA arrival-rate throttle: simultaneous arrivals treated as one millisecond apart · case 02

Two requests with the same timestamp spike the estimate to 1000 requests per second.

Member previewVariant 2 · 3 implementations · 6 checks per implementation

Case contract

Input {tau_ms, limit_rps, requests [t_ms]}. The rate estimate r starts at 0. For each arrival strictly after the latest seen time, dt = t - last, a = exp(-dt/tau) and r = a*r + (1-a)*(1000/dt) requests per second. Simultaneous or late arrivals leave r unchanged and never move the clock backwards. Every arrival reports deny if r > limit else allow, with r rounded to 3 decimals.

Why this case matters

Irregular-interval EWMA estimators throttle clients by smoothed request rate; time-unit and duplicate-timestamp handling decide whether bursts are detected.

One recorded failure

Sample boundary fixture

This sample comes from the broken implementation of a controlled reproducer.

Boundary fixture	Actual	Expected	Outcome
simultaneous arrivals	`[["allow", 0.0], ["allow", 1.106], ["allow", 2.354], ["allow", 3.6], ["allow", 3.901], ["deny", 5.146]]`	`[["allow", 0.0], ["allow", 1.106], ["allow", 1.106], ["allow", 1.106], ["allow", 1.964], ["allow", 1.964]]`	Failed

MEMBER ARCHIVE

The complete case is available to members.

This record includes three runnable implementations, regression fixtures, execution results, and source hashes.

Member access is invitation-based. Sign in with your invited account to inspect the sources.