LoadBudget
@forge/monorepo / backend/src / LoadBudget
Type Alias: LoadBudget
LoadBudget =
object
Defined in: backend/src/loadtest/metrics.ts:253
Properties
errorRate
readonlyerrorRate:number
Defined in: backend/src/loadtest/metrics.ts:257
The error rate an operator would accept. Refusals are not errors.
minCompletionRatio
readonlyminCompletionRatio:number
Defined in: backend/src/loadtest/metrics.ts:269
How far completions may fall short of offered load before the step is not sustained.
The check latency alone does not make. A step can post an excellent p99 while completing a third of what was offered, because the work that never got picked up contributes no latency sample at all — the fast runs are measured and the queued ones are invisible. My first version of this had exactly that hole and reported "sustainable 20/s, p99 5ms" for a step whose queue reached 101 jobs and whose real throughput was 6.9/s.
0.9: completions must be within ten percent of offered load. Not 1.0, because the drain window makes the ratio slightly noisy at low rates and a knife-edge threshold would make the envelope flap.
p99Ms
readonlyp99Ms:number
Defined in: backend/src/loadtest/metrics.ts:255
The p99 an operator would accept. Above it, the step is outside the envelope however much completed.