Skip to content

BENCHMARKS.md

Latest commit

 

History

History
50 lines (37 loc) · 1.91 KB

BENCHMARKS.md

File metadata and controls

50 lines (37 loc) · 1.91 KB
Raw

Benchmarks

news-api includes a deterministic local benchmark so performance claims can be reproduced without a live GNews key, provider quota, or public-internet latency.

The benchmark:

  1. Builds the TypeScript app.
  2. Starts a local GNews-compatible fake provider.
  3. Starts news-api with GNEWS_BASE_URL pointed at the fake provider.
  4. Measures cold unique searches and warm cached searches.
npm run benchmark:local

Useful overrides:

BENCHMARK_PROVIDER_DELAY_MS=25 \
BENCHMARK_COLD_REQUESTS=500 \
BENCHMARK_COLD_CONCURRENCY=25 \
BENCHMARK_WARM_REQUESTS=1000 \
BENCHMARK_WARM_CONCURRENCY=100 \
npm run benchmark:local

Latest Local Run

Run this on your machine before publishing numbers in a profile or portfolio; benchmark results depend on host CPU, Node version, and local load.

Command: npm run benchmark:local

  • Date: 2026-06-24T00:07:47.186Z
  • Commit: 07a171e
  • Node: v26.0.0
  • Host: Darwin 25.5.0 (Apple M1)
  • Fake provider delay: 15ms
  • Rate limiting: disabled for benchmark
Scenario Requests Concurrency Upstream calls Success Failed p50 ms p95 ms p99 ms Throughput req/s
Cold unique searches 300 20 300 300 0 28.6 54.2 67.7 611
Warm cached search 500 50 0 500 0 9.1 101.8 180.0 2484

Metric Notes

  • Cold unique searches use a different query per request, so every successful request should call the upstream provider.
  • Warm cached search performs one warmup request, then measures repeated identical searches. The warmup request is excluded from the table, so upstream calls should be 0 during the measured run.
  • Rate limiting is disabled by the benchmark process because the goal is service/cache behavior, not limiter behavior.
  • The fake provider adds a configurable delay (BENCHMARK_PROVIDER_DELAY_MS, default 15) to make cache behavior visible.