Benchmark Harness

Measuring GitWire's production performance — webhook RPS, queue throughput, API p95 latency.

Overview

GitWire ships with a built-in benchmark harness at scripts/benchmark.js that measures real production performance. It runs against a live GitWire instance (local or remote) and reports:

• API latency — per-endpoint p50/p90/p95/p99 response times
• Webhook ingestion — throughput for synthetic push events
• Queue throughput — BullMQ job enqueue rate
• Mixed workload — realistic combination of reads and writes

Usage

# Full benchmark suite
npm run benchmark

# Individual benchmarks
npm run benchmark:api
npm run benchmark:webhook

# Custom parameters
node scripts/benchmark.js --iterations 500 --concurrency 20

# Custom target
GITWIRE_BASE_URL=http://localhost:3000 node scripts/benchmark.js --api

CLI flags

Flag	Description	Default
--api	API latency benchmark only	—
--webhook	Webhook ingestion only	—
--queue	Queue throughput only	—
--iterations N	Requests per benchmark	200
--concurrency N	Max concurrent requests	10
--api-key KEY	API key for auth	API_KEY env

Production numbers

Benchmarks run against gitwire.erlab.uk via Cloudflare Tunnel (RTT ~130ms).

Health endpoint (no rate limit)

Metric	Value
p50	134 ms
p95	143 ms
p99	146 ms
RPS	7.5

Webhook ingestion (no rate limit on /webhooks)

Metric	Value
p50	138 ms
p95	146 ms
p99	152 ms
RPS	7.3

API endpoints (rate-limited: 100 req/min)

Endpoint	p95	Notes
GET /api/repos	162 ms	Heavy query with CI stats
GET /health	143 ms	Lightweight
POST /webhooks/github	146 ms	Signature verify + enqueue

Mixed workload

Metric	Value
p50	139 ms
p95	156 ms
p99	168 ms
RPS	7.1

Performance breakdown

The ~135ms baseline is dominated by the Cloudflare Tunnel round-trip (~120ms). Actual Express processing time is 5–30ms on top of that:

Client → Cloudflare Edge → Tunnel → Express → PostgreSQL → Response
  ~60ms      ~30ms         ~10ms     ~5-30ms      ~1-5ms

Rate limiting

The production rate limiter is set to 100 requests per minute per IP. The health endpoint (/health) and webhook endpoint (/webhooks/*) are exempt.

When benchmarking, the harness automatically throttles to stay under the limit. For raw performance numbers without the rate limiter, run directly against the Express backend:

# Port-forward to the Docker container
ssh -L 3000:localhost:3000 root@your-server

# Benchmark locally
GITWIRE_BASE_URL=http://localhost:3000 npm run benchmark:api

Output

The harness prints a summary table and saves detailed results to a JSON file:

benchmark-results-2026-05-25T22-41-45Z.json

The JSON file contains per-endpoint stats suitable for tracking over time:

{
  "timestamp": "2026-05-25T22:41:45Z",
  "config": {
    "baseUrl": "https://gitwire.erlab.uk",
    "iterations": 200,
    "concurrency": 10
  },
  "results": {
    "api": { "GET /api/repos": { "p95": "162.2", "rps": "6.9" } },
    "webhook": { "p95": "146.4", "rps": "7.3" }
  }
}

Architecture

graph LR
    A[Benchmark Script] -->|HTTP burst| B[Cloudflare Tunnel]
    B --> C[Express :3000]
    C -->|rate limiter| D{Rate limit?}
    D -->|pass| E[Route handler]
    D -->|429| F[Reject]
    E --> G[PostgreSQL]
    E --> H[Redis / BullMQ]

See also

• Workers — BullMQ queue processing
• Security — rate limiter configuration