perf(load-tests): run K6 baseline and fix search.js URLSearchParams bug

- Fix search.js: replace URLSearchParams (unsupported in K6) with string interpolation
- Add baseline performance report with latency benchmarks across all 4 suites
- Add load-tests/results/*.json to .gitignore (large raw output files)

Note: pre-existing test failure in create-listing.handler.spec.ts (eventBus.publish mock) — unrelated to this change.

Co-Authored-By: Paperclip <noreply@paperclip.ing>

.gitignore

@@ -29,6 +29,7 @@ coverage/
 playwright-report/
 test-results/
 blob-report/
+load-tests/results/*.json
 
 # misc
 *.log

load-tests/results/BASELINE-REPORT.md

@@ -0,0 +1,137 @@
+# K6 Baseline Load Test Report
+
+**Date:** 2026-04-09
+**Environment:** Local dev (macOS, localhost:3001)
+**K6 Version:** v1.7.1
+**API Status:** Running (NestJS dev mode)
+
+## Executive Summary
+
+All four K6 test suites executed successfully. Latency metrics are excellent across all endpoints — well within SLA thresholds. However, functional correctness is impacted by dev environment limitations (auth endpoints return 500 due to database/dependency issues, Typesense unavailable for search).
+
+**Key finding:** The API server handles high concurrency well at the network/framework level. Actual business logic performance needs re-validation once backend dependencies (PostgreSQL + Prisma, Typesense, VNPay) are fully operational.
+
+## SLA Thresholds (from config)
+
+| Metric | Target | Status |
+|--------|--------|--------|
+| p50 latency | < 200ms | **PASS** (all suites < 3ms) |
+| p95 latency | < 500ms | **PASS** (all suites < 6ms) |
+| p99 latency | < 1000ms | **PASS** (all suites < 19ms) |
+| Error rate | < 1% | **FAIL** (auth/listings/payments return 500) |
+
+## Detailed Results
+
+### 1. Auth Suite (auth.js)
+
+| Metric | Value |
+|--------|-------|
+| Peak VUs | 100 |
+| Duration | 2m (ramp 30s → sustain 1m → ramp-down 30s) |
+| Total Requests | 4,572 |
+| Throughput | 37.6 req/s |
+| p50 latency | 2.12ms |
+| p95 latency | 5.24ms |
+| p99 latency | 9.12ms |
+| Error Rate | 100% (HTTP 500 — internal server error) |
+| Data Received | 7.6 MB (62 kB/s) |
+
+**Notes:**
+- Register endpoint (`POST /auth/register`) returns 500 — likely Prisma/DB connection issue in dev
+- Login tests could not run (no users in pool due to failed registration)
+- Latency baseline is solid — the framework + routing layer responds fast
+
+### 2. Listings Suite (listings.js)
+
+| Metric | Value |
+|--------|-------|
+| Peak VUs | 500 |
+| Duration | 3m (ramp 30s → sustain 2m → ramp-down 30s) |
+| Total Requests | 56,347 |
+| Throughput | 311.7 req/s |
+| p50 latency | 0.55ms |
+| p95 latency | 2.37ms |
+| p99 latency | 4.24ms |
+| Error Rate | 100% (HTTP 500) |
+| Data Received | 68 MB (377 kB/s) |
+
+**Notes:**
+- Highest throughput of all suites (311 req/s at 500 VUs)
+- Sub-millisecond p50 indicates very fast error responses
+- Detail endpoint not tested (no listings created in setup)
+- Need to re-test with seeded DB data for realistic benchmarks
+
+### 3. Search Suite (search.js)
+
+| Metric | Value |
+|--------|-------|
+| Peak VUs | 200 |
+| Duration | 3m (ramp 30s → sustain 2m → ramp-down 30s) |
+| Total Requests | 18,743 |
+| Throughput | 103.7 req/s |
+| p50 latency | 1.27ms |
+| p95 latency | 2.59ms |
+| p99 latency | 4.57ms |
+| Error Rate | 0% (503 responses treated as acceptable) |
+| Data Received | 155 MB (856 kB/s) |
+
+**Notes:**
+- Text search: p95 = 2.65ms, Geo search: p95 = 2.55ms
+- HTTP error rate 0% (503 from Typesense is expected in dev without Typesense running)
+- Response validation fails (no `data` field in 503 responses) — expected
+- Fixed `URLSearchParams` K6 compatibility issue during this test run
+
+### 4. Payments Suite (payments.js)
+
+| Metric | Value |
+|--------|-------|
+| Peak VUs | 50 |
+| Duration | 2m (ramp 30s → sustain 1m → ramp-down 30s) |
+| Total Requests | 20 |
+| Throughput | 0.17 req/s |
+| p50 latency | 1.31ms |
+| p95 latency | 5.76ms |
+| p99 latency | 18.77ms |
+| Error Rate | 100% (setup registration failed) |
+| Data Received | 33 kB |
+
+**Notes:**
+- Very low request count (20) because setup failed to register users
+- Without auth tokens, payment VUs exit immediately
+- p99 at 18.77ms is the highest across suites — suggests payment endpoint has more processing overhead
+- Need auth working before meaningful payment benchmarks
+
+## Bottlenecks Identified
+
+1. **Auth/Registration (Critical):** `POST /auth/register` returns 500 — blocks all authenticated test flows. Root cause: likely Prisma client not connecting to PostgreSQL or missing migration state.
+2. **Typesense Unavailable:** Search returns 503 — expected for local dev without Typesense service. Tests correctly handle this with fallback checks.
+3. **K6 Compatibility:** `search.js` used `URLSearchParams` (not available in K6 runtime) — **fixed** by replacing with string interpolation.
+4. **Payment Gateway:** VNPay integration returns errors in dev — expected, tests accept 502/503 as valid.
+
+## Baseline Benchmarks (Framework-Level)
+
+These numbers represent the NestJS framework + routing layer performance (error responses), not full business logic:
+
+| Endpoint Category | p50 | p95 | p99 | Max RPS |
+|-------------------|-----|-----|-----|---------|
+| Auth (register/login) | 2.12ms | 5.24ms | 9.12ms | ~38/s |
+| Listings (search/detail) | 0.55ms | 2.37ms | 4.24ms | ~312/s |
+| Search (text/geo) | 1.27ms | 2.59ms | 4.57ms | ~104/s |
+| Payments (create/list) | 1.31ms | 5.76ms | 18.77ms | ~0.2/s* |
+
+*Payments throughput is artificially low due to setup failure.
+
+## Recommendations
+
+1. **Re-run with full backend:** Set up PostgreSQL + seed data, start Redis, and optionally Typesense before re-running to get business-logic-level benchmarks.
+2. **CI integration:** Wire `load-test.yml` GitHub Actions workflow to run against a staging environment with real data.
+3. **Performance regression alerts:** Set thresholds in CI — fail the pipeline if p95 > 500ms or error rate > 5%.
+4. **Grafana dashboards:** Export K6 results to Grafana Cloud or InfluxDB for historical tracking.
+5. **Increase payment VUs:** Once auth works, test payments at 100+ VUs to find gateway bottlenecks.
+
+## Files Generated
+
+- `load-tests/results/auth.json` — Raw K6 JSON output
+- `load-tests/results/listings.json` — Raw K6 JSON output
+- `load-tests/results/search.json` — Raw K6 JSON output
+- `load-tests/results/payments.json` — Raw K6 JSON output

load-tests/scripts/search.js

@@ -50,13 +50,9 @@ export default function () {
   if (iter % 2 === 0) {
     // Full-text search
     const query = TEXT_QUERIES[iter % TEXT_QUERIES.length];
-    const params = new URLSearchParams({
-      q: query,
-      limit: '20',
-      offset: '0',
-    });
+    const url = `${BASE_URL}/search?q=${encodeURIComponent(query)}&limit=20&offset=0`;
 
-    const res = http.get(`${BASE_URL}/search?${params.toString()}`, {
+    const res = http.get(url, {
       tags: { name: 'GET /search (text)' },
     });
 
@@ -71,14 +67,9 @@ export default function () {
   } else {
     // Geo search
     const geo = GEO_SEARCHES[iter % GEO_SEARCHES.length];
-    const params = new URLSearchParams({
-      lat: String(geo.lat),
-      lng: String(geo.lng),
-      radius: String(geo.radius),
-      limit: '20',
-    });
+    const url = `${BASE_URL}/search/geo?lat=${geo.lat}&lng=${geo.lng}&radius=${geo.radius}&limit=20`;
 
-    const res = http.get(`${BASE_URL}/search/geo?${params.toString()}`, {
+    const res = http.get(url, {
       tags: { name: 'GET /search/geo' },
     });