load-test - SKILL.md Agent Skill

name: load-test description: "Runs k6 performance and load tests against MaaS API and mesh endpoints for x0tta6bl4. Measures P95 latency, throughput, error rates under sustained load. Use when user asks to: run load test, performance test, stress test, k6, benchmark API, measure latency, MaaS performance, нагрузочное тестирование, тест производительности, запусти k6, проверь производительность API."

Load Test

k6-based load testing for x0tta6bl4 MaaS API.

Prerequisites

# Install k6 (if not present)
snap install k6  # Ubuntu
# or
brew install k6  # macOS

# Verify
k6 version

Step 1: Start the API

# Local dev
uvicorn src.core.app:app --host 0.0.0.0 --port 8080

# Or Docker
docker-compose up -d api

Wait for health check:

curl -s http://localhost:8080/health | python3 -m json.tool

Step 2: Run Load Tests

# Run all load tests
bash tests/load/run_load_tests.sh

# Run specific scenario (k6 directly)
k6 run tests/load/maas_load_test.js \
  -e BASE_URL=http://localhost:8080 \
  -e API_TOKEN=test-token-here \
  --vus 10 \
  --duration 30s

Available scenarios in tests/load/maas_load_test.js:

smoke — 1 VU, 1 min (sanity check)
load — ramp up to 50 VU, 5 min
stress — ramp up to 200 VU, 10 min
soak — 50 VU, 30 min (detect memory leaks)

Step 3: Read Results

scenarios: (100.00%) 1 scenario, 50 max VUs, ...

     data_received..................: 2.3 MB  77 kB/s
     http_req_duration..............: avg=45ms  min=12ms  med=38ms  max=890ms  p(90)=95ms  p(95)=142ms
     http_req_failed................: 0.00%   ✓ 0  ✗ 5000
     http_reqs......................: 5000    166/s
     vus............................: 50      min=1  max=50

Pass criteria:

http_req_failed < 1%
p(95) latency < 200ms for read endpoints
p(95) latency < 500ms for write endpoints

Step 4: Identify Regressions

Compare with baseline:

# Save baseline
k6 run tests/load/maas_load_test.js --out json=/tmp/baseline.json

# After changes
k6 run tests/load/maas_load_test.js --out json=/tmp/current.json

# Compare (Python helper)
python3 -c "
import json
b = json.load(open('/tmp/baseline.json'))
c = json.load(open('/tmp/current.json'))
# Look for p95 degradation > 20%
"

Step 5: Profile Slow Endpoints

If specific endpoints are slow:

# Target single endpoint
k6 run -e ENDPOINT=/api/v1/maas/nodes --vus 20 --duration 60s tests/load/maas_load_test.js

Check API logs for slow queries:

docker-compose logs api | grep -E "slow|timeout|>500ms"

Troubleshooting

High error rate (> 5%):

Check API logs: docker-compose logs api --tail=100
Verify auth token is valid
Reduce VU count: --vus 5

k6 not found:

export PATH=$PATH:$(go env GOPATH)/bin
# or use npx k6
npx @k6/k6 run tests/load/maas_load_test.js

Connection refused:

API not started — run curl http://localhost:8080/health
Wrong port — check BASE_URL env var

Reference

tests/load/maas_load_test.js — main load test script
tests/load/run_load_tests.sh — orchestration script