Eve library

Hathor Load Tests

Other

Revision 1

On this page

k6 suites that assert the Hathor worldbuilding SLOs documented in docs/performance/baselines.md §3.

Script Exercises SLO gate
world-load.js POST/GET /api/v1/worlds, /:id, /:id/export create p95 ≤ 400 ms, export p95 ≤ 2,000 ms
narrative-load.js POST /api/quests quest create p95 ≤ 3,500 ms (LLM-backed)
simulation-load.js POST /api/jobs, GET /api/jobs/:jobId (first-frame poll) enqueue p95 ≤ 500 ms, simulation first-frame p95 ≤ 1,500 ms

Run locally#

bash
docker compose -f docker/docker-compose.dev.yml up -d

# World API — CRUD + export
k6 run testing/performance/k6/hathor/world-load.js

# Narrative API — quest creation (LLM)
k6 run testing/performance/k6/hathor/narrative-load.js

# Simulation worker — enqueue + first-frame latency
k6 run testing/performance/k6/hathor/simulation-load.js

Environment variables#

Var Default Purpose
HATHOR_WORLD_URL http://localhost:3600 world-api origin
HATHOR_NARRATIVE_URL http://localhost:3601 narrative-api origin
HATHOR_SIMULATION_URL http://localhost:3602 simulation-worker origin
JWT_SECRET hathor-load-test-secret HS256 secret for load-test tokens
HATHOR_EXPORT_RATIO 0.25 Fraction of world iterations that run /export
HATHOR_SIM_POLL_INTERVAL_MS 50 First-frame poll cadence
HATHOR_SIM_FIRST_FRAME_MAX_WAIT_MS 8000 Abort a simulation iteration after this wait
HATHOR_WORLD_SEED_ID unset Pre-seeded world id to use instead of ad-hoc

What counts as a regression#

A run regresses the Hathor SLO if any thresholds block in the scripts fails — k6 exits non-zero. The numeric budgets match the per-domain section in docs/performance/baselines.md; a budget change requires updating both.

What the tests deliberately don't cover#

  • Merge-request workflows (low traffic in practice — covered by contract tests).
  • Version snapshot/restore loops (storage-bound, tracked separately).
  • Dialog generation (covered under Isis voice-synthesis tests).