Eve library

Bellona Load Tests

Other

Revision 1

On this page

k6 suites that assert the Bellona engine-bridge SLOs from docs/performance/baselines.md §3. Routes match the OpenAPI-owned contract in libs/bellona/client/src/api/generated.ts.

Script Exercises SLO gate
bridge-load.js Full session → build → status poll → export → end-session per iter session start p95 ≤ 250 ms, build start p95 ≤ 500 ms, export start p95 ≤ 400 ms
build-hotpath.js High-rate GET /v1/builds/:id (+ a slice of /logs) against seeded builds status p95 ≤ 150 ms, logs p95 ≤ 500 ms

Run locally#

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

# End-to-end bridge flow
k6 run testing/performance/k6/bellona/bridge-load.js

# Hot-path polling (status/logs)
k6 run testing/performance/k6/bellona/build-hotpath.js

Environment variables#

Var Default Purpose
BELLONA_API_URL http://localhost:3006 Bellona build-api origin
JWT_SECRET bellona-load-test-secret HS256 secret for load-test tokens
BELLONA_HOTPATH_RPS 80 Hot-path sustained RPS
BELLONA_HOTPATH_SEED_COUNT 8 How many builds the hot-path suite seeds before ramping
BELLONA_HOTPATH_LOGS_RATIO 0.15 Fraction of hot-path iterations that also fetch logs

What counts as a regression#

k6 exits non-zero if any thresholds block fails. The thresholds in config.js encode the baselines directly — update both config.js and docs/performance/baselines.md §3 when adjusting.

Why sessions are opened and closed on every iteration#

The bridge-load scenario intentionally spans the full session lifecycle per VU iteration: opening/closing is cheap relative to build start, and keeping one long-lived session per VU hides session-start regressions. The hot-path scenario is the place to pay the amortisation — it seeds fewer ids and hammers the read path.