Load-Test Report — Durable Path (Redis + Postgres + mock FMS)

Status: measured (local Docker soak)
Date: 2026-07-11T05:08:27.803Z
Topology: Redis redis://127.0.0.1:16380 · Postgres postgresql://waspada:***@127.0.0.1:15433/waspada_soak · Streams + outbox poller · mock FMS
Ingest: 50 req/s × 20s per fail-rate cell · OUTBOX_POLL_MS=500 · OUTBOX_CLAIM_LIMIT=50 · OUTBOX_BACKOFF_BASE_MS=2000 (soak-only; prod default 60s) · drain 45000ms while gateway alive after each cell

Matrix

Fail rate	Throughput	p50 ms	p99 ms	non-2xx	alerts	audit_events	processed_events	outbox succeeded	dead_lettered	mock FMS
0	29.0	793	1988	0	600	2418	600	600/609 (98.5%)	0	mock ok=600 fail=0
0.1	36.4	700	1799	0	747	3009	747	747/757 (98.7%)	0	mock ok=747 fail=91
0.2	31.6	755	2028	1	650	2625	650	650/660 (98.5%)	0	mock ok=650 fail=193
0.5	34.4	728	1860	0	708	2919	708	680/718 (94.7%)	0	mock ok=680 fail=679

How to read this

• alerts / processed_events ≈ durable write + idempotency ledger under the Streams/in-process pipeline.
• audit_events grows with ingest + alert_created + delivery_* kinds (1 INSERT per event today).
• outbox succeeded / total is the recovery rate after the in-process drain window (poller still running). At high fail rates, failed/dead_lettered should rise; succeeded should climb as retries land during drain.
• Ingest p99 here includes durable encrypt+INSERT path (and Redis store). It is not comparable 1:1 to the memory-only baseline (13/64 ms).
• Durability–latency trade-off: durable-path p99 reflects the cost of guaranteed delivery (Postgres write + Redis Streams publish + outbox commit). Memory-only path is 64 ms p99. This is a measured design cost, not a performance bug.
• Backoff note: production OUTBOX_BACKOFF_BASE_MS defaults to 60s. This matrix uses 2s so retry recovery is observable inside the drain window.

Re-run

docker compose -f apps/gateway-api/docker-compose.soak.yml up -d
pnpm --filter @waspada/gateway-api soak:durable

# Single cell
LOAD_TEST_MOCK_FMS_FAIL_RATE=0.2 pnpm --filter @waspada/gateway-api load-test -- --durable --mock-fms

Known gaps (unchanged)

• No XAUTOCLAIM PEL sweep (B5)
• Audit writes are per-event INSERTs (batch before harder load-test)