# Router chaos testing

Purpose
- Validate backpressure, recovery, and cache failure behavior for the router.

Test categories
- Load testing with spike scenarios (baseline, 10x, 50x, recovery).
- Backpressure verification for 429 and 503 with Retry-After.
- Recovery tests to ensure queues drain quickly.
- Valkey failure injection with graceful fallback.

Expected behavior
- Normal load returns 200 OK.
- High load returns 429 with Retry-After.
- Critical load returns 503 with Retry-After.
- Recovery within 30 seconds, zero data loss.

Metrics
- http_requests_total{status}
- router_request_queue_depth
- request_recovery_seconds

Alert cues
- Throttle rate above 10% for 5 minutes.
- P95 recovery time above 30 seconds.
- Missing Retry-After headers.

CI integration
- Runs on PRs touching router code and nightly staging runs.
- Stores results as artifacts for audits.

Related references
- operations/router-rate-limiting.md
- docs/operations/router-chaos-testing-runbook.md