reliability.md

Reliability Profile

Test Layers

• Unit: logic and mapping behavior
• Integration: sequencing across peers, rooms, signaling/media interactions
• Full: runtime process behavior and websocket/netsocket integration

Primary signaling suites:

• 0-peer: media server registration lifecycle
• 1-peer: full peer lifecycle and media start/stop/leave
• 2-peer: peer discovery + fanout + control path
• 3-peer: fanout correctness beyond pairwise interactions
• 10-peer: timing variance and churn
• 100-peer: high-cardinality sequencing and branch behavior
• multi-region: regional routing and capacity behavior across configured region pools (tests use local/local-2)
• room-fanout-scale: 99 peers across 33 rooms

What Reliability Means Here

A run is considered healthy when:

• no protocol invariant violations are thrown
• expected join/attach/fanout transitions converge
• cleanup paths complete without stale transport/producer mappings
• diagnostics provide actionable failure reasons for user-impacting issues

Current Limits (Practical)

• baseline coverage targets room control and signaling correctness, not media quality scoring
• stress tests focus on control-plane correctness under concurrency and churn (not media traffic)
• browser/device-specific codec/network variance is not fully modeled in integration tests

Runbook Commands

• Full repo check path:
  • make test
• Signaling only:
  • npm --prefix signaling run test:unit
  • npm --prefix signaling run test:integration
  • npm --prefix signaling run test:full
• Media only:
  • npm --prefix media run test:unit
• Target one signaling integration suite file:
  • npm --prefix signaling exec tsx --test tests/integration/signaling.zeroPeerLifecycle.test.ts
  • npm --prefix signaling exec tsx --test tests/integration/signaling.userImpactFailures.test.ts

Related Docs

• Signaling happy-path matrix
• Signaling error taxonomy