Deployment Topology

:::info Source Sourced from services/authoring-service/12-DEPLOYMENT_TOPOLOGY.md in the documentation repo. :::

1. Containers

Container	Purpose
authoring-api	REST API for drafts, modules, lessons, blocks
authoring-collab	WebSocket Yjs awareness server (M5)
authoring-worker	Outbox relay, publish-saga coordinator, SCORM import worker
authoring-ai-worker	AI co-author request fan-out (via AIClient)

2. Scaling Rules

Container	Min	Max	HPA Trigger
authoring-api	3	30	CPU > 60% or active-editors > 100/pod
authoring-collab	2	20	active WS connections > 500/pod
authoring-worker	2	10	outbox backlog > 5000
authoring-ai-worker	2	15	AI request queue > 200

• Collab server sticky-session by draftId hash (consistent hashing).
• API stateless; collab stateful via Redis-backed Yjs awareness coordination.

3. Resource Requirements

Workload	CPU req/limit	Mem req/limit
authoring-api	500m / 2000m	512Mi / 1.5Gi
authoring-collab	500m / 2000m	1Gi / 4Gi (Yjs docs in memory)
authoring-worker	200m / 1000m	256Mi / 1Gi
authoring-ai-worker	300m / 1500m	512Mi / 2Gi

4. Caching & Storage

Layer	Contents	TTL
Redis (per region)	Draft locks, Yjs awareness state, AI result cache	sliding
Postgres pgbouncer	Transaction-mode pool	—
S3/R2	Draft snapshots (for forensics), SCORM import staging	90 days
CDN	Block previews, thumbnails	5 min

5. CDN Usage

• Preview URLs for draft blocks (read-only, signed, TTL 10 min).
• SCORM import package staging (owner-only signed URL).

6. Regional Deployment

• Per-region: US, EU, ME, AP.
• Drafts live in tenant's homeRegion.
• AI calls route through regional ai-gateway → fall back to local model on budget exhaustion.

7. Service Mesh

• mTLS within cluster.
• Egress to media-service, content-service, ai-gateway, catalog-service.
• WebSocket connections through regional ingress (websocket-aware LB).

8. Release Strategy

• Blue/green for API.
• Collab server: drain sessions (notify clients 5 min before) → recreate.
• Workers: rolling; outbox is idempotent.

9. Topology Diagram

 Browser ─ HTTPS ─▶ CDN ─▶ API GW ─▶ authoring-api (3–30)
    │                                     │
    │                                     ├─▶ Postgres (drafts)
    │                                     ├─▶ Redis (locks, AI cache)
    │                                     ├─▶ ai-gateway-service
    │                                     ├─▶ media-service
    │                                     └─▶ NATS (outbox → publish saga)
    │
    └─ WSS ──▶ authoring-collab (2–20, sticky) ─▶ Redis (Yjs awareness)

 Publish saga steps are coordinated by `authoring-worker` consuming
 `content.play_package.built.v1` + `catalog.course_version.published.v1`.

10. DR

• RPO 5 min, RTO 60 min.
• Draft snapshots hourly to cold storage.
• Live collab sessions rebuildable from event log + last snapshot.