Before: 3 pods ran ENABLED_SERVICES=all — gateway, auth, and every
business-logic service co-hosted. This defeated the purpose of the
gateway pattern and prevented independent scaling.
After:
sor-gateway (2 replicas) — ENABLED_SERVICES=gateway,auth
• WebSocket edge, accepts client connections
• Auth co-located (JWT validation is a localhost gRPC call, no network hop)
• Lighter resource footprint: 150m/128Mi request, 500m/512Mi limit
• Ingress at hub.aeternalisgames.com → sor-gateway:80
sor-workers (3 replicas) — ENABLED_SERVICES=chat,collection,bazaar,
content,versioning,matchmaking,game
• Subscribes to NATS, handles all business-logic messages
• Holds DB connection pools for all schemas
• Internal only — no ingress
• Heavier resources: 300m/384Mi request, 1000m/1Gi limit
configmap.yaml: removed ENABLED_SERVICES=all and Services__Auth (each
deployment now sets these directly in its own env block)
ingress.yaml: hub.aeternalisgames.com backend sor-host → sor-gateway
ArgoCD prune=true will delete the old sor-host Deployment + Service
from the cluster on next sync.
Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)
Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>