Roadmap
Scale-Out Plan
Multi-node scaling strategy for Chelar.
Current State
Single Hetzner AX42 bare metal server. Can handle ~50-100 concurrent tenants depending on memory usage patterns.
Scale-Out Strategy
Full details in docs/SCALE_OUT_ROADMAP.md.
Key decisions:
- Nomad multi-node — add Nomad client nodes, keep single server
- JuiceFS is already shared — S3 backend means any node can mount tenant data
- Caddy stays single — reverse proxy on the primary node
- No node pinning — tenants can schedule to any node with capacity
Triggers for Scaling
- Memory utilization consistently above 80%
- More than 50 concurrent running containers
- Tenant boot time degraded (Docker pull contention)