Tenant Compute #
The tenant compute layer provides resources for application workloads, experiments, and user-facing services. Unlike the management plane, tenant compute is designed for higher churn and experimentation.
Tenant Compute Components #
| Component | Purpose |
|---|---|
| Tenant Hypervisors | Proxmox hosts for VM-based tenant workloads |
| Raspberry PIs | Edge/IoT compute for specialized workloads |
Characteristics #
Tenant compute infrastructure is:
- Tolerant of rebuilds — Expect frequent provisioning and teardown
- Fast change cadence — Experimentation and iteration encouraged
- Terraform-managed — Declarative VM lifecycle (future)
- VLAN-isolated — Tenant traffic separated from management plane
Workload Types #
| Category | Examples |
|---|---|
| Application development | IoT backend, services, APIs |
| Experiments | Test environments, sandboxes |
| Ephemeral workloads | Short-lived or rebuildable VMs |
| Edge/IoT | Sensor collection, local processing |
Separation from Management Plane #
Tenant compute is deliberately separated from management plane infrastructure:
| Aspect | Management Plane | Tenant Compute |
|---|---|---|
| Change cadence | Slow, deliberate | Fast, experimental |
| Blast radius | Must be minimized | Tolerable |
| Rebuild tolerance | Low — avoid rebuilds | High — expect rebuilds |
| Provisioning | Ansible | Terraform (future) |
This separation ensures that tenant experimentation cannot impact substrate stability. If a tenant VM misbehaves or a tenant hypervisor fails, the management plane remains operational.
Tenant Networking #
Tenant networks are EVPN/VXLAN overlays owned by the tenant hypervisor’s own SDN fabric — one VRF per tenant, an anycast gateway hosted by the fabric, and addressing from fabric IPAM. The core router is the perimeter for the aggregate transit network only.
See ADR-0001 and the tenant fabric implementation.