Advanced Features (Cluster, HA, Ceph, SDN)¶
Eight.ly OS includes several enterprise-grade multi-node features, described under Settings > Advanced. This isn't a beta in the "might not work" sense — HA failover (fence → relocate → Rehome), cross-node CephFS migration, and tunnelled live migration have all been proven in multi-node drills on real hardware. They're grouped here because they're advanced capability aimed at multi-node setups, not because they're unproven.
Settings > Advanced also has a genuinely experimental section, unrelated to clustering: Nova's bigger, GPU-only models. Those carry their own warning and stay off until you opt in — see Nova.
Multi-Node Clustering¶
Manage multiple Eight.ly OS servers as a single cluster.
- Create clusters — designate a primary node and form a cluster
- Join/leave — add or remove nodes from the cluster
- Heartbeat monitoring — nodes send heartbeats to detect failures
- Cross-node API proxy — manage any node's resources from any other node's dashboard
- Unified management — see all VMs, containers, and storage across nodes in one view
Use case¶
Run a media server on one node, AI workloads on a GPU node, and backups on a third — manage them all from one dashboard without switching between browser tabs.
Ceph Distributed Storage¶
Ceph provides distributed, redundant storage across multiple nodes.
- Cluster initialization — set up a Ceph cluster across your nodes
- OSD management — add and manage Object Storage Daemons (one per disk)
- Pool configuration — create pools with configurable replication levels
- CephFS — distributed filesystem for shared storage across nodes
- Monitor management — manage Ceph monitor daemons
Use case¶
Three or more nodes with direct-attached storage that you want to present as a single, redundant storage pool. Overkill for most homelabs, but powerful for larger setups.
Rehome¶
Rehome — high availability that moves a failed node's workloads to a survivor and brings them home when the node recovers. Automatic failover (HA) when a node goes down.
- Background health monitoring — Eight.ly watches node health continuously
- Automatic resource migration — VMs and containers move to healthy nodes on failure
- Affinity rules — control which resources prefer which nodes
- Fencing — isolate a failed node to prevent split-brain scenarios
- Start/stop/migrate — manual Rehome resource management
Use case¶
Critical services (Home Assistant, Nextcloud, databases) that need to keep running even if one server fails.
Storage Replication¶
Replicate data between nodes for disaster recovery.
- Rsync-based — efficient delta sync between nodes
- Configurable schedules — replicate hourly, daily, or on custom cron
- Per-volume — choose which storage volumes to replicate
Software-Defined Networking (SDN)¶
Advanced networking for multi-node setups.
- Zone management — define network zones (simple, VLAN, QinQ, VXLAN, EVPN)
- VNet configuration — create virtual networks within zones
- Controller management — configure SDN controllers
- IPAM — IP Address Management for automated IP assignment
- VNet firewall — per-VNet firewall rules
- Fabric management — configure the underlying network fabric
Use case¶
Complex multi-VLAN environments where you need software-defined network segmentation across multiple nodes.
Resource and Directory Mappings¶
Map physical hardware to VMs for direct passthrough.
- PCI device passthrough — pass GPUs, NVMe controllers, or other PCIe devices directly to a VM
- USB device passthrough — pass USB devices (Zigbee sticks, hardware keys) to VMs
- Directory passthrough — mount host directories into VMs, from its own Directory Mappings tab
Use case¶
Pass your NVIDIA GPU to a Windows VM for gaming, or pass a Zigbee USB stick to a Home Assistant VM.
Where these live¶
Every feature above has its own tab under Infrastructure > Datacenter (Cluster, Ceph, SDN, Rehome, Replication, Resource Mappings, Directory Mappings) — see Datacenter Management. The Settings > Advanced page is where you'll find a plain-English description of each and the toggle for Nova's experimental models. Toggling a switch there is audit-logged.
Some features need more than one node to do anything useful (Ceph wants 3+ nodes with dedicated storage devices; SDN wants cluster mode). On a single-node box, their tabs are there but there's nothing for them to manage yet.