Proxmox and Ceph Cluster
Status: Historical configuration notes consolidated October 2026; verify versions and health on the cluster.
Cluster
The BarnesBears environment has five Proxmox nodes (prox-1 through prox-5).
| Node | LAN | Ceph network | Hardware notes |
|---|---|---|---|
prox-1 |
192.168.10.41 |
192.168.14.41 |
Ryzen 5 5625U mini PC |
prox-2 |
192.168.10.42 |
192.168.14.42 |
Ryzen 5 5625U mini PC |
prox-3 |
192.168.10.43 |
192.168.14.43 |
GMKtec M5 Ultra, Ryzen 7 7730U replacement |
prox-4 |
192.168.10.44 |
192.168.14.44 |
Ryzen 5 5625U mini PC |
prox-5 |
192.168.10.45 |
192.168.14.45 |
GMKtec M5 Ultra, Ryzen 7 7730U |
The original Ryzen 5 systems were documented with a 32 GB RAM maximum and BIOS FP656V507 (2024-09-05). prox-5 has 32 GB RAM, a 512 GB OS SSD and dual 2.5 GbE. The replacement prox-3 used a 512 GB OS disk and 1 TB WD Black SN850X Ceph disk.
Ceph
- Five monitors, one on each node.
- Five OSDs (0 through 4) were recorded.
- Pool
nc-1-pool: size3,min_size 2. - Dedicated
192.168.14.0/24network. - The cluster was reported
HEALTH_OKafter rebuildingosd.2following theprox-3replacement.
Do not assume the historical healthy state describes current health.
Notable incidents
prox-3 replacement
The former prox-3 experienced hangs and unexpected reboots. RAM and SSD replacement did not fix the problem. A GMKtec M5 Ultra replacement was joined to the cluster, and Ceph osd.2 was rebuilt.
prox-1 link negotiation
A previous cable problem limited the prox-1 network link to 100 Mbps. Verify link speed after cabling or switch changes:
ethtool <interface>
Operations
pvecm status
pveversion -v
ceph -s
ceph osd tree
ceph mon dump
ceph osd pool ls detail
Power
The home lab is supported by a whole-house generator. UPS runtime was targeted at approximately two minutes to bridge generator startup, not for prolonged standalone operation. Verify current UPS load, transfer behavior, and shutdown policy before relying on this target.
Backup and recovery
Ceph replication is not a backup. Record Proxmox backup destinations, schedules, retention, restore tests, and disaster-recovery procedures separately; these details have not been verified here.