Compute & Storage

Compute & Storage

Proxmox cluster layout, Site A ZFS storage, Site B Ceph cluster, and PBS cross-site backup replication.

Compute and storage split by site role: Site A anchors management workloads with ZFS on sa-stor-01; Site B runs Ceph across five nodes for distributed block storage and Kubernetes. Two independent Proxmox clusters: sa-pve and sb-pve: anchor each site. Proxmox Backup Server (PBS) at each site replicates cross-site for disaster recovery.

Proxmox Clusters

One Cluster per Site: No Cross-WAN Proxmox Clusters

One Proxmox cluster per site: sa-pve at Site A, sb-pve at Site B. Never stretch a Proxmox cluster across WAN or WireGuard. Corosync requires LAN-like latency; a WAN/VPN cluster risks quorum instability and split-brain.

SiteClusterNodesCreate on
Site Asa-pvesa-edge-01, sa-cmp-01, sa-cmp-02, sa-stor-01sa-stor-01
Site Bsb-pvesb-edge-01, sb-cmp-01, sb-cmp-02, sb-cmp-03, sb-cmp-04, sb-cmp-05sb-cmp-01

Cluster Only After Final IPs and /etc/hosts

Do not create or join a cluster until every node in the site has its final 10.x.20.x Proxmox Management IP and /etc/hosts is correct. Never join nodes across sites. Never run Corosync over WireGuard or WAN.

See Proxmox Clusters for the full creation and join procedure.

Site A: ZFS Storage

Site A storage runs on sa-stor-01 (Supermicro 5049A-T), which hosts ZFS mirror vdevs, PBS-A, DNS VMs, monitoring, and core databases. ZFS is chosen over Ceph here because Site A is management-focused: simpler administration, strong data integrity, and native snapshots for PBS storage pools.

DetailValue
Hostsa-stor-01
Drives8-12 × Samsung SM863 1.92 TB SSDs
LayoutZFS mirror vdevs
Services hostedPBS-A, sa-dns-01/sa-dns-02, monitoring, databases

See Site A ZFS for pool layout, vdev configuration, and PBS dataset setup.

Site B: Ceph Storage

No Stretched Ceph

Site B Ceph stays local. Do not stretch Ceph across sites. Use PBS replication for cross-site disaster recovery instead.

Site B runs Ceph across all five compute nodes. Ceph public/client traffic uses VLAN 60 and cluster replication uses VLAN 65; both are L2-only with no SVI or default gateway. Consumers attach directly to VLAN 60. PBS cross-site synchronization uses routed VLAN 90 instead.

HostHardwareCeph Role
sb-cmp-01SYS-5019D-4C-FN8TPMON, MGR, OSD
sb-cmp-02SYS-5019D-4C-FN8TPMON, MGR, OSD
sb-cmp-03SYS-5018D-FN4TOSD, K8s worker
sb-cmp-04SYS-5018D-FN4TOSD, K8s worker
sb-cmp-05SYS-5018D-FN4TOSD, K8s worker

Plan: 4-6 enterprise 1.92 TB SSDs per node; replication size 3. Ceph is provisioned via the Proxmox ceph-tentacle no-subscription repo (Ceph release: tentacle, decided 2026-06-28).

See Site B Ceph for OSD layout, CRUSH map, and Kubernetes RBD integration.

Backup & Replication

PBS runs at each site as a dedicated service VM. Each instance backs up its local Proxmox cluster, and the two instances replicate to each other for cross-site disaster recovery.

InstanceProxmox mgmt IPBackup-data IP
PBS-A (Site A)10.10.30.2010.10.90.40
PBS-B (Site B)10.20.30.2010.20.90.40

Cross-Site Replication Is Required

Do not rely on one site for all backups. PBS cross-site replication is a design requirement: Site A and Site B each hold backup copies of the other.

See PBS Backups for backup job configuration, datastore layout, and cross-site replication setup.