Compute & Storage

PBS Backups & Replication

Proxmox Backup Server instances, IPs, VLAN 90 allocation, and the cross-site replication strategy for disaster recovery.

Each site runs Proxmox Backup Server (PBS) for local VM backups and cross-site disaster-recovery replication. PBS-A serves Site A and PBS-B serves Site B; neither site is the sole backup target.

No Stretched Ceph: Use PBS Replication for Cross-Site DR

Site B Ceph stays local to Site B. Do not stretch Ceph across sites. Cross-site disaster recovery uses PBS replication, not stretched Ceph.

PBS Instances

Each PBS instance has two network interfaces: a management interface on VLAN 30 (VM Services) and a backup-data interface on VLAN 90 (Backup / Replication).

InstanceSiteMgmt IP (VLAN 30)Backup-data IP (VLAN 90)
PBS-ASite A10.10.30.2010.10.90.40
PBS-BSite B10.20.30.2010.20.90.40

PBS-A runs on sa-stor-01 (the Supermicro 5049A-T), co-located with ZFS storage, the Site A DNS VMs (sa-dns-01 / sa-dns-02), monitoring, and core infrastructure VMs. PBS-B runs as a service VM within the sb-pve cluster; the specific Proxmox node that hosts the PBS-B VM is not pinned in the current design.

VLAN 90: Backup / Replication

Backup and replication traffic flows on VLAN 90, separate from management (VLAN 20), Corosync (VLAN 25), and general VM services (VLAN 30).

VLANNameSite ASite B
90Backup / Replication10.10.90.0/2410.20.90.0/24

The .40 address within each /24 is reserved for the PBS backup-data interface (10.10.90.40 for PBS-A, 10.20.90.40 for PBS-B). This follows the address-block convention: .40-.49 is the band for infrastructure service VMs.

Backup Strategy

Both PBS instances run local backups at their own site. PBS data is replicated cross-site for disaster recovery so that a full site failure does not result in complete data loss.

  • Run local backups at each site independently.
  • Replicate PBS data cross-site for disaster recovery.
  • Do not rely on one site for all backups.

Local-First Backup

Each PBS instance operates locally within its site. Local backup jobs continue regardless of the inter-site WireGuard tunnel state.

The cross-site PBS replication path uses the WireGuard tunnel (10.255.0.0/24) that connects sa-fw-01 and sb-fw-01. See WireGuard for tunnel configuration.

Operating Contract

The source design fixes the failure boundary and traffic path without claiming an implementation schedule that has not been chosen.

AreaDecided posture
Local operationEach site runs backups to its local PBS instance; a WireGuard outage does not stop local backup jobs.
Cross-site DRPBS data is replicated between sites so neither location is the only recovery copy.
Network pathBackup data uses routed VLAN 90 and crosses the site-to-site WireGuard tunnel.
Storage boundarySite B Ceph remains site-local; PBS replication is the cross-site protection mechanism.
PBS-A placementPBS-A runs on sa-stor-01.
PBS-B placementPBS-B is a service VM in sb-pve; its specific Proxmox host is not pinned.

Open Operating Decisions

Backup Policy Is Not Yet Complete

The architecture sources do not define retention, prune or garbage-collection cadence, synchronization schedule or direction, datastore encryption and credentials, recovery objectives, or restore-test frequency. These remain open design work rather than implicit defaults.

Open itemCurrent status
Retention policyNo keep-last/daily/weekly/monthly values are decided.
Prune and garbage collectionNo cadence or maintenance window is decided.
Cross-site synchronizationReplication is required; job direction, order, bandwidth window, and retry policy are not specified.
Credentials and encryptionThe authentication, key custody, and datastore encryption procedure is not specified.
RPO / RTONo recovery-point or recovery-time objective is committed.
Restore verificationNo recurring restore-test procedure or evidence cadence is committed.
PBS-B host and datastoreThe service VM address is reserved, but its hosting node and backing datastore are not pinned.