pbs-zabbix

Zabbix monitoring for Proxmox Backup Server (PBS). Alerts when a scheduled backup did not complete successfully — including the case where it never even started (e.g. the network path between the hypervisor and the PBS datastore was down).

How it works

pbs-zabbix-backup-check reads Proxmox Backup Server task logs directly from /var/log/proxmox-backup/tasks. Those files are world-readable, so the tool needs no root privileges and does not call the PBS API.

It supports three modes, wired into Zabbix Agent 2 via zabbix/pbs-backup.conf:

• discovery — Zabbix low-level discovery (LLD): lists {#VMID} for every VM/CT that has at least one recorded backup task.
• age <vmid> — seconds since the last successful (TASK OK) backup of vmid. Returns a large sentinel value if no successful backup was ever recorded, so a staleness trigger fires immediately.
• status <vmid> — the last recorded task line for vmid (success or failure), for alert context.

Installing

apt install pbs-zabbix
systemctl restart zabbix-agent2

Zabbix setup

The package installs a ready-to-import template at /usr/share/pbs-zabbix/zabbix/template_pbs_backup.yaml. On the Zabbix server: Data collection → Templates → Import it, then link it to the PBS host. It provides:

• A discovery rule (pbs.backup.discovery) with item prototypes pbs.backup.age[{#VMID}] (Numeric, unsigned, units s) and pbs.backup.laststatus[{#VMID}] (Text).

• A trigger prototype that fires when a backup is either stale or simply missing data:

  last(/<host>/pbs.backup.age[{#VMID}])>{$PBS.BACKUP.MAXAGE:"{#VMID}"}
  or nodata(/<host>/pbs.backup.age[{#VMID}],{$PBS.BACKUP.MAXAGE:"{#VMID}"})=1
  

  The nodata() half matters: a plain last() comparison only re-evaluates when new data arrives, so if PBS never even ran the backup (host powered off, network path down, agent unreachable) the item just stops updating and last() silently freezes below the threshold, never alerting. This is not a "PBS host must always be up" check — a host that is normally powered off and only wakes for its nightly backup is expected to report at least once per {$PBS.BACKUP.MAXAGE} window, so it will not false-alarm during its normal off periods; it only fires if a full window passes with no confirmed backup at all.

• A user macro {$PBS.BACKUP.MAXAGE} (default 108000, 30 hours) which covers both the staleness threshold and the nodata() window. Use context overrides ({$PBS.BACKUP.MAXAGE:"<vmid>"}) for VMs/CTs on a different backup cadence (e.g. weekly jobs).

If the PBS host is only powered on intermittently (e.g. nightly), configure zabbix-agent2 there for active checks (ServerActive=<zabbix-server> in zabbix_agent2.conf) rather than relying on the server polling it passively — otherwise the server may simply miss the host's short wake window, which delays data as much as an actual outage would.

Resulting alerts in the Zabbix problem view:

License

MIT — see LICENSE.