Monitor the Whole Proxmox Cluster, Not One Node at a Time

Point the agent at the Proxmox API with a read-only token and it reports the entire cluster - every node, guest and storage pool - not just the machine it runs on. Install it on more nodes and the extras become redundant reporters: the most complete one is authoritative, and if it goes quiet another takes over automatically, with no duplicate rows.

That changes what an alert means. A cluster losing quorum opens one incident, not one per reporting node. An offline node opens one incident for that node. A stopped guest opens one incident for that guest, and it keeps the same identity if the guest migrates to another node.

The agent reads each guest's most recent backup and its age from the Proxmox node - across every storage it can see, local, NFS and PBS-backed alike - so you can answer the question that actually matters: has every critical VM and container had a successful backup in the last N hours? A guest past that age - or with no successful backup at all - raises its own alert, and any guest that sits in no backup job is flagged so it cannot slip through unnoticed.

It stays honest about what it cannot see: a storage the token cannot audit reads unknown rather than a false "never", and the backup-stale alert freezes instead of clearing while visibility is lost - so a missed read never looks like a fresh backup.

What you see. Each datastore shows its usage and PBS's own estimated-full date, plus garbage collection. Each backup group shows its newest backup, its size, whether it is encrypted and whether it passed verification. Each sync, verify and prune job shows its last run, and charts keep the history. A group the latest collection did not read shows as last known, never as overdue, and a usage PBS did not report reads Unknown, never zero.

To connect it, add the backup server from the Proxmox area or from an instance's settings, and pick the agent that reads it: the one on the PBS host, or any other that reaches its API, with a read-only token limited to Datastore.Audit and Remote.Audit (PBS's built-in Audit role is too wide, and the agent refuses it). That answers what the Proxmox VE backup view cannot: is the latest backup intact, is the offsite copy still syncing, when will the datastore be full - and what about the machines Proxmox VE never sees, which back up to the same server.

What alerts. With the one-click Proxmox Backup Server Needs Attention workflow, an incident opens when the backup server can no longer be read. A failed verification or a filling datastore is shown, not alerted on. For verify, sync, prune and garbage-collection results, point PBS's own webhook notification target at a Fivenines task per job: success to the task's ping URL, errors to its /fail URL, and a job that has run before and then stops trips its task's schedule. A Task Failed workflow turns an error into an incident.

See which processes on the Proxmox host consume the most resources. Identify runaway QEMU processes, unexpected CPU consumers, or memory-hungry services competing with your VMs.

Six Proxmox triggers cover the cluster from top to bottom: quorum lost, node offline, guest stopped, guest backup stale, storage pool inactive, and a per-guest metric threshold on CPU, memory, disk read/write or network in/out. All six triggers ship as one-click templates.

Node alerts fan out per node, so a cluster losing two nodes opens two incidents, each naming the node it is about - you know what to look at before you open anything. And putting the hosts into maintenance freezes that cluster's alerts rather than firing a false recovery when the work finishes.

Alerts go where your team works: email, Slack, Microsoft Teams, Telegram, Discord, Pushover, PagerDuty or webhooks.