Remote Support Start download

Proxmox Datacenter Manager in production: five tips after one year of multi-cluster operation

ProxmoxPDMPraxis
Proxmox Datacenter Manager in production: five tips after one year of multi-cluster operation

When we published our first practical review of the Proxmox Datacenter Manager — PDM for short — about a year ago, the headline was: mature enough for production, but not a vCenter replacement with feature parity. The time since then has confirmed that view, and it has shown where the real difference between “PDM installed” and “PDM is actually useful” lives: in five concrete operational decisions we now make for every multi-cluster environment.

This article is deliberately not an introduction. It is a follow-up for teams that already have PDM connected to several clusters and are looking for the next maturity step. Five tips from our workbench, with the pitfalls we hit ourselves and cleared out of the way.

Tip 1: Set up SSO cleanly before the first side user is created

The most common mistake we see when we take over PDM environments from other providers: local users in PDM, parallel local users on the individual clusters, and somewhere in between an API token whose owner nobody can confidently name. This works for a while, until the first rotation is due, until an admin leaves the company, or until an auditor asks. Then it is cleaned up — and that costs more than a clean SSO integration from day one.

Our recommendation: PDM gets exactly one identity source. Either OIDC against a self-hosted Authentik server or a Microsoft Entra tenant, or LDAP against classic Active Directory. Local PDM users exist only for emergencies — a break-glass account with a hardware token, documented in the safe, never used in daily operations.

Which groups move from the IdP into PDM

In practice, three or four groups have proven themselves: an admin group with full rights, an operator group with read and start rights, a read-only audit group, and optionally a migration group for colleagues who are allowed to move VMs between clusters. More granularity sounds nice on paper but creates maintenance overhead without real benefit. Less granularity leads to everyone being admin eventually — the classic dam break.

The trap: two IdPs in parallel

Things get particularly annoying when PDM binds against two identity sources because “we still have a few users from the old system during the transition”. Users click the wrong login button, end up with an empty session and call support. Better to set a hard cut-off date, migrate, and remove the second source.

Tip 2: Think of roles per cluster — not per PDM

The second tip sounds trivial but is often overlooked: PDM roles and cluster roles are two different things. Someone who is “admin” in PDM does not automatically have deep rights in the individual clusters — and in many setups they should not.

In our standard configuration, PDM gets its own API token with a dedicated role for each connected cluster. For the production cluster, the token is allowed to read and migrate but not to create users. For the DR cluster, it is additionally allowed to restore snapshots. For a customer cluster that we run as a managed service, it is allowed to read and start but not to change configurations — changes continue to run through the direct login, documented in the customer’s change history.

Role matrix as a living document

We maintain the role matrix in a simple Confluence or markdown document next to the network diagram. Rows are clusters, columns are roles, cells are concrete permissions. That sheet decides, not the GUI. If someone “quickly” extends a role, it shows up in the next review — and is either included in the matrix or rolled back.

Technical side effect: clean token rotation

A dedicated token per cluster also means that a rotation affects only one cluster. We rotate yearly or when staff change, document the rotation in Git, and afterwards test a migration between two clusters — that way we make sure the new token works before the old one is revoked.

Tip 3: Migration workflow — the wizard is only half the way

Central cross-cluster migration is one of the most visible features of PDM. In practice, it turns out that the clickable wizard in the GUI is the easy part. The real work sits in the frame around it.

Before the migration: pre-check

Before every non-trivial migration we check four things: is the PVE version on the target the same or newer? Are the storage types compatible — ZFS to ZFS, Ceph to Ceph, or is a deliberate conversion planned? Are the network bridges on the target present with the same VLAN tags? Is enough bandwidth available or do we run into the window where the nightly backups kick off?

During the migration: one window for the GUI, one for the shell

Alongside the PDM wizard we open an SSH session to the target node and watch pvesm status, iostat and the log of the migration task. The GUI shows progress, but the command line shows whether the storage keeps up and whether network throughput drops. Both views together make the difference between “looks good” and “runs stable”.

After the migration: the follow-up

Often forgotten: after a migration, the VM must be moved in the monitoring configuration, backup jobs must point to the new cluster, DNS entries for VM shortnames may need adjustment. We keep a small checklist that runs alongside the ticket. For the actual backup rework, our article on PBS sync jobs between sites is a good reference, because sync chains often need re-planning after a cluster migration.

Tip 4: Alert consolidation — fewer mails, more information

PDM shows open alerts from all connected clusters in its dashboard. That is useful, but it does not replace proper monitoring. We regularly see environments where PDM alerts go out by mail, the clusters themselves also send mails, and a Prometheus Alertmanager runs in the background on top. Triple mail means triple noise — and in the end nobody clicks any more.

One source, several channels

Our policy: every technical alert flows into exactly one consolidation point. Usually that is a Prometheus plus Alertmanager stack or a Checkmk server, depending on what the customer already runs. PDM reports its states there, the individual clusters as well, PBS too. From there, alerting branches into mail, Teams, Slack or SMS — but according to a clear logic, not ad hoc.

PDM dashboard as the second verification

In our workflow, PDM itself remains the second verification stage. When an Alertmanager alert fires, the on-call colleague takes a look at the PDM dashboard and immediately sees whether the alert really hits several nodes or is just a local phenomenon on one host. This split — consolidation in one system, visualization in PDM — noticeably reduces false alerts.

Actively dampen the noise

Not every state is worth an alert. A brief load spike on a node during the nightly backup is normal operation. We keep a list per environment of known, tolerated states with time windows and thresholds. That is maintenance work, but it pays off: an inbox with two relevant alerts per week is worth more than one with twenty irrelevant ones.

Tip 5: Backup status overview — the blind spot gets smaller

The Proxmox Backup Server remains its own system with its own GUI. PDM shows backup states in its overview but no control. That is one of the most common criticisms — and at the same time the point where a little bit of custom work provides the biggest lever.

PBS state as a Prometheus metric

PBS exports metrics via its built-in endpoint. We pull the last successful backup timestamps per datastore, the deduplication rate and the free capacity from there, store them in Prometheus and build a Grafana panel that we embed into the PDM dashboard as an iframe — technically simple, operationally a noticeable win. For a deeper look at backup practice, our article on backup strategy for SMBs sets the frame.

Restore drills are part of the deal

Backup status alone is only half the truth. What counts is a working restore. We run a documented restore test once per quarter — usually a small test VM that comes back from backup, boots and responds to ping. PDM helps indirectly here: the status view shows on which cluster the test VM lands, and central migration later makes it easy to push the VM back.

3-2-1 remains the rule

A second PBS at a different site, ideally with a different hardware generation, and a third copy offline or on tape — that is still the frame we recommend. If you need a reliable storage foundation for this, the TrueNAS configurator covers the right variants; for support of the whole picture we offer TrueNAS support, and for virtualization and backup infrastructure our IT service for virtualization remains the central contact.

What has changed in our view over the past year

A year ago we described PDM as a promising but incomplete tool. Today we would say: PDM is the standard for multi-cluster environments — but only when the five operational decisions above are made consciously. A simple “install and let it run” does not deliver the value PDM promises. Only the frame of SSO, role matrix, migration workflow, alert consolidation and backup status overview turns an additional dashboard into real productivity.

For customers coming from VMware, the message is the same as a year ago: anyone who honestly maps the vCenter features down to what they really use every day will do well with PDM plus the native cluster GUIs. The gaps are not trivial — but what is there is stable and gets more useful each quarter.

Frequently asked questions about practical Proxmox Datacenter Manager tips

Which of these tips saves the most time in daily operations?

In our experience, alert consolidation. As soon as the flood of double and triple mails dries up, attention for the truly relevant alerts rises noticeably. The second biggest lever is the clean migration workflow, because it clearly lowers the error rate during cluster moves.

Is PDM worth it for only two clusters?

Yes, PDM provides clear value from two separate clusters onwards — especially when the second cluster serves disaster recovery. For single-cluster operation, the native GUI remains the better tool.

How often do you rotate the API tokens between PDM and clusters?

By default once per year and additionally whenever an administrator leaves the team or a token has accidentally shown up in a public log file. The rotation is documented in Git and verified with a short test migration.

Which SSO mistakes do you see most often?

Two identity sources in parallel, local PDM users used in daily operations and missing break-glass accounts. These three points have come up repeatedly in customer projects and are the first part of our takeover audit.

Is the alert consolidation approach tied to a specific stack?

No, it works with Prometheus plus Alertmanager just as well as with Checkmk, Zabbix or any other mature monitoring system. What matters is that one place bundles the alerts and that distribution follows a clear logic.

How does PDM fit into the backup strategy when PBS cannot be centrally controlled?

PDM shows the state, not the control. We complement that with exported PBS metrics into Prometheus and a Grafana panel attached to the PDM surface. That way, two tools form one consistent view without one replacing the other.

More on these topics:

Need IT consulting?

Contact us for a no-obligation consultation on Proxmox, OPNsense, TrueNAS and more.

Get in touch