// the take
An Ansible role that installs Proxmox VE on Debian and manages the full lifecycle of a cluster afterward: users/ACLs, storage backends, Ceph, HA watchdogs, PCIe passthrough, and datacenter.cfg. It's for homelab and small-shop operators who want their PVE cluster state declared in Ansible instead of clicked through the web UI, and who are comfortable with multi-node Debian/PVE setups rather than single-box installs.
It ships real custom Ansible modules (proxmox_user.py, proxmox_acl.py, proxmox_storage.py, proxmox_role.py, proxmox_metric_server.py) backed by a pvesh wrapper in module_utils, so user/ACL/storage changes are properly idempotent instead of shell_out-and-hope. Version coverage is wide and explicit — separate vars files per Debian release (buster through trixie) tracking PVE 6.x through 9.x — rather than one script that happens to work on whatever's current. It also goes past the installer into real cluster mechanics: corosync ring/link address priority, IPMI/NMI watchdog wiring into pve-ha-manager, and IOMMU/VFIO passthrough config including driver blacklisting, which is the kind of detail that's easy to get wrong by hand. CI runs ansible-lint plus a Vagrant/libvirt harness that actually stands up multi-node clusters across Debian releases, not just a syntax check.
Ceph support is explicitly labeled experimental in the README and isn't exercised in CI at all — the maintainer says there's no usable block device in CI runners, so you're trusting Ceph behavior that's only been validated by user reports, not automated tests. Clustering requires root-over-SSH between every node, which the role sets up for you but is a real security posture to accept, not an edge case you can opt out of if you want >1 node. It also doesn't manage NTP itself; you're expected to bolt on a separate role (geerlingguy.ntp in the examples), which is one more thing that has to be correct before the cluster will even form reliably. Kernel-update reboots are handled with a shell sleep+shutdown and wait_for_connection poll rather than any cluster-aware draining — pve_reboot_on_kernel_update_delay is a flat timer, so a slow-booting node or a reboot that hangs stalls the whole playbook with no real recovery path besides going in by hand.
We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →