Fixed
- A node that cannot be read no longer takes every other entity with it. The device description read a processor name that had never been assigned, and the resulting error failed the setup of the sensor and binary sensor platforms — so one unreadable node cost the entry all of its entities, the cluster's included. Thank you, @bilogic, for the log that showed it (#712).
- Saving Add or remove Nodes, VMs, Containers or Storages no longer deletes the node's disk and pool devices. The switch for physical disks moved to the Advanced configuration page in 5.2.0, and the selection step kept reading it from its own form — so every save looked like "disks turned off". Home Assistant restores the device, which is why this looked harmless, but the entities come back as new: with Enable newly added entities off they come back disabled and disk health stops reporting without an error. Thank you, @ItsColby, for a report that named the commit (#713).
- The
Software updateentity no longer says "up to date" while packages are pending. The version it reports carries the pending counts as-p7-d2, which Home Assistant's version comparison reads as a pre-release — ranking it below the installed version, so with nopve-manageramong the pending packages the entity compared itself into silence. Thank you, @DrNiKa, for working the mechanism out (#711). - The hardware sensors are found again under PVE-mods v2. It renamed the field twice —
PveMod_JsonSensorInfo,PveMods_JsonSensorInfo,PveMods_SensorInfo— and v2.1.0 also changed the shape: the chips moved toenhanced_sensors, keyed by the raw sensor id, and a reading became an object instead of a flat number. All three are read, and the label still names the entity, so upgrading PVE-mods neither renames the sensors nor loses their history. Thank you, @atfua, for finding the rename (#706). - The QEMU guest agent is only asked while it can answer. Proxmox answers every
agent/*read with a 500 while the agent is not enabled or the VM is not running, and theagentflag in the status read is the VM's configuration — set on a stopped VM too, which cost some 2,900 failed requests a day for a single stopped VM. Thank you, @BadFlo, for the fix and the measurements (#708).