Fixed
- A restart while a node is down no longer takes the integration with it. The addresses of the cluster's other nodes were read from
cluster/statusonce a connection stood and were forgotten with the session, so a restart began against the configured host and had nowhere else to go — however many nodes were still running. They are kept in the config entry now, refreshed whenever the cluster can be asked, and a reload that cannot read them keeps the last ones. The first request of a setup can then go to another node, and the entry loads instead of waiting. The configured host stays the one every start goes to first, and the addresses are redacted from the diagnostics download, like the host itself. Thank you, @mundschenk-at, for the report (#692). - A guest renamed in Proxmox is renamed here too. The device name was only ever written when the device was created, so a container created and named afterwards kept showing
LXC CT516 (516)until the next restart. It now follows within a polling interval, and the displayed names of its entities with it. A name given in Home Assistant is untouched, and entity ids keep theirs.
Also in this release
Ruff 0.16.8 and pip 26.2.1, and the last Crowdin sync was reverted: three of its files carried placeholders the English strings do not, which failed the translation check. The strings need correcting in Crowdin, otherwise the next sync brings them back.
Documentation
The comparison against the core integration gained the rows for the pressure sensors and the memory a VM costs its host, which 5.3.0 shipped without.