dougiteixeira/proxmoxve 5.2.0

on GitHub

The follow-up to 5.1.0, from going through the open issues and discussions once more: an entity id scheme, a polling interval, fewer requests, and the small features people had asked for. Developed and tested in my fork and run on my four-node Proxmox VE 9 cluster since; the pull request (#679) has the details.

Please read first

• The setup form asks for an entity id scheme — a required choice, no default, for new setups only. Standard is Home Assistant's device-then-name, as today. Extended puts a common prefix first and the id before the name (sensor.pve_qemu_108_win11_cpu_used), so a search or recorder filter catches everything by the prefix, a list sorts by id, and shared storage carries its node in the id. The scheme is a suggestion Home Assistant takes when an entity is registered for the first time; entities that exist keep their ids whatever is chosen, an existing setup is on Standard and stays exactly as it is, and there is deliberately no bulk rename. (#573, #604, #283)

Added

• Polling interval — 30, 45, 60, 90 or 120 seconds for nodes, guests, storage, backups, cluster and discovery; 60 unless changed. Certificates, subscriptions and Ceph stay hourly, failed tasks at five minutes. Proxmox refreshes guest figures about every ten seconds, so nothing faster than 30 is offered. (#591, #431, #339)
• Advanced configuration — a new options page for physical disks, failed tasks, package updates, the interval, the guest file, the backup storage and the entity id scheme; the selection page is the selection again, and each page saves only what it shows. The setup form slims down the same way, and the disk and task options say what they read.
• Monitor package updates can be switched off — for a setup that holds Sys.Modify but does not want the update entity, or does not want to grant the privilege. (#588, second half)
• Snapshots per VM and container — the count, with the names newest first and the time of the newest as attributes. Diagnostic, off by default. (#290)
• IP address per VM and container — a VM's through the guest agent, a container's from its own interface list; loopback and link-local left out, the first IPv4 shown, all of them by interface as attributes. Diagnostic; created once the address could be read. (#140)
• Guest agent per VM with the agent configured — on while it answers. Diagnostic. A refused read leaves it unknown; the 5.1.0 repair names the privilege and now mentions the addresses too. (#140)

Changed

• cluster/resources is read once per polling burst, not once per coordinator. Every VM, container and storage coordinator, the cluster summary and discovery read the same list themselves — about forty identical requests a minute on a modest cluster. One shared read per config entry serves them for fifteen seconds; a failure is shared the same way, so a dead host is hit once per burst. Discovery still reads afresh. (#595, #618)

Documentation

docs/behaviour.md has sections on entity ids and on how often the integration polls; the README's option list, the permissions table and the comparison with the core integration are updated. Only en.json carries the new strings; the other languages fall back until Crowdin catches up.

Full changelog: 5.1.0...5.2.0