ProxMenux v1.2.8

ProxMenux 1.2.8 is a short corrective release built on what users reported after 1.2.7. OCI manager Apps gains the most: Compose files with several services, updates that keep what you change in Proxmox, a VLAN tag for the access interface and seventeen new applications. Log2RAM 2.0 keeps the journal and the Proxmox logs on disk, and ProxMenux Monitor adds ZFS pool details, storage controllers and Proxmox updates that carry on after signing out.

A special thank you goes to @Vaso73, @f3rs3n, @jcastro and @DrOwl for their contributions and reviews, and to everyone who reported a problem or suggested an application.

📦 OCI manager Apps

• Custom image installs Compose files with several services (PR #550, contributed by @jcastro): one LXC per service on a private network, started in dependency order, like the multi-container applications of the catalog. Image references, environment files, dependencies, shared data and one-shot jobs are kept, and updates, backups, recovery after a restore and removal work on the whole stack.
• Your own variables stay in an OCI container. Add or change a variable in the Proxmox options of the container and it is kept: it is saved in the oci-compose of the application and the update carries on with it, manual or scheduled, as with Docker. This works on single applications and on every member of a stack. A variable the image defines and that you removed comes back with its image value.
• Updates keep the rest of what you change in Proxmox: the start command, network options, DNS, hostname, features, protection, disk sizes, added host directories and lxc.* lines stay on the rebuilt container, also in scheduled updates.
• The access interface takes a VLAN tag (#565, requested by @xistoso): the advanced installation asks for it next to the address, and Modify changes or removes it on a single application. With a tag, the application is checked from inside its container, so the host needs no access to that VLAN.
• Ollama and llama.cpp run on Intel and AMD integrated GPUs: a new Vulkan profile uses the GPU through its render node, and on Radeon iGPUs such as the 680M and 780M the ROCm profile is offered as experimental.
• Seventeen new applications. As stacks: Authentik (PR #563), TeslaMate (PR #555), Umami (PR #541) and Scanopy (suggested by @gabglus), whose daemon scans the network from an unprivileged LXC. As single containers: Zigbee2MQTT (PR #545), with the adapter as an optional USB device, Scrypted (PR #546), NetAlertX (PR #547), and Activepieces, Bytebase, Campfire, Chhoto URL, Ech0, Kopia, Metabase, OpenHabitTracker, Restic REST Server and Watcharr (PR #538). All but Scanopy were contributed by @jcastro.
• New stacks get readable container names (PR #555, contributed by @jcastro), as in teslamate, teslamate-db and teslamate-grafana.
• The Ollama entry describes what the image provides (PR #566, contributed by @jcastro): its HTTP API on port 11434, with its link and startup check on /api/version.
• An update clears the snapshot an interrupted backup leaves (#565, reported by @kiosvantra). A snapshot made by the user stops the update, because a rebuilt container cannot keep it.
• A container that ignores the shutdown signal is stopped after 60 seconds, as docker stop does, during an update and when an interrupted operation is recovered.

🖥️ Monitor — Proxmox updates and pending reboots

• The Proxmox update runs as a job on the server (PR #551, contributed by @Vaso73): closing the terminal or signing out leaves it running, and signing back in reopens the same update. A finished update opens as a read-only result.
• A pending reboot is shown when the running kernel is older than the installed one, also after updates made outside the Monitor. The restart button checks the host, active updates and package locks before rebooting.
• The ProxMenux version row stays compact when the installed version is the latest and shows both versions when a newer one is available.

💽 Monitor — storage

• ZFS pools show their layout and health (PR #561, contributed by @DrOwl): vdevs and their members, the last scrub or resilver, error counts per device, fragmentation and dedup ratio, with OpenZFS 2.3 or later.
• Each disk shows the storage controller it hangs off (PR #562, contributed by @DrOwl), in its details and in the Hardware page. Storage can group the physical disks by controller, an option in Settings.
• VM disk thresholds in Settings (#543, suggested by @byte21516).
• Every disk is monitored unless it is excluded in Settings: a mechanical disk with no activity keeps showing its type, model, serial number, temperature and SMART data (#542, reported by @ovikiss; #553, reported by @peterurbanec). A disk in Disk exclusions receives nothing, so it can spin down on its own timer, and a disk that is already parked stays asleep.

📝 Log2RAM 2.0

• The journal and the task logs of Proxmox VE and Proxmox Backup Server stay on disk (#540, reported by @shtefko), so they keep their history. Log2RAM holds the logs that change all the time, such as the pveproxy access log.
• A guard watches the ramdisk every 10 minutes: past 85% it rotates the logs, compresses the ones just rotated and writes everything to disk. No log is emptied.
• Every sync is flushed to disk, so a finished sync survives a power cut. An installed Log2RAM updates to this design with ProxMenux, and the change takes effect at the next boot.

🔔 Notifications

• AI Suggestions now reaches the runtime formatter, for queued and direct notifications. Backup errors are explained separately from the guest inventory, so statuses, sizes, durations, paths and totals stay unchanged. Matching local vzdump logs can supply a cause hidden behind a generic exit code, only when the complete timestamped failure line matches the same task.
• Explicit storage quota errors have a localized explanation even when AI is unavailable. Guests without evidence keep their cause unconfirmed; detailed email reports retain the original diagnostics, and failed rows no longer show null as an archive path. Exhausted API credits or quota are distinguished from temporary rate limits and are not retried.
• Ollama (Local) accepts an optional API key for servers behind an authenticated proxy or gateway. The key is stored encrypted and shown masked, and the connection test tells an authentication failure apart from a server it cannot reach.

🛠️ Fixes

• Host backup jobs run on their schedule: the menu checks the schedule when a job is created, and updating ProxMenux repairs saved jobs whose timer systemd did not accept, such as daily 02:00.
• A service reloading or restarting counts as running (#567, reported by @Joshua1264): a restart gets a few seconds to finish, and nothing is concluded while a package update runs. A service that has failed or never finishes starting is still reported.
• The Intel GPU view ends intel_gpu_top with each reading (#568, reported by @FromTheCToTheG), which also makes the reading half a second faster.
• VM disk usage counts every filesystem of the guest (#559, with the fix proposed by @TheFoxCraft), also on guests whose agent reports no device name, such as Alpine.
• The LVM check keeps the Proxmox filter for zvols and RBD images on hosts whose only volume groups belong to their VMs (#552, reported by @osipovdm-dot).
• A volume group with no active volumes counts as healthy, such as a disk left over from another installation, stopped guests or shared LVM. A storage Proxmox cannot use is still reported by the storage check.
• The NVIDIA installer stops before touching the driver when Secure Boot is enabled (#571, reported by @Balethorn), because the kernel would refuse the unsigned module, and any installer failure stays on screen with its reason.
• Bitwarden recognizes the 2FA field on the Monitor login (#570, reported by @shtefko): the backup code hint keeps its text and place, outside the group of the field.
• Secure Gateway installs on Proxmox hosts on BTRFS (#564, reported by @78v5): the Alpine template goes to an active storage that holds container templates.
• A disk I/O error from the kernel is recorded on its disk, with its serial number, also when the error names a partition.
• The Monitor installs as an app (PWA) from the browser: it serves its own web app manifest and service worker, with the 192 and 512 px icons Chrome requires.
• The Y-axis titles stay inside the overview charts (PR #539, contributed by @f3rs3n).
• The application version-tracking catalog builds from the current Helper-Scripts repositories.
• The system update confirmation opens with Yes selected, in post-install and in the Proxmox VE 8 and VE 9 update scripts, and the package list stays aligned.

🙏 Acknowledgments

• @jcastro — the import of multi-service Compose files, tested on real stacks, sixteen of the new applications and the guide for proposing applications to the OCI catalog.
• @Vaso73 — the server-side Proxmox update job and the pending reboot detection, tested through a real upgrade and reboot, and the Slovak translation.
• @DrOwl — the storage controller mapping and the ZFS pool details.
• @f3rs3n — the overview charts and the Italian translation.
• Carlos Cáceres — the report and configuration example that prompted the API key for local Ollama.
• Everyone who reported a problem or suggested an application, named next to each change.