Proxmox ditches Ubuntu kernels, Posteryard polishes posters, and the .lan debate

Homelab Highlights for 2026-11-20: self-hosting and homelab notes worth knowing.

The Lead: Proxmox Is Moving Away From Ubuntu for Its New Kernel. Here’s Why.

Proxmox VE has long been Debian underneath, but its kernel package was built from Ubuntu’s kernel sources. With Proxmox VE 9, an opt-in Linux 7.3 kernel builds directly from the upstream stable branch at kernel.org and sits in the test and no-subscription repos, while 7.0 remains the default. Proxmox says 7.3 is a likely long-term support candidate.

Why it matters: Tracking upstream means security and driver fixes can reach homelab hosts faster, but newer kernels can also change how NICs, storage controllers, passthrough devices, and ZFS on root behave—so test on a spare node first.

Source: Virtualization Howto

News Roundup

Posteryard Brings Clean, Consistent Posters to Plex, Jellyfin, and Emby

Posteryard is a self-hosted, open-source tool that automatically generates consistent poster artwork for Plex, Jellyfin, and Emby. It pulls art mainly from TMDB, adds title logos in a uniform spot, applies quality badges like 4K and Dolby Vision, and starts in dry-run mode so you can preview changes before they touch your library.

Why it matters: It’s a small polish layer that saves manual poster management across growing media libraries, though it shouldn’t overlap with Kometa or other overlay tools.

Source: GitHub

Pulse Monitors Your Proxmox and Docker Servers From One Dashboard

Pulse is a self-hosted monitoring dashboard that pulls Proxmox, Docker, TrueNAS, and other platforms into one view. It tracks resources, backup history, and container health, with alerts via ntfy, Telegram, Discord, and more. The free Community edition is MIT licensed with no system limit and seven days of metric history.

Why it matters: If your homelab monitoring is scattered across multiple tabs, Pulse consolidates Proxmox and Docker visibility without requiring agents on every Proxmox host.

Source: GitHub

Tool & Software Highlight: I replaced my homelab notes with a self-hosted Wiki.js 3.0 instance, and the long-awaited rewrite was worth the wait

An XDA writer migrated his homelab notes from Trilium to Wiki.js 3.0, the from-scratch rewrite that just left beta after four years. The move wasn’t one-click: Markdown imports 404’d until he renamed the root folder to a locale code, and deploying without admin environment variables created a placeholder admin requiring a database wipe and redeploy.

Why it matters: Wiki.js 3.0 offers Markdown pages, built-in page history, and folder or Git sync, making it a viable self-hosted knowledge base for homelab documentation—if you’re willing to work through early migration friction.

Source: XDA

Community Highlight: What’s your opinion on local domains like cloud.lan?

A Lemmy selfhosted thread asked about local domains like cloud.lan. Short names are convenient and easy to add to local DNS, but the poster ran into HTTPS trust pain with Caddy’s internal certificate authority across Windows and family devices, leading him to consider public subdomains with Let’s Encrypt instead.

Why it matters: It’s a classic homelab tradeoff: clean local names keep traffic off the public record, but private CA trust is fiddly per device and becomes a support burden with non-technical household users.

Source: Lemmy.World - selfhosted

Sources

Transcript

Host A: Welcome to Homelab Highlights, the show that surfaces practical homelab wins and useful self-hosted tools.

Host B: Big architecture shift to start with: Proxmox is moving away from Ubuntu as the base for its new kernel. This one is more interesting under the hood than the headline suggests. The context is that Proxmox VE has always been Debian underneath, but the actual kernel package was built from Ubuntu’s kernel sources. That was the pipeline for years. In a forum post on November 18, Proxmox announced an opt-in Linux 7.3 kernel for Proxmox VE 9, and that one is built directly from the upstream stable branch at kernel.org. It lives in the test and no-subscription repos, and the supported 7.0 kernel is still the default. 7.3 is the first public opt-in release built this way, though they had already quietly packaged and evaluated 7.2 using the same upstream approach. Proxmox also says 7.3 is a likely candidate for upstream long-term support, which gives them a branch with ongoing maintenance to watch over a longer stretch.

Host A: So why does the starting point matter? Think about a fix landing upstream for your NIC or storage controller. If Proxmox is already tracking that stable branch, there’s a lot less reconciliation work compared to merging from Ubuntu’s separate downstream tree. And it works the other direction too - a regression someone narrows down on a Proxmox host can get to the actual driver upstream faster. The compatibility side is what homelab folks should actually watch. Proxmox points out that Ubuntu patches and kernel build settings can be part of why a specific device behaves the way it does today. So a newer kernel might fix an NVMe controller and simultaneously cause trouble with a NIC that’s been rock solid for years. Both on the same box. That’s real give and take. If anyone’s tempted to try it, the sensible move is a spare node, not the main hypervisor. Especially if running device passthrough, oddball NICs or storage controllers, or ZFS on root. Stability issues may not show up in a short test - these things need soak time. If something does break, the homework is to report with the working and failing kernel builds, the device, and the driver, because the problem may be in an updated driver rather than the kernel itself. For anyone running Proxmox at home, this is worth watching even if the upgrade itself waits.

Host B: Quick pair for the feed, starting with a polish tool. Posteryard is a new open-source project that automatically generates clean, consistent poster art for Plex, Jellyfin, and Emby. It hunts down artwork without embedded text, adds the title logo in a uniform spot, and layers on badges for 4K, HDR, Dolby Vision, and Atmos. It checks every fifteen minutes by default and even starts in dry-run mode so previews can be inspected before anything touches the library. One caveat worth flagging: if Kometa or another overlay tool is already managing posters, don’t let both fight over the same artwork.

Host A: And on the monitoring side - Pulse is a self-hosted dashboard that pulls Proxmox, Docker, TrueNAS and friends into one view, with backup history tracking and alerts through ntfy, Telegram, Discord, and the like. The free Community edition is MIT licensed with no limit on systems, just seven days of metric history. There’s also an optional AI monitoring feature called Pulse Patrol that can examine infrastructure health on a schedule using a local model or external provider. If monitoring is currently scattered across three tabs, that’s worth a glance.

Host B: For today’s tool highlight, Wiki.js 3.0 is out of beta after four years of development - a complete rewrite - and an XDA writer took it for a spin by migrating his homelab notes onto it. The setup here is relatable. He was running Trilium for notes across thirty-plus containers, six machines, a few VPSes, and networking gear - and before that, the information was genuinely scattered: recovery codes in a phone notes app, API keys in a documents folder, network config buried in a chat. Classic. What pulled him toward Wiki.js 3.0 was the rewrite’s highlights: a Markdown editor, a file manager, content blocks, and what he calls a better page history than Trilium’s - page diffs and edit summaries, very Wikipedia-shaped.

Host A: The migration did not go one-click, though, and that’s the useful part. He exported from Trilium as Markdown even though Trilium recommends HTML for preserving formatting, pointed Wiki.js’s local file system storage at the folder, ran the import - and the dashboard said eighteen pages existed, but every one of them 404’d. The culprit: Trilium uses a tree structure in URLs, and Wiki.js read the top folder name as the locale. Renaming the root folder to the locale code fixed it. There was also a setup hiccup where deploying without the admin email and password environment variables created a placeholder admin, and the fix was wiping the database and redeploying with them set. Which is exactly the kind of friction that only shows up in a real writeup, not the docs. His verdict after all that: Trilium is on its way out, and Wiki.js takes over. He notes Trilium has genuinely good features, including LLM integration, but a familiar workflow won out. Worth saying this is one writer’s early impression - he admits he hasn’t explored everything yet. But for anyone whose homelab documentation lives across four apps, a self-hosted wiki with Markdown pages and built-in page history is a solid pattern, and the Git sync option means the notes can be versioned like code.

Host B: Last up, a community thread from the selfhosted community on Lemmy - a user named marius asking for opinions on local domains like cloud.lan. The replies got into one of the classic homelab debates. The appeal is obvious. Short names beat a long public subdomain, the good short domains are all taken anyway, and dropping entries into a local DNS server is trivial. The problem he ran into is HTTPS. He set up Caddy with its own internal certificate authority, but even after installing the root certificate on Windows, plenty of programs still refused to trust it - and most of them throw useless error messages while doing it.

Host A: That’s the pain everyone in that thread recognizes. Getting a private CA trusted is fiddly per device, per platform, and per application, because not everything honors the system trust store. And his kicker is the family angle: try explaining root certificate installation to your mom so she can use the self-hosted cloud. At that point the friction isn’t technical anymore, it’s support burden. Which is why his conclusion was to consider switching back to public subdomains, where Let’s Encrypt certificates just work on anything that can reach the internet. And that’s the honest tradeoff the thread lands on: local names are cleaner and keep things off the public record, but the certificate story is rough. Public subdomains mean a real domain in DNS and publicly issued certs, but zero trust-store wrestling. There are middle paths people run - a public domain with split-horizon DNS resolving internally, for instance - though whether that fits depends on whether the services should be reachable from outside at all. For a household with actual non-technical users on it, the thread’s lean toward Let’s Encrypt is hard to argue with.

Host A: That’s Homelab Highlights for Friday. Until next time, happy hosting!