What’s up, what’s down and what are you not sure about?
Let us know what you set up lately, what kind of problems you currently think about or are running into, what new device you added to your homelab or what interesting service or article you found.
Personally I’m finally reaping the fruits of my labour and enjoy my stable homelab without doing much. One node went down recently and the other took over until I restarted so I was not in a hurry to fix things. Enjoying family time and only running updates that aren’t automated (yet). I’m about to dig a bit deeper into logging, probably setting up central log collection like Loki at some point, but not yet.
I was also enjoying my stable homelab until… well lets just say I got cheap parts here, nice stuff there and now I am building myself a new system and I started by stripping a case I got for 20 bucks and totally spray painting it, got some nice black and white cables, wanna display my nas this time instead of hiding it in the cupboard. After that I will put in the parts I got and then I need to migrate everything from the old nas (well hopefully I just put the drives in and it works). Soooo… Yeah 😀
Started looking at Gemini-protocol over the weekend. (It’s like a newer version of gopher) Now I’m looking for a problem to fit the tool.
I started writing a science fiction, choose your own adventure, short story to fit the platform But that’ll take ages to finish.
I’m also eyeing a meshtastic client proxy. But you only get about 200 bytes per message so I’m not entirely sure it’s worth it.
The last thing, it would be kind of cool is a Zim tie-in. It would be cool to have a canned Wikipedia that could be accessed via Gemini protocol.
Found out Ghost 6.0 is out today and now it supports ActivityPub. It’s time to set up a new blog I’ll never write once more!
Oh exciting, finally!
Got my HPE DL380 G9 networked and configured with hardware RAID 0 and Debian running under ProxMox for a test run (need more disks for RAID 5). Thing had an advanced iLo license intact from the previous owner.
Deployed a docker container of linkwarden to it to try out and it seems pretty nice.
I recently started setting up home server on Raspberry Pi 5. Having issues with raid1. I have 2 nvme PCIe gen 4 SSDs. There was power outages while writing. Now second disk keeps randomly falling. Though, I’m not sure if that’s the reason because I don’t know what was raid status before outage, also disk passes checks. First time it degraded, it tried to recover and it failed. I removed that disk from raid, recreated partition run some test using nvme-cli. Disk looked healthy. I re-added disk, rebuild started and completed successfully. Then I’ve written around 500Gbs of data and it degraded again. At that point I took a break.
There are two things I’m yet to try:
- Change configs to use gen 2 PCI, currently it’s set to gen 3, but AFAIK pi 5 does not support gen 3 officially.
- Remove, format and write data to problematic disk directly. I hope this will give me an idea is this hardware issue or software issue
I’m frustrated and will appreciate any hints.
Recently set up my own nextcloud instance!
Nextcloud AIO?
Yup!
Realized today that borgbackup failed for almost 2 months straight on one of my servers (was a simple case of a lock being stuck). Finally setup push notifications via Pushover to notify on success/fail.
Healthchecks is incredibly nice for this kind of thing, it’ll notify you if it doesn’t receive a ‘success’ ping on whatever interval you specify.
I use it for all my Restic backups.
Yeah that sounds even better. What service do you use?
Backblaze B2 for storage, and I host Healthchecks myself at home.
Same. I’d rather be alerted because something expected didn’t happen, not silence because something failed so hard it didn’t even send an alert.
This is worth it. Had this happen on OS backup. Lost my data. Notifs should be default.
Getting ready to move from out of the woods and back to civilization with my partner.
Not looking forward to having neighbors above or below me but I’m very excited to have internet that doesnt fucking suck.
Once were moved and a bit more settled, I’m gonna start really digging into to selfhosting things. I have the hardware, a couple HP mini PCs that will run home assistant and probably a server for various docker things. Nextcloud and immich seem to be the things I’ve found i wanna use so far. I already have a NAS set up, but was having am issue with it not booting if a monitor isnt plugged in. I bought a dummy plug for it but haven’t tried it out yet.
Will also be setting up an AI server for local LLM use. Hope to train one to fit my needs once I pull the trigger on 3060 12GB card but need to figure out what other parts I’ll use. Might upgrade my main rig and use the parts from that, or maybe I’ll buy a old dell and fix it up. Not sure yet.
Lots of ideas, so little time lol.
Might want a bigger GPU, I have a 3080ti and the 12gb is pretty limiting in terms of how large a model you can use, or like one thing I was hoping to do was essentially replace Google Assistant/Gemini and can’t realistically run a good model and the STT/TTS off the one gpu.
Thats why i was considering training my own model if possible. Ive been toying around with kobold.CPP and gpt4all which both have RAG implementations.
My idea is to essentially chat with documentation and as a separate use case, have it potentially be a AI search engine but locally hosted. I do still prefer to search myself, but fuck man, searches have gotten so bad, and the kobold.CPP web lookup feature was pretty neat IMO.
So yea you’re not wrong, I’m just hoping that if in train it and or give it documentation it can reference when answering, it will be suitable. Mostly AI has been good for me as kind of a rubber ducky when troubleshooting and helping me search for things when I have some specific question and in don’t want “top 5 things vaguely related to your question” results.
Interesting, I mainly have used text generation webui which has a search support plug in, kinda nifty to use my searxng instance for it. It’s a bit finicky though.
Another thing to keep in mind then (apologies if this is just repeating info you already know), you’d also want to keep in mind your total potential context size in relation to the model size, since both take up VRAM. Reading search results/pages can eat up a lot
Yea I’m aware but I appreciate the insight :) so far my local ai experience has been lack luster so I’m hoping that training and RAG will make up for the context size at least a little. Ifnit can answer accurately in the first place, it may not need as big of a context window.
If you haven’t tried using RAG in some form, I would recommend giving it a go. Its pretty cool stuff, helps make models answer more accurately based on the documentation you give them though in my case, ive had limited success. Tbh, chatgpt has become my last resort when I just wanna get something done but I don’t like using it due to the privacy concerns, not to mention the ethical issues I have with ai training in general from big tech.
How is searxng BTW? Would you say its good to host or do you use a normal search engine more often? Or do you just use it for the AI search plugin?
Ive actually been thinking about using it rather than duckduckgo but was also hopeful the search index they are working on would be enough to satisfy my needs, or that a self hosted AI enabled search engine would work well enough when I need it.
I’ve completely replaced my searching with searxng, it is a little slower and ofc if I have an outage or something at home I have to go back to a different search temporarily but overall I like it a lot.
It was one of the first things I set up last year with my homelab because I am attempting to degoogle a fair amount, the Ai search stuff was just a fun test
Thats rad, thanks for the info. I may follow suit, been trying to degoogle myself lately.
For sure, good luck and have fun :D
I’ve been hosting immich for a long time and finally decided to make a website so people could sign up for paid monthly accounts and upload their stuff to the server that I’m going to run anyway. Maybe it’ll make me beer money.
Ngl this would skeeve me out. The chances of someone uploading CSAM while slim scare the fuck outta me.
Straight up CSAM would be pretty brazen. You could probably reduce the chances to zero by just saying that if there’s any thing like that uploaded it will go straight to the police. You probably wouldn’t need to invade anybody’s privacy. The warning itself would set the bar.
I’m kind of surprised there’s not an open source model out there capable of identifying it. Cloudflare has it as a free service if you use them, But I’m not seeing anything that you could just self host.
Models need content for training.
Models that can identify things can generate things if run in reverse. (This is blatant oversimplification.)
Those models already exist. It’s one of the things that everybody’s worried about trying to stop.
With the amount of companies out there willing to fund stopping it I’m surprised somebody hasn’t stepped up to spend a few million dollars to train one specifically to catch people and make it available.
Turns out making money off of it is more important I guess.
Oh, wow, I was expecting a comment about privacy from the preview I got of your message and then it went on to talk about risk to the provider instead! Yes, you’ve definitely identified a risk and I hope to mitigate it with hopes, prayers, and as anonymous access logs as I can get while still identifying public, popular images.
Another glorious day of not having to worry about my nice and stable Debian server. It runs on an old Dell thin client I got on ebay, which isn’t much, but it gets the job done.
I wanna get into it but man, the mountain of knowledge I need to even understand what people are talking about is hard to climb. I’m trying to just get some stuff running in docker and it fails to launch and I’m like… How?! Isn’t that the whole point of docker lol. Baby steps I guess
It’s messy. Dockers superpower: You can write a crazy ass python application that needs dozens of dependencies and weird software configured. You put it into a container, you can update and publish the container with a single script call. Other people can install that, set some variables and not have to install the dozens of other pieces of software. They also don’t have to worry about updates.
But that’s not to say you don’t have to worry about networks, storage and ports.
Then the simplicity of the configuration of containers depends upon the person that made the container. Maybe they wanted to be very flexible and there are dozens of things you need to set. Maybe they didn’t include the data store internally in that container and you need your own data store in another container.
I felt exactly the same when i started - the learning curve is real! Try TrueCharts.org or linuxserver.io for reliable docker templates with good docs that actually work, saved me so much troubleshooting headache.
Thanks will do!
I’ve learnt it from scratch in my week off, spending 2 or 3 hours on it every night for a week (although this might be underselling it as I had become familiar with desktop Linux over the past year and had a superficial idea of Docker containers with my Synology NAS). But still it’s not as big a deal as you think once you find some good resources. I’m going to comment about my setup after this in this thread… Have a look.
Main resource that helped me was Marius Hosting and ChatGPT got me out of trouble when I got stuck by deciphering logs for me when things didn’t work.
Thanks. Yeah I’m just trying to work at it slowly in my downtime instead of just watching YouTube all night.
Check out Cosmos, I struggled piecing things together but when I restarted from scratch with this as the base is has been SO much easier to get services working, while still being able to see how things work under the hood.
It’s basically a docker manager with integrated reverse proxy and OpenID SSO capability, with optional VPN and storage management
Im at the level where I don’t know what SSO means. I can follow instructions to change a DNS. But what a DNS actually is I don’t know. Which is fine, until I need to work out what’s broken
SSO is single sign on, so you don’t need individual username and password for every service. It’s a bit more advanced so don’t worry about it until you have what you want working properly for a while.
DNS is like the yellow pages of the internet - when you type www.google.com your computer uses a DNS server to look up what actual IP address corresponds to the website name. The point of Adguard or pihole is that when a website tries to load an ad your custom DNS server just says it doesn’t recognize the address
Oh like a custom yellowpages, sick!
SSO is “single sign on”. DNS is “domain name service”, which is just a way to turn a hostname (like www.google.com) into an IP address. It’s sort of like a phone directory, but for the Internet.
Sometimes you just need to start small and not worry about over complicating things. I started my journey in 2011 running Plex on a crappy laptop
Are you doing things through docker compose? If so, feel free to PM me or reply here with your compose file and I’ll help as best I can
Docker should be trivial to run. Hopefully it gives you some useful messages in the logs.
Realized last week that my fail2ban settings are too strict – I get banned immediately if I visit my funkwhale (music server) domain without being logged in. In fact, I think much of my “downtime” might have actually just been me banning myself for 15 minutes now and then…
I was thinking about getting rid of Grafana, which is overkill for my server, and replacing it with Logdy this weekend, but didn’t get around to it.
Let us know what you set up lately, what kind of problems you currently think about or are running into
I noticed that my link collector nears perfection (for my use case) - not much stuff required to be done lately. Which is a good thing.
Finally retired proxmox (actually I just removed pve packages and repos). Left the nfs export on there and hardened the whole thing.
Now I’m slowly working to get all my installs into layered ansible playbooks. Fortunately, there exists an incus ansible module.
With separate, mounted, persistent data, it’s getting very close to docker in easy deployment.
I installed a new server at home and went with NixOS. It looks super cool but it takes so much time to learn everything. The only thing keeping me from going back to Debian is how easy it was to permanently mount drives (and save a configuration for any future install or mishaps).
(I.e. mount,
nixos-generate-config,nixos-rebuild switchand done!)NixOS […] learn everything
I don’t think it’s possible to learn everything for NixOS as a casual user / admin. It’s massive. I was luckily able to sneak a NixOS project into work which gave me some paid time on the topic. But there’s always room to learn more about it. Which is a good thing - by its nature, it’s just more powerful than conventional distributions.
More powerful = more mental burden and capacity used to know how to run and manage its unique syntax and structure.
Sincerely NIX user daily. Switching away from nix and off to fedora kinoite.
You might have some luck with Suse, their Yast configuration is very easy and was stable for years for me. Now I’m running on an M1 Mac mini which was more of a pain than a regular setup for sure. Unfortunately the Linux support just isn’t there yet.
