Lead admin for https://lemmy.tf, tech enthusiast
Uhh… if your script is subbing to 24k remote communities, those will continue to grow from then on, unless you start purging communities at some point. After one user subscribes to a community, all new content gets indexed and stored on your instance. Pict-rs can cache images short term (and eventually clear them out), but Postgres will start growing very quickly and never slow down until it fills up disks.
I’ve had access to 4 for several weeks and it’s not really much better. Maybe I’m just asking too much of it though.
Strongly disagree about gpt being excellent for code, it’s extremely confident about the wrong answer most of the time. I’ve found it to be mildly useful as a Stack Overflow alternative (for asking general questions and having it point me in some direction) but it’s code outputs are garbage.
These instructions won’t work in anyone’s unraid box, even if they compile compose from source. Not sure why people think posting random chatGPT’d instructions is remotely useful.
I use Internetbs.net, sometimes Name.com if they have a particular TLD way cheaper.
I’m just letting mine do whatever it wants, got plenty of local storage. If/when I have storage issues I’ll add an s3 bucket, pretty easy to modify the entrypoint for pictrs to pass s3 connection info in the docker-compose deployment.
I’ve got a baremetal server with OVH running VMware, so it’s just a VM that I manage. I’m paying more for it than I’d like, but it’s running far more than just Lemmy. If I wind up ditching it in the future, it’s just a quick vMotion off to another machine + DNS updates.
Here’s a current output of my storage about a week into hosting the instance. It’s growing slower than I expected, and I do have plans to move volumes/pictrs up to an s3 bucket whenever I start running low on local storage.
[jon@lemmy lemmy]# du -sh volumes/*
2.5G volumes/pictrs
2.2G volumes/postgres
I would recommend locking down SSH on your Lemmy server, I have mine restricted to allow logins from VPN only. Otherwise you’ll get probed 24/7 with a public server.
Do you need access control? If not, a simple Apache/Nginx directory listing is nice and easy, just drop your files in your webroot and you’re set. h5ai is a nice addition if you go that route.
If you need access control (or at least some sort of obfuscated URLs), Nextcloud is a good option. Pretty easy to get up and running, and there’s a ton of plugins available.
Got a Veeam community instance running on each of my VMware nodes, backing up 9-10 VMs each.
Using Cloudberry for my desktop, laptop and a couple Windows VMs.
Borg for non-VMware Linux servers/VMs, including my WSL instances, game/AI baremetal rig, and some Proxmox VMs I’ve got hosted with a friend.
Each backup agent dumps its backups into a share on my nas, which then has a cron task to do weekly uploads to GDrive. I also manually do a monthly copy to an HDD and store it off-site with a friend.
I spun up Firefly a few months ago and had about three weeks where I was actively categorizing transactions and reconciling everything and then my ADD kicked in. Really cool tool but I just need something low-maintenance for budget tracking.
From what I’ve seen and read, server to server traffic is less taxing on instances than client to server. So even if your instance is JUST you, it would be your instance talking to everything else so it would have some net benefit on the federation. But it would take a lot of users self-hosting solo instances for this to help in any noticeable way, I’d think.
There is certainly no downside to running a solo instance, if you’re even slightly interested I would say go for it!
Mine is public, yes. Not sure how many active users I have, 28 signups but my sidebar shows 5 monthly active users so far. I imagine this will pick up once people start commenting and posting more.
I’m interested, but I don’t know Rust and haven’t done frontend work in years. Might be able to do some work around scalability and contribute to a Kubernetes deployment guide (and/or Helm chart).
Yes, I’ve got separate subnets & vlans for a few things. My PCs/phone/tablets/etc, homelab, IoT devices (i.e. loads of Govee bulbs/ropes, gaming consoles, oven, etc), Guest (all isolated from everything else internal) and one for my roommate. I’m on a Unifi Dream Machine Pro so setting up traffic rules to allow certain traffic from PC vlan to homelab (and the other way) was pretty straightforward.
As for the VPN, yes a full tunnel would force all traffic over the VPN, but for all but my *arr stuff that’s overkill. I just join all my VMs to Zerotier and force traffic from the public LB in via their VPN IP, but the VMs can still pull yum updates and anything else they want over my WAN link.
4vcpu (Ryzen), 8GB RAM, 256gb disk (which will be expanded when it gets to like 60% full). Not too worried about storage unless I get a bunch of image-happy users, text all comes in as json and goes straight to Postgres so it’s not a concern.
I run all my lab servers/services/etc in their own /16 on my home net. Nothing is publicly routed in over my WAN IP- if I want to expose a service, it goes through Nginx Proxy Manager to my local service via a ZeroTier tunnel.
I would strongly encourage you to not expose any of the *arr services (particularly your download node) to your WAN IP. PIA’s desktop app does a pretty good job of forcing a full tunnel with a VPN kill switch, so you never have to worry about your ISP catching onto what you’re doing.
All the things! I’ve got a hybrid VMware cluster (two nodes at home and one in a DC) with a bunch of VMs for stuff like Plex, Plesk, Gitlab, Lemmy, Stable Diffusion, etc. also running a 5-node Rancher k8s cluster.
Some of my public services do actually run from home but are routed through ZeroTier to my Nginx Proxy Manager appliance.
Pretty much everything is running RHEL8 or CoreOS after a recent migration. Veeam for backups (two community instances since I’m too cheap to pay for licensing for personal stuff).
Edit:
Not sure this one counts but…
Small instances don’t seem to require anything major, I’m running mine on a VM with 4c/6gb ram/256gb disk with no issues- it’s just a few Docker pods. Just make sure you use a dynamic DNS provider if you’re hosting from home, as valid SSL is required to connect to the federation.
UI doesn’t come up until database migrations fully complete. Can take half an hour or more depending on how much content is indexed in your instance.