terribleplan

Laptops/desktopes: no real naming scheme, they use non-static DHCP leases anyway.

Physical servers: NATO phonetic alphabet. If I run out of letters something has gone terribly wrong right.

VMs: I don;t have many of these left, but they are named according to their function and then a digit in case I need more. e.g. docker1, k3s1. This does mean that I have some potential oddities like a k3s cluster with foxtrot, alpha, and k3s1 as members, but IMO that’s fine and lets me easily tell if something is physical or virtual. I am considering including the physical machine name in the VM name for new things as I no longer have things set up such that machines can migrate… though I haven’t made a new VM in some time.

Network equipment: Named according to location and function. e,g, rack-router, rack-10g, rack-back-1g, rack-ap, upstairs-10g, upstairs-ap. If something moves or is repurposed it is likely getting reconfigured so renaming at that point makes sense.

Pretty sure it needs to be https://$user:$pat@github.com/username/repo.git#branch.

I have owned and otherwise dealt with a few different Startech 4-post open racks and have been very happy with them. I currently use one of their 25U racks for my lab, but am running out of space…

I started on Gitlab, which was a monster to run. I moved to Gitea, until the developers started doing some questionable things. Now I’m on Forgejo (a fork of Gitea).

Yeah, all I know is that I am definitely seeing images loaded in from domains other than that of my instance as I load/scroll pages, which I want to be loaded via my instance for privacy reasons.

I believe the Pictrs is a hard dependency and Lemmy just won’t work without it, and there is no way to disable the caching. You can move all of the actual images to object storage as of v0.4.0 of Pictrs if that helps.

Other fediverse servers like Mastodon actually (can be configured to) proxy all remote media (for both privacy and caching reasons), so I imagine Lemmy will move that way and probably depend even more on Pictrs.

IIRC Lemmy preloads all thumbnails for posts in communities you subscribe to into pictrs to be cached for like a month or something. So, yeah…

The servers aren’t even identified in the listing as R610s (or E01S, they misread that as “EOLS”), so who knows…

Lemmy has a feature/setting called “Private instance” that I think could be used to achieve this, but I think that got broken at some point because it got tied to turning federation off… not sure what the current state is but may be worth looking into.

I switched from Plex to Jellyfin several years ago and haven’t really looked back. Overall I just didn’t like the direction plex kept going (pushing shit streaming services, central auth, paywalling features), and dropped it even though I grabbed a lifetime plex pass back in the day. The only thing I miss about plex was the ease of developing a custom plugin for it since you could pretty much just drop python scripts in there and have it work, though their documentation for plugin development was terrible (and I think removed from their site entirely).

I love tinc, it’s so simple. I wish there were something just as easy that leveraged wireguard instead of whatever custom VPN/tunneling stuff tinc uses, as using it scares me with how seemingly little maintenance tinc gets. Like if tailscale/headscale and tinc had a baby, haha.

Is there a way to run tinc on your phone or similar? To me that’s another bonus of tailscale at least.

Having a “source of truth” makes many things easier but less resilient. One place to go get the latest version of something mutable. The fediverse/ActivityPub needs to get on board with some form of DID or something similar before worrying about improving the ID system (and the ID system is inherently tied to JSON-LD, so AP would need to stop using that or there would need to be a new version of it) IMO.

It depends on what specific thing you want to add geoblocking to, but often something like the MaxMind GeoIP database, which then can feed into a firewall to pre-emptively geo-block at a connection level, or as part of e.g. nginx geolocating the IP a of the connecting IP then making the blocking decision at request time.

There’s a project that works with Traefik’s forward-auth middleware to do this, which is probably how I would go about it if I wanted it at an HTTP level.

Basically, no:

It can cause some wackiness… basically you will need to maintain that old domain forever and everything will still refer to that old domain.

For example, your post looks like this from an ActivityPub/federation perspective:

{
    [...]
    "id": "https://atosoul.zapto.org/post/24325",
    "attributedTo": "https://atosoul.zapto.org/u/Soullioness",
    [...]
    "content": "<p>I'm curious if I can migrate my instance (a single user) to a different domain? Right now I'm on a free DNS from no-ip but I might get a prettier paid domain name sometime.</p>\n",
}

The post itself has an ID that references your domain, and the the attributedTo points to your user which also references your domain. AFAIK there is no reasonable way to update/change this. IDs are forever.

It would also break all of the subscriptions for an existing instance, as the subscriptions are all set to deliver to that old domain.

IMO your best bet would be to start a new instance on the new domain, update your profile on the old one saying that your user is now @Soullioness@newinstance.whatever and maintain that old server in a read-only manner for as long as you can bear.

Lemmy and Akkoma, both in docker with Traefik in front.

Ext4 because it is rock solid and a reasonable foundation for Gluster. Moving off of ZFS to scale beyond what a single server can handle. I would still run ZFS for single-server many-drive situations, though MDADM is actually pretty decent honestly.

A few of these servers were stacked on top of each other (and a monitor box to get the stack off the ground) in a basement for several years, it’s a journey.

No. - sent from my iNstance

Things don’t get backfilled, so until a new action happens on an old post/comment/etc they won’t show up on your instance. New things should make their way in eventually though.

Taking the link of a specific post/comment from the community instance and searching for it from your instance should populate it on your instance, just like you probably had to do to get this community to show up so you could subscribe/post at all.

There are backfill tools/scripts, but unless you really want old posts I wouldn’t use them. It unnecessarily increases the load on already struggling popular/overloaded instances like lemmy.world.

Business in the front:

• Mikrotik CRS2004-1G-12S+2XS, acting as a router. The 10g core switch plugs into it as well as the connection to upstairs
• 2u cable management thing
• Mikrotik CRS326-24S+2Q+, most 10g capable things hook into this, it uses its QSFP+ ports to uplink to the router and downlink to the (rear) 1g switch.
• 4u with a shelf, there are 4x mini-pcs here, most of them have a super janky 10g connection via an M.2 to PCIe riser.
• “echo”, Dell R710. I am working on migrating off of/decomissioning this host.
• “alpha”, Dell R720. Recently brought back from the dead. Recently put a new (to me) external SAS card into it, and it acts as the “head” unit for the disk shelf I recently bought.
• “foxtrot”, Dell R720xd. I love modern-ish servers with >= 12-disks per 2u. I would consider running a rack full of these if I could… forgive the lack of a label, my label maker broke at some point before acquiring this machine.
• “delta”, “Quantum” something or other, which is really just a whitelabeled Supermicro 3u server.
• Unnamed disk shelf, “NFS04-JBOD1” to its previous owner. Some Supermicro JBOD that does 45 drives in 4u, hooked up to alpha.

Party in the back:

• You can see the cheap monitor I use for console access.
• TP-Link EAP650, sitting on top of the rack. Downstairs WAP.
• Mikrotik CRS328-24P-4S+, rear-facing 1g PoE/access switch. The downstairs WAP hooks into that as well as the one mini-PC I didn’t put a 10g card on. It also provides power (but not connectivity) to the upstairs switch. It used to get a lot more use before I went to 10g basically everywhere. Bonds 4x SFP+ to upllink via the 10g switch in front.
• You can see my cable management, which I would describe as “adequate”.
• You can see my (lack of) power distribution and power backup strategy, which I would describe as “I seriously need to buy some PDUs and UPSs”

I opted for a smaller rack as my basement is pretty short.

As far as workloads:

• alpha and foxtrot (and eventually delta) are the storage hosts running Ubuntu and using gluster. All spinning disks. ~160TiB raw
• delta currently runs TrueNAS, working on moving all of the storage into gluster and adding this in to that. ~78TiB raw, with some bays used for SSDs (l2arc/zil) and 3 used in a mirror for “important” data.
• echo, currently running 1 (Ubuntu) VM in Proxmox. This is where the “important” (frp, Traefik, DNS, etc) workloads run right now.
• mini-pcs, running ubuntu, all sorts of random stuff (dockerized), including this Lemmy instance. Mounting the gluster storage if necessary. They also have a gluster volume amongst themselves for highly redundant SSD-backed storage.

The gaps in the naming scheme:

• I don’t remember what happened to bravo, it was another R710, pretty sure it died, or I may have given it away, or it may be sitting in a disused corner of my basement.
• We don’t talk about charlie, charlie died long ago. It was a C2100. Terrible hardware. Delta was bought because charlie died.

Networking:

• The servers are all connected over bonded 2x10g SFP+ DACs to the 10g switch.
• The 1g switch is connected to the 10g switch with QSFP+ breakout to bonded 4x SFP+ DAC
• The 10g switch is connected to the router with QSFP+ breakout to bonded 4x SFP+ DAC
• The router connects to my ISP router (which I sadly can’t bypass…) using a 10GBASE-T SFP+.
• The router connects to an upstairs 10g switch (Mikrotik CRS305-1G-4S+) via a SFP28 AOC (for future upgrade possibilities)
• I used to do a lot of fancy stuff with VLANs and L3 routing and stuff… now it’s just a flat L2 network. Sue me.