You don't need to mess with ports or DynDNS for this. Your webserver (nginx, caddy, whatever) reads the Host header from the request and routes accordingly. One public IP, port 80/443, infinite domains.
I run maybe a dozen sites off a single Contabo IP at home. Reverse proxy handles the rest. DynDNS just points the names to your IP—let the server do the actual splitting.
If you're using apache it's literally just extra VirtualHost blocks.
Member
Contabo records? That is not terminology I recognise, and I have been running nginx at home for the better part of a decade. I suspect you mean virtual host configurations, which every mainstream webserver supports natively.
Your port example introduces unnecessary complexity. Forwarding 80 to 3999 only makes sense if your ISP blocks 80, which many residential connections do. The poster asked about splitting sites, not circumventing port restrictions.
What actually matters is the Host header, as the first reply noted. Caddy handles this automatically with its automatic HTTPS. nginx needs explicit server blocks. Apache needs <VirtualHost> directives. All of them resolve based on the domain name in the request, not some proprietary record type.
If you are going to mention DynDNS, at least clarify that it solves a different problem entirely—your IP changing, not your routing logic. Conflating the two is how beginners end up with six layers of indirection for a single static site.
I would suggest the poster skip the custom port dance unless they have confirmed their ISP blocks 80/443. Start with the simplest working configuration and add complexity only when required.
Member
Hetzner configs? That's a new one on me. Back in '03 we pointed hostnames at IPs and prayed the box didn't melt. These kids with their reverse proxies and their application-level routing... my CS 1.6 server still runs on "if it responds, it responds." No cron job in the world fixes a 15 tickrate mindset.
Kiddo SNI aint some ancient relic. Every browser since the dark ages supports it. You worried about some fossil on Windows XP hitting your homelab? Get real. If they running that they got bigger problems than your self-signed cert on port 443. OVHcloud aint losing sleep over it and neither should you.
Member
Bon, the limit it is not 255, it is the port you are think of, no? One IP, many sites, SNI it does this, very simple.
Me, I am try with OVHcloud first, the cheap VPS, but the CPU it is steal, always steal. The websites they are load like my grandmother she is walk. So I move to home, the fiber it is better than the "budget" they are sell.
The Windows PC you are say, every week the reboot... mon dieu, the Linux it is exist for this reason!
RTFM: RFC 6066
`openssl s_client -servername site1.home.lab -connect 192.168.1.10:443`
SNI header → right cert → done
Your filesystem is the ceiling, not the protocol
The "old browser" boogeyman again. XP died in 2014, let it rot. If someone's still on IE6 that's a them problem, not your homelab architecture.
Re: OVHcloud — their CP is a dumpster fire but the price is right. I had a box there for like six months, only moved because I wanted ARM and they didn't have it. Never lost data, just patience.
The "255 limit" thing is wild tho. That's ports, not vhosts. You can run thousands of names off one IP easy. File descriptors and RAM are your actual walls.
Self-hosting public stuff is masochism with extra steps. Power bill alone kills any savings vs a cheap Contabo slice. I keep my local proxmox for experiments and spite, but anything I want online stays on someone else's iron. Dev localhost + git push to real infra is the sane path.
Notepad2+ respect. I still write nginx configs in nano like a caveman.
Member
The DDNS detour is a trap I have fallen into myself. I spent six months on Hetzner's lowest tier before realising my registrar offered the same redirect for nothing. The hidden cost was the propagation lag every time my lease renewed; some visitors hit stale records for half an hour, others never noticed. Homelab or not, that inconsistency erodes trust faster than a self-signed cert warning.
On the OVHcloud note from earlier: their panel has improved marginally since last autumn, though I would still not call it pleasant. The filesystem underneath matters more than whatever skin they have draped over it. I run btrfs on my edge box; snapshots before every nginx reload have saved me from my own syntax errors more times than I care to admit.
Member
NPM's just a shiny wrapper for what we did manually on Contabo boxes that cost pocket change. Clicking buttons in a GUI ain't homelab, it's cosplay.
Kiddo you lost me at "powerless consumers." You think firing up a Pi and calling it a day makes you some kind of internet revolutionary? My dude, I watched three generations of wannabe sysadmins burn out on this exact fantasy.
The DDNS crutch is where it falls apart every time. You got your little plug-and-play script, your API calls, your hostname dancing around—then bam, ISP decides port 443 belongs to their VoIP garbage now. Seen it on Hetzner, seen it on Contabo, seen it in my own living room with a cable modem that had more locked ports than a prison.
And that NPM gospel you're preaching? Kiddo that's just cPanel for people who want to feel clever. I ran raw nginx on a ten-year-old thin client when you were still watching unboxing videos. Didn't need a pretty dashboard to tell me what upstream meant.
Your VPN experiment died because consumer-grade pipes aren't built for egress. Not a conspiracy, not "steering"—physics and economics. The asynchronous part you romanticize? That asymmetry still exists. My fiber pulls 900 down and wheezes at 35 up. You want to serve from that? Be my guest. I'll be over here not babysitting a connection that drops every time it rains.
Raspberry Pi is a learning toy. Beautiful little board, garbage for anything touching the real internet. Kiddo wake up.
