I just use Docker Compose with explicit port mappings and a .env file for the base port. If my provider changes the range I change one variable and redeploy. Not fancy but I am not fancy either.
services:
nginx:
ports:
- "${BASE}:80"
- "${BASE_SSL}:443"BASE=30000, BASE_SSL=30001, done.
Of course you did.
Docker Compose for 20 ports is how you end up with 20 containers when you needed 5. I run bare nginx and a handful of static binaries. My entire port map fits in a 12-line script that I have memorized.
Kubernetes solved problems I never had and created ones I do.
In Turkey we have same issue with providers giving high ports. I use HAProxy on a cheap Hetzner VPS in Finland (€3.50/month) as my entry point. Finland to my NAT in Germany is 25ms. Total cost still less than one "normal" VPS with dedicated IP.
The trick is `accept-proxy` on HAProxy side so you keep real client IPs in logs.
I have understand now. So I need a second VPS for the real IP, or use something like Cloudflare Tunnel. I will check prices because my budget is tight. Thank you all for explain SNI to me, I was confuse.
Cloudflare Tunnel works and is free but remember: you are now locked to Cloudflare's network path. I measured 15% slower speeds to EU users compared to direct routing from my Hetzner proxy. For a game server that matters. For a blog, who cares.
Member
This is why I self-host the proxy. My Rust config generator outputs WireGuard configs too, so the tunnel is mine end to end. 847 lines was a lie, it is 903 now. I added Prometheus metrics.
I am following this thread because I have the exact same setup. Has anyone tried using a IPv6-only VPS as the front end? Some providers give those for $1/month and you get a full /64. Then use NAT64 or something for IPv4 users?
Might be cheaper than dual-stack proxy.
Tried it. Pain. Half the mobile networks in the US still do not do IPv6 properly. Your game server players will complain. Stick to dual-stack for anything user-facing.
TAYGA on a $3 Racknerd box worked for me as experiment. Throughput was 200 Mbps, CPU-bound on the translation. Not worth for production. Ampere handles it better but why waste the cores.
ARM efficiency does not help if you create unnecessary work.