I'm hosting few hobbyist public websites. One is the library of hymns, songs, ... - mostly PDF, MP3, MIDI.
OpnSense router -> Traefik -> Docker container.
That Docker container runs on NAS, which is connected via 1Gbps port.
Suddenly, last week I started to get over 1000 RPS. Was it scanner, parser, crawler, DDOS? I can't tell, as user agent was default Mac OS.
It was using ~1Gpbs bandwidth. From many IPs. If I would have it hosted at AWS / Azure, it would cost me > $40 / hour for the bandwidth. Unless I use some of their CDN.
My router CPU went high, but it handled it ok. NAS CPU went high, but it handled it ok. Traefik loaded the server heavily: it had to both receive and send 1Gbps. And write access logs for all requests. My Alloy & Loki logging system got loaded heavily.
NAS already have cache. I tried to enable cache on Traefik. It went OOM in 2 minutes: https://github.com/darkweak/souin/issues/844
When I searched, I found that Traefik cache with proper memory handling is available only in Enterprise version.
Solved the issue by putting the site behind Cloudflare proxy and enabling it's magic to protect from the bot.
The best is almost not to do it at all. Just have a docker image serve http/1.1 and document that a reverse proxy is required to expose the service.
There are simply too many ways to do it and every sys admin/hobbyist has their own preferred way.
For the anonymous and authenticated caching issues the author goes into, I think once again it is useless for self hosting. Just embed a performant web server like nginx for example that is pre configured to serve static files. Use proper headers, and predictable/simple paths. Self-hosted versions are going to be low traffic and just need to work.
Have advanced settings for more control, but keep the default simple and working out of the box with minimal dependencies.
If you want to put caddy in front of yours? Great it works.
For me, running a much larger setup? Great, it also works.
For users who never expose it beyond an IP address on lan? Great it also works.
You say caddy. The next person will say nginx-proxy-manager. The next will say DNS challenge let's encrypt behind wireguard VPN. The next will say Traefik. The next will say CloudFlare tunnels.
Everyone has their preferred solution and its always the best and simplest.
I think at this point it's not a hobby, it's a burden.
Somebody hasn't read the docs and doesn't understand how nginx's "if" works, because it's fantastically useful.
I guess that breaks the plug-and-play nature, but you can't expect something internet-facing to be plug-and-play. (Maybe that's the real travesty here.)
I just wish nginx / caddy / etc had better out-of-the-box support for letsencrypt. It would be way easier to set up if it was just built in to nginx. Let me specify my domain in my nginx config, and just make it work.
An art professor I knew used to say about overwrought paintings "too much sugar for a dime". Technical people are often prone to this as well.
We'got all the world's knowledge compressed and queryable with LLMs: there are way less configuration issues that I cannot solve today than used to be the case.
Did those in charge of those standards royally fuck up HTTPS/TLS? Of freakin' course they did. The interests of the big players, not the small ones, have been taken care of. And the result is an ultra complex mess of fucktarded standards.
But it is what is: it's what we have. And we'll make it work.
I personally don't like apps that ship the entire sink (complex stack) when I already have an opinionated setup that already includes TLS/access control/etc.
Hobby scale you said ? It doesn't have to be that complicated.
My homelab is also currently ~40 nix service VM's plus a handful of Ubuntu, etc long tail, and some external DO droplets for NAT, and a more robust vlan config than I'd ever been able to put time towards before... which I haven't had to do more than a dozen hours of manual setup on IN TOTAL thanks to Fable, Sol and local Qwen.
There's truth in this post, but I doubt it's longevity.
It's DEFINITIVELY overkill, and fills up fast, but nix making everything near fully declarative to roll fwd/back or shuffle between proxmox hosts at any time makes for a clean and easy mental model. Both for me and the AI, because it can infinitely retry a nix flake config until it gets it perfect.
This is the way (while also allowing people to use their own proxies etc., but being clear about those not being officially supported).
Ship your back end container, your front end container, tell people how to set up the off the shelf DB containers or whatever else your software needs and if you don’t need 20 different components like Sentry self-hosted versions do, you’ll be fine. The DB can be a container or not. The reverse proxy can be a container or not. Hell, if someone wants to build the back end or front end from source and serve them differently they can, just on their own time. That’s pretty close to the 12 Factor App principles.
Just avoid a lot of complexity when possible: do traditional SSR, or with some progressive enhancements or what I like to do in this day and age (unless maximum accessibility is paramount) only use your back end for the API and have a traditional SPA, none of that complex Next.js stuff. Your front end just becomes a bundle of files that you can put in a web server container to be proxied by whatever the ingress is. At most the Docker entry point would change some values in config.json that the app loads on startup in the browser.
If someone asks for Kubernetes support or whatever, just wish them the best of luck in setting it up themselves!
I can fiddle with the page content in Developer Tools to make it work, but frankly, WTF, especially since Immich expects people to set up a reverse proxy for HTTPS access on their own and does not help them in any way (like providing a pre-configured caddy in their compose.yaml, I don't know)
You could theoretically upstream the fixed compose.yaml if you felt like it, but the project seemed unusually hostile towards the one fix I tried to upstream.
photos.example.com {
reverse_proxy 127.0.0.1:8080 {
header_up X-Real-IP {remote_host}
}
}It’s fine to publish recommendations how to tune configurations for specific usages. But I only need the app, not the various things that you may think I need. And I really want the app to only solve its essential problem. It should not take care of the IT part other than providing configuration samples.
I just call /publish-local-docker subdomain.mydomain.local
And if there isn't a docker compose, it will create one, and then run it, and update caddy server on its own.
Unless you also want to deal with security, backups, GDPR compliance, etc.
No other.
photos.example.com {
reverse_proxy 127.0.0.1:8080
}
Do you want a caddy service embedded with every web app you self host or are you running a single one serving all your apps?
From my understanding if you want acme http challenges to just work with let's encrypt you probably need to use a single, separated caddy to front all your apps.
In that case if I ship a caddy service pre configured to do TLS termination with my app it's just more trouble for you no?