I see so many posts and people who run NGINX as their reverse proxy. Why though? There’s HAProxy and Apache, with Caddy being a simpler option.
If you’re starting from scratch, why did you pick/are you picking NGINX over the others?
I see so many posts and people who run NGINX as their reverse proxy. Why though? There’s HAProxy and Apache, with Caddy being a simpler option.
If you’re starting from scratch, why did you pick/are you picking NGINX over the others?
NGINX can really do a lot of things out of the box while being pretty easy to configure. NGINX can serve static files, it can proxy emails, it can do FastCGI, it can do UWSGI, it can do HTTP proxying, you can run Lua code inside NGINX to do things, there’s a module for RTMP live streaming. You can also implement some stuff like external authentication to protect your services/authenticate them at the proxy level. It can also do caching. Not all that useful with all those Rust and Go apps with their own built-in web server but if you run large legacy apps at scale it’s great, you can offload a lot of stuff away from your slow ass PHP app.
Caddy’s simpler but the current battle tested popular option is NGINX.
HAproxy is good at what it does but it’s only good at proxying and simple rules. For the most part, it’s used as a load balancer and router and doesn’t really process the requests itself. It can alter some things in it but it’s limited, and it only does HTTP and TCP. So you can’t really run PHP or Python or Ruby or whatever applications directly behind HAproxy. That makes NGINX a better choice there because NGINX deals with HTTP and only passes the request details to the application which doesn’t have to do HTTP on its own. I usually see HAproxy load balancing to NGINX hosts with some PHP/Python/Ruby app behind them.
Apache is old. It’s gotten better but the way it works just doesn’t reflect most modern use cases. I remember when NGINX popped off like 15 years ago and just how much more resource efficient it was and how happy I was with the upgrade. So it exists and still works but not very popular anymore. It’s a bit easier to set up but also a bit weird with things like mod_php which runs directly inside Apache instead of a dedicated user that can be better sandboxed.
Traefik is getting traction in big part because it fits well with the Docker ecosystem and just sets itself up automatically.
There’s also Envoy if you want some serious proxying and meshing but setting that one up is truely headache inducing.
They’re all pretty good web servers regardless, it comes down to preference. There’s no right choice because everyone’s needs are different.
Not sure why you say haproxy can’t serve python. I do it all the time. You just use something like python waitress and then point haproxy to it’s port.
It depends on what you use on the Python side. Classically that would have been uWSGI or one of the *SGI interfaces, and lately ASGI.
Sure, one can totally make Python apps that serve HTTP directly. The same can be done with PHP (and Ruby and others) as well, but most people still run their PHP through PHP-FPM over FastCGI because you can offload a lot of the work to the much faster NGINX side. A fair amount of apps make use of
X-Accel-Redirectto serve private files, so you don’t tie up a PHP worker for an hour serving the user’s 2GB file.But yes, as those languages all move to async computing and away from worker pools, it’s more common to see those serve HTTP directly, and there’s less and less need for a proxy that supports those other protocols. The async event loop is what made NGINX special when it came out, so naturally languages that moves to that model greatly reduce the need for that as well, they too can easily handle thousands of concurrent connections no problems. Plus these days people slap a CDN in front anyway so static file performance doesn’t matter quite as much.
Ye pretty much. I was just quite astounded at that statement as the AI Horde is basically just a lot of python processes behind a very low powered haproxy server.
Personally, I understand people like to stay with the familiar, which is perfectly fine for a non-demanding service, but when something becomes demanding, I find the haproxy specialization serves better. I wish lemmy deployment by default utilized haproxy myself.
To add something here: HAProxy’s ACLs are more powerful than anything nginx, Apache or even Envoy can do. Of course HAProxy is not a web server but “just” a reverse proxy that speaks HTTP (and TCP) but what you can do with its ACLs is often extremely impressive in its simplicity and elegance. A single-line ACL in HAProxy would require loading additional modules in nginx and writing a screenful of configuration directives. Though the average self-hoster will probably never need any of the power HAProxy offers.
In the past 20 years I have professionally used all four of these as web servers and/or reverse proxies and I am pretty confident that HAProxy beats all others when it comes to request processing. Though Envoy might be getting there.
Having used HAProxy for 15 years commercially, I absolutely agree with this. There are lots of complex features of HAProxy that only a dedicated proxy can provide. The acls, deep packet inspection and stick tables are a few.
Whilst it doesn’t directly “serve” PHP or Python - it’s a load balancer so can just have regular Apache or nginx backends serving content which is arguably its main use case. For homelab this doesn’t always make sense but I would pick nginx for high traffic commercial environments.
Traefik’s marketing as the “Docker reverse-proxy” put me off since I like technologies to stay agnostic of each other (personal preference).
Your arguments are correct, and usually I’d run a separate web server but I suppose for a homelab having less things to manage is great
Traefik does auto discovery and you can register different configuration providers. Don’t need docker? Then don’t use the docker label-based provider. It is really flexible and has sensible defaults. Other than a few quirks in the basic auth support I haven’t had any problems. And at work it powers our globally utilized infrastructure without any hiccups.
It’s possible to write very complex rules/ACLs with HAproxy… stick-tables, ACLs with regexes on whatever HTTP header, source or destination ACLs, map files, geoblocking, lua scripting, load-balancing from round-robin to host header load balancing, dynamic backend servers provisionning through DNS… Not that you can’t do it with Nginx (it started as a reverse-proxy before becoming a jack of all trades), nor that nginx isn’t a great tool (it is!), but HAProxy can do very complex things too. It also follows the good ol’ UNIX philosophy of “one program to do one thing and do it well” and thus doesn’t try to be a webserver, hence why you need a webserver behind it to serve anything from static files to PHP/Python/whatever.