For a while now I’ve been using either haproxy or nginx depending on my needs. I’ve hit instances with both where the functionality I want is in the paid version.

I use Synology integrated reverse proxy, stupidly simple and always works for me (only if IPv6 doesn’t fuck up itself, I can’t fallback to IPv4 because that is CGNATED), if I am missing features that other options have I would like to know :)

I highly recommend npm. It’s also the only one I’ve used, so please keep that in mind.

having tried many in past, i always go back to haproxy. it has everything required as proxy and load balancer while also being very efficient.

I use both, Traefik on my docker host that’s also used for trying out new stacks, and NPM at work for a config that won’t change (ever, probably).

Yes, the NPM web ui is somewhat easier in regard to proxying targets outside Docker.

I use nginx as the internet facing proxy, write my own config and manage it with source control. Also use traefik in docker land with service labels to configure it

In my experience, all the 3 big ones work just fine. Caddy, Traefik, Nginx. I use Nginx.

I’ve been using caddyserver for awhile and love it. Config is nicely readable and the defaults are very good.

Nginx for my intranet because configuration is fully manual and I have complete control over it.

Caddy for the public services on my vps because it handles cert renewal automatically and most of its configuration is magic which just works.

It is unbelievable how shorter caddy configuration is, but on my intranet:

1. I don’t want my reverse proxy to dial on internet to try to fetch new SSL certs. I know it can be disabled, but this is the default.
2. I like to learn how stuff works, Nginx forces you to know more details but it is full of good documentation so it is not too painful compared to Caddy.

I use NPM in a docker container. It could not be easier in my opinion but then again, I did not use any of the alternatives so I might be missing out on something, who knows. I did manage a couple of proxy servers in the past based on Apache and I can tell you that NPM is much easier and logical to me than that.

Just create a compose file and start it. Create DNS records pointing to your NPM IP address/exposed IP and make a host in NPM sending traffic to the right container IP:port. The compose file is super simple, could not be easier. Here’s mine for example:

services:
  nginx-proxy-manager:
    container_name: nginx-proxy-manager
    image: 'jc21/nginx-proxy-manager:latest'
    restart: always
    ports:
      - '80:80'
      - '443:443'
    volumes:
      - ./data:/data
      - ./letsencrypt:/etc/letsencrypt

I just make sure ports 443 and 80 are exposed on my router so DNS records can point to that IP adrdess. All traffic on port 80 gets re-routed to 443.

I’m probably stating all the obvious things here 😀

This the main reason I switched from traefik, I can have certificates on all my internal stuff and not just on my docker host. I personally love NPM but maybe I’ll give NPMPlus a try, I have never heard of it.

Caddy all the way!

Traefik is a PITA.

Caddy all the way. If you build it with Docker support (or grab the prebuilt), you can use docker container names to reverse proxy using names instead of any IP addresses or ports. It’s nice because if the IP updates, so does caddy. All automatically.

Here’s what my caddyfile looks like;

{
        acme_dns cloudflare {key}
}

domain.dev {
        encode zstd gzip
        root * /var/www/html/domain.dev/
        php_fastcgi unix//run/php/php8.1-fpm.sock
        tls {
                dns cloudflare {key}
        }
}
*.domain.dev {
        encode zstd gzip
        tls {
                dns cloudflare {key}
        }
        @docker host docker.domain.dev
        handle @docker {
                encode zstd gzip
                reverse_proxy {portainer}
        }
        @test host test.domain.dev
        handle @test {
                encode zstd gzip
                reverse_proxy 127.0.0.1:10000
        }
        @images host i.domain.dev
        handle @images {
                encode zstd gzip
                reverse_proxy 127.0.0.1:9002
        }
        @proxy host proxy.domain.dev
        handle @proxy {
                encode zstd gzip
                reverse_proxy proxy
        }
        @portal host portal.domain.dev
        handle @portal {
                encode zstd gzip
                reverse_proxy portal
        }
        @ping host ping.domain.dev
        handle @ping {
                encode zstd gzip
                respond "pong!"
        }
}

DNS hosted by cloudflare but because caddy handles ACME certs, all the subdomains automatically get SSL.

I like NPM, it’s simple, but also allows for more complex configs as well if needed. I run it in its own LXC because I have other non-dockerized things that are exposed.

it seems easier to manage stuff not in docker

Read into Traefik’s dynamic configuration. Adding something outside of Docker is as easy as adding a new config file in the dynamic configuration folder. E.g. jellyfin.yml:

http:

  routers:

    jellyfin:
      rule: Host(`jellyfin.example.org`)
      entrypoints: websecure
      tls:
        certResolver: le
      service: jellyfin

  services:

    jellyfin:
      loadbalancer:
        servers:
          - url: "http://192.168.1.5:8096/"

The moment you save that file it will be active and working in Traefik.