GitHub - subdavis/selfhosted: docker compose + traefik + tailscale · GitHub
Skip to content

subdavis/selfhosted

Repository files navigation

docker selfhosted services

with docker-compose and traefik

Uptime Robot ratio (30 days) Build Status

This repo contains my production docker services accessible from anywhere over HTTPS using traefik. These services (and others) run on a single server. It used to be rootless-mode but slirp4net was too slow and too much of the docker advanced configuration (permissions flags, mostly) were missing.

  • Jellyfin
  • Navidrome
  • Linkding
  • Miniflux
  • Sonarr, Radarr, Prowlarr
  • Calibre Web
  • Kobo book downloader (kobodl)
  • Transmission torrent server
  • AdGuard Home DNS
  • Drone CI and runner
  • Backrest (restic)
  • Watchtower
  • Cloudflare DNS Automation
  • Dozzle

Documentation

I've also written some intermediate to advanced generic usage docs for traefik, docker, pihole, and home networking. These articles are generally applicable, but some may be more useful than others.

More great documentation.

Prerequisites

  • A recent version of ubuntu server with Docker CE installed (see below)
  • A router or firewall capable of dnsmasq. I use a Ubiquiti EdgeRouter X.
  • A domain name.
  • A cloudflare account.

Home network prep

  • You need to make sure that ports 80 and 443 are port-forwarded through your router to whatever host this will be on.
  • Your server should be assigned a static private IP by DNS. ifconfig will list your interfaces.
  • Refer to the docker-pi-hole docs and my docs for further network setup related to that service. Even though I use AdGuard Home, those docs are relevant.

DNS Configuration

UPDATE: This is now done automatically with Docker Traefik Cloudflare Companion. Instructions below are left as an explanation of how this works.

In this setup, each container's service will serve from a different subdomain of your Cloudflare hosted zone dyndns subdomain.

  • Create an A record for core.mydomain.com to point to your public IP.
  • For each service, you'll need to create CNAME records for each service.mydomain.com to point to core.mydomain.com because all of your services are running on the same host but the host needs to be able to do virtual host routing based on domain name.
  • Your services will be publically available on https://servicename.mydomain.com.

Dynamic DNS (recommended)

Resolving the IP address of your home network is annoying because most DNS providers change your IP every now and again. Services like No-IP combat this, but they aren't the most reliable. However, setting DNS programatically is pretty easy with Cloudflare API.

Installation

  1. start with ubuntu lts
  2. Enable Unattended Upgrades
  3. clone this repo
  4. Sign into any private docker registries
  5. install docker a Understanding UID remapping a. ignore the env exports it says to set, see below
  6. make sure UsePAM yes is set in /etc/ssh/sshd_config read more
cd selfhosted
cp .env.example .env # edit this

# make mount points
mkdir /media/local /media/primary /media/secondary

# install mounts
systemctl link media-primary.mount
systemctl link media-secondary.mount

# install logrotate
systenctl --user link $HOME/selfhosted/logrotate.timer
systenctl --user link $HOME/selfhosted/logrotate.service
systemctl --user enable logrotate.timer --now

# enable traefik logrotate
cp etc/traefik-logrotate.conf /etc/logrotate.d/traefik

# Add to .profile
# export DOCKER_HOST=unix://$XDG_RUNTIME_DIR/docker.sock
nano .profile

Set up docker daemon.json. Otherwise, you may end up with subnet ranges inside your containers that overlap with the real LAN and make hosts unreachable.

{
    "default-address-pools": [
        {"base":"172.16.0.0/16","size":24},
        {"base":"172.20.0.0/16","size":24}
    ]
}

Sysctl customizations

# Allow containers to bind privileged ports (e.g. 80, 443)
echo 'net.ipv4.ip_unprivileged_port_start=0' | sudo tee -a /etc/sysctl.d/99-sysctl.conf

# Increase max_map_count for Jellyfin/Elasticsearch
echo 'vm.max_map_count=262144' | sudo tee -a /etc/sysctl.d/99-sysctl.conf

sudo sysctl --system

Edit /lib/systemd/system/user@.service to include dependencies on mounts

[Unit]
Requires=user-runtime-dir@%i.service media-primary.mount media-secondary.mount

Automatic deployments and drone

  • Create a github api app. Follow drone setup instructions.
  • Make sure the user filtering config is set correctly so other users can't log in
  • Add secrets ssh_key, ssh_host, ssh_user for your deploy user.
  • Open drone.yourdomain.com and finish configuring your repo.

Adguard DNS

You may need to disable ubuntu's default dns service and remove resolf.conf read more.

After disabling systemd-resolved.service, I ususally set a different DNS server in /etc/resolv.conf so that DNS doesn't break when I screw up the stack.

systemd-resolve --help is your friend.

Static /etc/resolv.conf (required)

With systemd-resolved disabled, /etc/resolv.conf must be a real file pointing at a proper upstream DNS server — not 127.0.0.1/AdGuard. Docker's embedded DNS resolver on user-defined networks (127.0.0.11) reads the host's resolv.conf at daemon start to find upstreams; if it can't reach an upstream, every container's DNS breaks. Pointing at AdGuard creates a chicken-and-egg: AdGuard can't start without Docker DNS, and Docker DNS can't resolve anything without AdGuard.

An Ubuntu release upgrade (e.g. 22 → 24) will re-symlink /etc/resolv.conf to stub-resolv.conf even though systemd-resolved is disabled, leaving a dangling symlink and breaking container DNS. Fix:

sudo rm /etc/resolv.conf
sudo tee /etc/resolv.conf <<'EOF'
nameserver 1.1.1.1
nameserver 8.8.8.8
EOF
sudo chattr +i /etc/resolv.conf   # prevents future upgrades from clobbering it
sudo systemctl restart docker      # re-read upstream DNS

Also disable AdGuardHome's Rate Limiting because miniflux will trigger it and throw server misbehaving errors.

Tailscale as Exit Node

I've switched to tailscale. In Tailscale DNS admin settings, I set the exit node's Tailnet IPv4 address as the DNS server. There should be no need to modify firewall rules on the exit node.

tailscale up --advertise-connector --accept-dns=false --advertise-exit-node --advertise-routes=192.168.48.0/20,63.231.151.187/32 --ssh

Troubleshooting: I've had issues with outdated clients not resolving DNS properly when connected to the exit node that were fixed by updating to the latest version.

Archived: WireGurad and subnet docs

  • use wg-quick for simplicity

  • May need to install or symlink resolvconf

  • Need to avoid overlapping subnets.

  • Set MTU down to 1280 for issues with cellular networks, on BOTH sides of the connection.

    • Update: As of September 19, had to drop to 1250 for TMobile LTE to work....
  • My subnet is 192.168.48.0/20

  • The mask is 255.255.240.0

  • The default LAN will be 192.168.52.0

  • The gateway is 192.168.52.1

Gateway: 11000000.10101000.0011 | 0100.00000001
Mask:    11111111.11111111.1111 | 0000.00000000
  • The upper 4 bits will be used for VLANs (16).
  • The lower 8 shoud belong to a single VLAN.

Using wireguard:

sudo systemctl enable wg-quick@peerN --now

I have aliases wgup and wgdown for this in my .bashrc.

IPv6

Some references I encountered while rolling out ipv6.

My full edgerouter config

Other useful nonsense

# set own IP, delete set
ifconfig eth0 192.168.1.5 netmask 255.255.255.0 up
ifconfig en1 delete 192.168.1.5

Contributors

Languages