This is an annoying quirk in the way docker handles networking between containers and I couldn’t find a good solution for this issue when I was trying out network_mode. I just couldn’t find a way to set docker up to automatically restart the dependent container. You can achieve this with services defined in the same stack (using depends_on), but I don’t know if it’s possible with your current setup.

That’s why I mentioned manual routing in my other reply. It’s annoying to set up, but more convenient because you avoid having to manage restarts (or figuring out how to get docker to do it, which may not be possible in this case).

Uhh, I think you might be confused. Let me explain a bit more:

1. Services and Containers aren’t the same thing. The distinction usually doesn’t matter in typical self-hosting scenarios, but in this case it does.

In short: Services are what you define in a compose file; Containers are what you spin up based on those service definitions.

2. network_mode is a service attribute and it can be defined for each service separately.
3. network_mode: "service:{name}" requires the service being referenced to be part of the same stack. This is probably what you were thinking of when you wrote this reply.
4. network_mode: "container:{name}" can freely reference any preexisting container. This helps you achieve what you want. You can define your gluetun container independently, along with any services you might want to be part of the same stack, and give it a unique identifier using container_name: myIndependentGluetun. After spinning it up, run your Qbittorrent container or whatever service you want to route through the gluetun container after adding network_mode: "container:myIndependentGluetun".

You could also route it manually. That’s a more advanced solution, but it’s more convenient than the network_mode approach. More on this here: https://discuss.tchncs.de/post/19039498

You could use network_mode: "container:{name}" instead of service:{name}. See here https://docs.docker.com/reference/compose-file/services/#network_mode

Service definitions have to be defined in the same compose file or merged into one file at some point in order to be able to reference each other. Containers don’t have that restriction.

Using Nicotine+ on my server. https://github.com/fletchto99/nicotine-plus-docker

Soulseek, among others. Putting my ~400 GB classical music collection out there.

I think you already have a kill-switch (of sorts) in place with the two Wireguard container setup, since your clients lose internet access (except to the local network, since there’s a separate route for that on the Wireguard “server” container") if any of the following happens:

• “Client” container is spun down
• The Wireguard interface inside the “client” container is spun down (you can try this out by execing wg-quick down wg0 inside the container)
• or even if the interface is up but the VPN connection is down (try changing the endpoint IP to a random one instead of the correct one provided by your VPN service provider)

I can’t be 100% sure, because I’m not a networking expert, but this seems like enough of a “kill-switch” to me. I’m not sure what you mean by leveraging the restart. One of the things that I found annoying about the Gluetun approach is that I would have to restart every container that depends on its network stack if Gluetun itself got restarted/updated.

But anyway, I went ahead and messed around on a VPS with the Wireguard+Gluetun approach and I got it working. I am using the latest versions of The Linuxserver.io Wireguard container and Gluetun at the time of writing. There are two things missing in the Gluetun firewall configuration you posted:

• A MASQUERADE rule on the tunnel, meaning the tun0 interface.
• Gluetun is configured to drop all FORWARD packets (filter table) by default. You’ll have to change that chain rule to ACCEPT. Again, I’m not a networking expert, so I’m not sure whether or not this compromises the kill-switch in any way, at least in any way that’s relevant to the desired setup/behavior. You could potentially set a more restrictive rule to only allow traffic coming in from <wireguard_container_IP>, but I’ll leave that up to you. You’ll also need to figure out the best way to persist the rules through container restarts.

First, here’s the docker compose setup I used:

networks:
  wghomenet:
    name: wghomenet
    ipam:
      config:
        - subnet: 172.22.0.0/24
          gateway: 172.22.0.1

services:
  gluetun:
    image: qmcgaw/gluetun
    container_name: gluetun
    cap_add:
      - NET_ADMIN
    devices:
      - /dev/net/tun:/dev/net/tun
    ports:
      - 8888:8888/tcp # HTTP proxy
      - 8388:8388/tcp # Shadowsocks
      - 8388:8388/udp # Shadowsocks
    volumes:
      - ./config:/gluetun
    environment:
      - VPN_SERVICE_PROVIDER=<your stuff here>
      - VPN_TYPE=wireguard
      # - WIREGUARD_PRIVATE_KEY=<your stuff here>
      # - WIREGUARD_PRESHARED_KEY=<your stuff here>
      # - WIREGUARD_ADDRESSES=<your stuff here>
      # - SERVER_COUNTRIES=<your stuff here>
      # Timezone for accurate log times
      - TZ= <your stuff here>
      # Server list updater
      # See https://github.com/qdm12/gluetun-wiki/blob/main/setup/servers.md#update-the-vpn-servers-list
      - UPDATER_PERIOD=24h
    sysctls:
      - net.ipv4.conf.all.src_valid_mark=1
    networks:
      wghomenet:
        ipv4_address: 172.22.0.101

  wireguard-server:
    image: lscr.io/linuxserver/wireguard
    container_name: wireguard-server
    cap_add:
      - NET_ADMIN
    environment:
      - PUID=1000
      - PGID=1001
      - TZ=<your stuff here>
      - INTERNAL_SUBNET=10.13.13.0
      - PEERS=chromebook
    volumes:
      - ./config/wg-server:/config
      - /lib/modules:/lib/modules #optional
    restart: always
    ports:
      - 51820:51820/udp
    networks:
      wghomenet:
        ipv4_address: 172.22.0.5
    sysctls:
      - net.ipv4.conf.all.src_valid_mark=1

You already have your “server” container properly configured. Now for Gluetun: I exec into the container docker exec -it gluetun sh. Then I set the MASQUERADE rule on the tunnel: iptables -t nat -A POSTROUTING -o tun+ -j MASQUERADE. And finally, I change the FORWARD chain policy in the filter table to ACCEPT iptables -t filter -P FORWARD ACCEPT.

Note on the last command: In my case I did iptables-legacy because all the rules were defined there already (iptables gives you a warning if that’s the case), but your container’s version may vary. I saw different behavior on the testing container I spun up on the VPS compared to the one I have running on my homelab.

Good luck, and let me know if you run into any issues!

EDIT: The rules look like this afterwards:

Output of iptables-legacy -vL -t filter:

Chain INPUT (policy DROP 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source               destination
10710  788K ACCEPT     all  --  lo     any     anywhere             anywhere
16698   14M ACCEPT     all  --  any    any     anywhere             anywhere             ctstate RELATED,ESTABLISHED
    1    40 ACCEPT     all  --  eth0   any     anywhere             172.22.0.0/24

# note the ACCEPT policy here
Chain FORWARD (policy ACCEPT 3593 packets, 1681K bytes)
 pkts bytes target     prot opt in     out     source               destination

Chain OUTPUT (policy DROP 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source               destination
10710  788K ACCEPT     all  --  any    lo      anywhere             anywhere
13394 1518K ACCEPT     all  --  any    any     anywhere             anywhere             ctstate RELATED,ESTABLISHED
    0     0 ACCEPT     all  --  any    eth0    dac4b9c06987         172.22.0.0/24
    1   176 ACCEPT     udp  --  any    eth0    anywhere             connected-by.global-layer.com  udp dpt:1637
  916 55072 ACCEPT     all  --  any    tun0    anywhere             anywhere

And the output of iptables -vL -t nat:

Chain PREROUTING (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source               destination

Chain INPUT (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source               destination

Chain OUTPUT (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source               destination
    0     0 DOCKER_OUTPUT  all  --  any    any     anywhere             127.0.0.11

# note the MASQUERADE rule here
Chain POSTROUTING (policy ACCEPT 0 packets, 0 bytes)
 pkts bytes target     prot opt in     out     source               destination
    0     0 DOCKER_POSTROUTING  all  --  any    any     anywhere             127.0.0.11
  312 18936 MASQUERADE  all  --  any    tun+    anywhere             anywhere

Chain DOCKER_OUTPUT (1 references)
 pkts bytes target     prot opt in     out     source               destination
    0     0 DNAT       tcp  --  any    any     anywhere             127.0.0.11           tcp dpt:domain to:127.0.0.11:39905
    0     0 DNAT       udp  --  any    any     anywhere             127.0.0.11           udp dpt:domain to:127.0.0.11:56734

Chain DOCKER_POSTROUTING (1 references)
 pkts bytes target     prot opt in     out     source               destination
    0     0 SNAT       tcp  --  any    any     127.0.0.11           anywhere             tcp spt:39905 to::53
    0     0 SNAT       udp  --  any    any     127.0.0.11           anywhere             udp spt:56734 to::53

Gluetun likely doesn’t have the proper firewall rules in place to enable this sort of traffic routing, simply because it’s made for another use case (using the container’s network stack directly with network_mode: "service:gluetun").

Try to first get this setup working with two vanilla Wireguard containers (instead of Wireguard + gluetun). If it does, you’ll know that your Wireguard “server” container is properly set up. Then replace the second container that’s acting as a VPN client with gluetun and run tcpdump again. You likely need to add a postrouting masquerade rule on the NAT table.

Here’s my own working setup for reference.

Wireguard “server” container:

[Interface]
Address = <address>
ListenPort = 51820
PrivateKey = <privateKey>
PostUp = iptables -A FORWARD -i %i -j ACCEPT; iptables -A FORWARD -o %i -j ACCEPT; iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE
PostUp = wg set wg0 fwmark 51820
PostUp = ip -4 route add 0.0.0.0/0 via 172.22.0.101 table 51820
PostUp = ip -4 rule add not fwmark 51820 table 51820
PostUp = ip -4 rule add table main suppress_prefixlength 0
PostUp = ip route add 192.168.16.0/24 via 172.22.0.1
PostDown = iptables -D FORWARD -i %i -j ACCEPT; iptables -D FORWARD -o %i -j ACCEPT; iptables -t nat -D POSTROUTING -o eth0 -j MASQUERADE; ip route del 192.168.16.0/24 via 172.22.0.1

#peer configurations (clients) go here

and the Wireguard VPN client that I route traffic through:

# Based on my VPN provider's configuration + additional firewall rules to route traffic correctly
[Interface]
PrivateKey = <key>
Address = <address>
DNS = 192.168.16.81 # local Adguard
PostUp = iptables -t nat -A POSTROUTING -o wg+ -j MASQUERADE #Route traffic coming in from outside the container (host/other container)
PreDown = iptables -t nat -D POSTROUTING -o wg+ -j MASQUERADE

[Peer]
PublicKey = <key>
AllowedIPs = 0.0.0.0/0
Endpoint = <endpoint_IP>:51820

Note the NAT MASQUERADE rule.

I set it up manually using this as a guide. It was a lot of work because I had to adapt it to my use case (not using a VPS), so I couldn’t just follow the guide, but I learned a lot in the process and it works well.

I’ve tried both this and https://github.com/jmorganca/ollama. I liked the latter a lot more; just can’t remember why.

GUI for ollama is a separate project: https://github.com/ollama-webui/ollama-webui

If anyone wants to achieve something similar without using Tailscale or with alternative VPN providers, the setup outlined in this LSIO guide is pretty neat: https://www.linuxserver.io/blog/advanced-wireguard-container-routing

Edit: Don’t be intimidated by the word “advanced”. I struggled with this a bit at first (was also adapting it to use at home instead of on a VPS that’s tunneling to home) but I got it working eventually and learned a lot in the process. Willing to assist folks who want to set it up.

Ooooh, good catch. I assumed “it’s been giving me the same message for over an hour” to mean that they’ve been monitoring the logs, not running in interactive mode. O_O

That log entry is unrelated to whatever issues you’re having. That’s what the default docker-compose.yaml uses for health checks:

  healthcheck:
      test: wget -nv --tries=1 --spider http://127.0.0.1:3000/api/v1/comments/jNQXAC9IVRw || exit 1
      interval: 30s
      timeout: 5s
      retries: 2

The fact that it returns a 200 probably means that Invidious is properly up and running. Could you elaborate further on what you mean by “setup isn’t completing”? How are you trying to connect to the web UI? Sharing your docker-compose.yaml might help us debug as well.

Edit: I just noticed that the default compose file has the port bound to localhost:

    ports:
      - "127.0.0.1:3000:3000"

which means you won’t be able to access it from other machines inside or outside your network. You’d have to change that to - "3000:3000" to enable access for other machines.

I’ve never heard of NextCloud Cookbook before. Looking at its Github page, it says it’s “mostly for testers” and is unstable, so no point in even considering it for regular use at this point in time. Besides, I’m assuming you’d need to have your own instance of Nextcloud up and running to use it; I don’t use Nextcloud.

As for Grocy and other more mature alternatives (Tandoori also comes to mind), I think I initially went with Mealie because it had the most pleasant UI out of all of them. I liked it and found that it satisfied all of my requirements, so I just kept using it.

One thing I need to publicly expose is my own instance of Mealie. It’s a recipe manager that supports multiple users. I share it with family and friends, but also with more distant acquaintances. I don’t want to have to provide and manage access to my network for each and every one of them.

I run Koreader on a Kobo Libra 2. I just connect to my OPDS catalogue on my Calibre-Web instance. It’s not exactly a sync setup; it just gives me access to my library whenever I need to download something, and that covers my needs. There are several other sync options; check out Koreader’s features here: https://github.com/koreader/koreader/wiki

If you like it and decide you want to it, go through the list of supported devices and see what sort of sync capabilities are available for them (support for Kobo devices seems to be the best/have the most options).

+1 for Joplin. I have a different setup since I don’t use Nextcloud: Run Joplin server in a docker container and back up the volumes mapped to it (as well as those of other containers) with rsync.