Could also be docker network-config. Docker should by default use the hosts resolver config if there’s nothing in /etc/resolve.conf
You can also supply dns server on the docker command or in your compose file if you’re using compose.
As a last resort you can enter server and ip i the container’s /ets/host file if the ip is static. But that’s gone once you rebuild the image.
Or maybe there’s env on the container you use for dns
That’s not too easy methinks. You get kvm vm-s though gui/shell/api with proxmox, but no docker (they use lxc). Unless you set up a podman or similar inside it.
If you’re made of money there’s always https://www.vmware.com/products/vsphere/vsphere-with-tanzu.html
Other than that I’d say go with a xen (xcp-ng). Proxmox or esxi host, and spin up a vm as docker host.
I ditched docker in my latest setup, just running 2 machines in a proxmox cluster. I like lxc - as it’s got the footprint of docker and behaves like a vm
On the host of the nginx rev proxy or in nginx config files. Something seems to block the lookup from name to ip, as ip works you know the proxy works. Check dns config and nginx config on that host
As you can forward by ip but not by name it sounds like resolver issue.
I think your best option would be a pi 4 compute for high speed bus. https://www.waveshare.com/cm4-nas-double-deck-c4a.htm
Its also a vital part of software defined networking
Head-less is without screen (head). You’d need to install the desktop of choice. You only get a screen. It’s for server admin not daily use - even with a fancy one like that.
This is awesome, I just want to point out that once you have a pikvm connected it’s no longer headless. That’s kinda the whole idea about any kvm - to provide screen(s), keyboard and mouse to servers so they’re no longer headless
And do use their guide for spf, dkim and dmarc
The Debian installer can be pre-seeded and be automated. You can use cloud-init for non cloud installs but why would you? Preseed or use fai and let your config system handle the rest.
I get that you love this board and think that “the establishment” is evil. But you come off as someone not having the knowledge to back your assumptions.
Sometimes this will be the right board, sometimes a Pi is better. And sometimes 2-3 microcontrollers are a better fit. But the choice should not be based on telemetry in an optional imager, or the fact that your headless setup requires editing of config files.
How is a pi (or other single-board computers) less critical than “a full system”? Do you have any idea how many pi’s are out there running serious stuff? Where I work I bump into them all over - including in security systems and door-access.
This one has two 2.5gb ports, 8 to 32gb ram. This is serious stuff for an sbc, clearly overkill for your pihole install. What’s not equally serious with banana pi is support. I went to their wiki, it lists Android and Debian (previous version) “images” but no download links, so it’s hard for me to verify that this board boots with sshd running or not. Like I said Debian does not, and for a good reason. Raspberry pi os pulls from raspbian, and they pull from Debian.
You can run Ubuntu LTS, fedora or others on your pi.
The telemetry is bad news - soon we will be out of food because someone knows what size of sd-cards you use, and the number of installs you do. So better go buys a silly board, track down some ancient image of an install someone did at some point where they managed to compile the nic drivers and include the binary blob. Because nobody gets to force you to add an empty file to your sd-card!
The extra menu in the flasher does the magic on the sd-card. I’ve been setting up headless pi’s since before 3b came out, and the same options are available today.
The idea that ssh being enabled by default is reasonable is just like your opinion. Did you know you have to enable it during installation on both Debian and canonicals derivative? Maybe it’s still on by default on fedora (with root login enabled to help you!)
If editing your config is fiddling then I struggle to see your use of an sbc.
This and support. My dad could set up a pi, and he doesn’t know what a kernel is or how to compile.
This is not true
This - why add another machine into the equation?
Or set up an account at Tailscale or similar. They let you have like 10 machines in your network w/o payment
Or route to your home network through the vpn interface on the VPS. So you can reference the windows ip, rather than NAT
Your current setup is very complicated. I did not check your rules at all but maybe you are setting up forwarding on your servers lan interface rather than the WireGuard interface.
Also you don’t say much about how the VPS setup is. Do you Nat the other working services? What IP/host do you forward to? Are you rdp’ing from the VPS or is that also some form of forwarder/revproxy?
Yes - I like bind9 with views so I can serve external and internal from same instance. As I only have services for my own use 1 ns on my dynamic ip is enough for my home subdomain.
Bind9 has ok scripting possibilities with rndc and nsupdate.
Why the unmanaged switch? Putting servers on a switch I like to know that the switch can handle VLAN and link aggregation.
Right - so the upstream server is a docker container on the same machine, and you proxy the connection to the servers up on the port forwarded through the magic docker iptables thingy. It might be here that you get the connection closed - maybe check logs on that. Don’t recall if it’s logged by default or you have to set it up
There’s also the possibility of the web service not being proxy friendly
I’m not very familiar with nginx, I use haproxy for my reverse proxy or load balancing needs.
Do the 504 get logged on the proxy? From your screen grabs it seems that nginx have a working connection to the upstream server.
Maybe that website needs special settings that nginx can’t handle or is not set up for?
If you can - separate host and storage. Run what ever hyper visor you like - Xcp-ng is also good. Any nas is good