Uncloud on Miek Gieben

Uncloud Server Setup

Tue, 07 Jul 2026 09:21:52 +0200

I’m thinking about “upgrading” my server from Ubuntu to Fedora and thereby also making Uncloud a part of my routine, but…… This can be done, but doing a container style setup, does mean I need some extra infra, like a container registry, also my DNS server needs to set source IP addresses for notifies unsure if that will still work.

And Caddy with L4 forwarding and getting TLS certificates can cause issues, because it interferes with each other (i.e. Caddy is starting, but not L4-forwarding, aks for a certificate, DNS is still not working, cert is not handed out - that kind of thing).

Uncloud Unpush

Mon, 15 Jun 2026 11:21:52 +0200

Unpush is an experimental attempt to write automatic CI/CD in Uncloud, it either works via web hooks or polls a repository for changes. When seen it redeploys a service via its compose.yaml file from the repository.

Now of course this works for GitHub, but we have GitLab and it reimplements a bunch of code from uncloud, which can all be removed if you just use the uc-tool.

I will probably butcher the code quite a bit, so seems fair to push this to a different repository at some point.

Uncloud Tips

Wed, 10 Jun 2026 18:38:34 +0200

Uncloud’s docs are great, but a few things are buried a bit. Hence a simple list of things that can improve your interactions with Uncloud.

SSH ProxyJump

Sometimes you can not directly connect to a cluster, you first have to SSH into a host that is directly connected to a cluster. If this is the case use a SSH proxy jump (see ssh(1)). This allows you to, e.g. use a local docker to build images that can then be pushed to the cluster.

Uncloud Prometheus, part II

Thu, 21 May 2026 11:22:52 +0200

Had some fun playing with my fork of Uncloud. And making metrics work.

Metrics are now part of Uncloud proper - uncloudplus can now drop it.

In prometheus we use dns_sd_configs to pick up new things automatically:

- job_name: uncloud
  dns_sd_configs:
  - names: ["m.internal"]
   type: A
   port: 51090
- job_name: caddy
  dns_sd_configs:
  - names: ["caddy.internal"]
    type: A
    port: 2019

For caddy metrics, we hack around a little… in the x-caddy file we deploy for Caddy add:

metrics {
    per_host
}

But this serves metrics on localhost inside the container. So we need to export this, we can’t add this snippet:

Uncloud TLS

Wed, 20 May 2026 10:11:23 +0200

Caddy in a distributed setup (i.e. running on more than one machine) is a nightmare together with Let’s Encrypt ACME challenges… What happens is that Caddy A starts a challenge, which ends up at Caddy B and this can ping-pong for days before any certificates are given - and then only to a single instance, so the other instance will try for even longer. Of course the more of these you have the worse the problem gets.

Uncloud Image Pruning

Thu, 14 May 2026 22:04:39 +0200

Currently uc doesn’t have a prune subcommand to prune images, and even if it had it would still be a manual operation. It would be better to automate this. This post shows how.

First create an image with this Dockerfile:

FROM --platform=$BUILDPLATFORM alpine:latest
RUN apk add curl

RUN <<EOF
echo \
'#/bin/bash

if ! [[ -S /var/run/docker.sock ]]; then
    echo "Docker socket /var/run/docker.sock not found" >&2
    exit 1
fi

while true; do
    curl -s -X POST --unix-socket /var/run/docker.sock http://localhost/images/prune
    sleep 24h
done
' > /prune
EOF
RUN chmod +x /prune

WORKDIR /
CMD "/prune"

Build and push that image to a registry and then use this Uncloud compose.yaml to deploy:

Uncloud Prometheus

Sun, 26 Apr 2026 18:22:52 +0200

Prometheus metrics are being worked on by me, and will hopefully soon be merged in Uncloud.

The uncloudd metrics endpoint is exposed into the cluster network, so this makes it possible for a Prometheus service running inside the cluster to scrape it. For this it just needs to be able to access the API to get all machines in the cluster. This can be done by mounting the uncloud socket inside the container, and using uc with structured output.

Uncloud L4

Sat, 25 Apr 2026 13:21:52 +0200

Natively Uncloud doesn’t do L4 forwarding. This is an open issue as we are figuring out how to actually do this. But it turns out you can already do this with the current code - well almost.

Buckle up, because there is quite some stuff you may need:

~~1. A patched Uncloud, see this PR for that.~~

A custom Caddy build, with the L4 plugin.
A DNS server, I use atomdns of course.

Caddy

With the following Dockerfile you create a custom caddy image that we will use.

Uncloud Private Registries

Tue, 14 Apr 2026 19:31:00 +0200

While pushing this PR, the answer was that this was already working! But kind of unintuitive, but smart as it re-uses whatever docker setup you got (i.e. credentials helpers).

So to get this working you:

% docker login registry.science.ru.nl
Username: bla
Password:
...

And then use uc deploy -f .../compose.yml just like you did before. Now it should pull from the private repo and things “just work”.

See this guide for canonical text on this.

Uncloud

Mon, 16 Mar 2026 14:00:40 +0100

I’m playing with Uncloud to see if it is something we can use at $WORK. Pretty impressed with it already, as I’ve written something worse pgo, that was (obviously) not good enough. Uncloud does: implement a subset of the Docker(-compose) API, uses gRPC and wireguard to create mini network; just brilliant. Haven’t seen prometheus metrics yet, although Caddy is used for the proxy, but more prometheus metrics would make sense. (Which you can then observe through a cluster hosted prometheus…?)