Cluster information:
Kubernetes version: 1.29.x and 1.30.x
Cloud being used: Bare Metal
Installation method: kubeadm
Host OS: Debian 12 Bookworm
CNI and version: Cilium 0.3.1 (1.15.6)
CRI and version: containerd GitHub - containerd/containerd: An open and reliable container runtime 1.6.20~ds1 1.6.20~ds1-1+b1
When using this to create a very standard Kubernetes Cluster and this to install Cilium in the simplest way possible on an empty and extremely standard Debian 12 system, CoreDNS is always stuck in ContainerCreating status and nothing works. All the network related pods go into crash loops, etc.
Essentially, pretty much everything complains about not being able to connect to Port 6443, i.e. the Kubernetes API Server. However, when checking its logs in various ways, there are no errors whatsoever. The only error seen, is that it also cannot connect to this port…
I have tried many different Cilium as well as kubeadm init options - like with and without setting Pod CIDR, setting different Pod CIDRs, enabling and disabling Kube Proxy, etc. - however nothing worked. What baffles me about this: the installation procedure is extremely standard. No specific modifications, no weird settings, nothing. Just a textbook Kubernetes and CNI installation, without any out of the ordinary configuration options whatsoever. Yet, nothing works.
Can anyone reproduce this issue on Debian 12?
This Debian 12 is freshly installed. There is nothing on it, except an SSH server and dependencies for Kubernetes.
Every time I tried a new Kubernetes initialisation, I completely re-applied the Debian 12 image. So, the operating system is clean and ready to go, every single try.
NO Firewall installed.
NO Firewall enabled.
NO Firewall!