Impact on Talos Kubernetes Cluster Stability When DHCP IP Addresses Change Due to Lease Expiration

[H-haonan]

Issue/Question
I'm running a 3-master Kubernetes cluster on Talos Linux where all node IP addresses are dynamically assigned via DHCP. I understand that Talos supports DHCP for IP allocation, but I'm concerned about what happens when the DHCP leases expire and the IP addresses change (e.g., due to lease timeout or DHCP server behavior).
Specifically:

What is the expected cluster state after IP addresses change? Will the cluster remain operational, or could it lead to issues like etcd peer disconnection, API server unavailability, or pod scheduling failures?
Is the cluster still stable in this scenario? For example, does Talos handle IP changes gracefully, or does it require manual intervention to recover?
If the IPs change, can I continue managing the cluster (e.g., via talosctl or kubectl) using the newly assigned IP addresses? Would I need to update any configurations, such as the kubeconfig or machine configs, to point to the new IPs?

From my understanding, Kubernetes components like etcd and the control plane rely on stable endpoints. If IPs change without static DHCP reservations, it might disrupt the cluster. I'd appreciate any insights, best practices, or documentation references on using DHCP in production Talos clusters.
Thanks for your help!

[steverfrancis]

Best practice is dont use DHCP, certainly for control plane nodes, as etcd communication will break.

[H-haonan]

The best option is definitely to use static addresses. The reason I had this idea is that my three-node cluster will have the need to change addresses, but directly modifying the addresses in a three-node K8s cluster is also a troublesome process. I saw that in Talos, there is a way to create a cluster using DHCP for address allocation, so my thought at the time was that if the DHCP lease expires, the addresses will definitely change. In that case, what would be the result? Can the cluster with the new nodes be managed using kubeconfig or talosconfig files with the new addresses?

[smira]

Talos itself will adapt to new addresses.

You might need to update your client-side configs (e.g. talosconfig) yourself to use new addresses.

The etcd cluster will break if more than one machine changes an address at a time, it needs to patch in new member addresses which requires quorum members to be up.

[H-haonan]

Alternatively, is there an efficient solution for fixing or updating static addresses in a three-master cluster created with Talos?

[smira]

You can always set addresses statically in the machine config, it doesn't require a reboot or anything, just make sure you put it right so that you don't lose connectivity.