How to make Talos ignore OpenStack metadata?

[biased-badger]

Context:
I'm running multiple Talos clusters on our internal OpenStack. We don't use DHCP, only static predefined IPs and we don't have OpenStack metadata service network available.
We have a Terraform configuration that can deploy cluster in a fully automated mode and that is a crucial part of our incident mitigation approach. Basically, if things were to fail, we would be able to recreate the cluster in a short time and since we use Flux, all the workload would be restored as well.
This TF configuration prepares individual config for each Talos node, specifically with individual network settings and this config is passed to each OpenStack instance through config drive.
Our worker nodes have two network connections. First one is a primary network that's doing basically everything. Second one is a data network, that connects the cluster to data storage on a single /24 subnet. As such it doesn't need a gateway or complicated routing.

Problem 1 (Talos v1.10.5)
It turns out when you use openstack image variant, Talos would pull metadata from OpenStack even if it has full machineconfig applied. Also due to what I understand to be an OpenStack quirk, its network can't be defined without the gateway. What happens as a result is an additional default route with data network gateway is being added on Talos node from layer: platform and it messes routing on the node.

Here's how this interface is defined in the talos config:

  - interface: eth1
    addresses:
      - 10.0.10.230/24
    dhcp: false

And here's the RouteSpec that gets added:

node: 10.0.20.4
metadata:
    namespace: network
    type: RouteSpecs.net.talos.dev
    id: inet4/10.0.10.254//1024
    ...
spec:
    family: inet4
    dst: ""
    src: ""
    gateway: 10.0.10.254
    outLinkName: eth1
    table: main
    priority: 1024
    ...
    layer: platform

Problem 2 (Talos v1.12.4)
When I updated Terraform config to start VMs from Talos v1.12.4 openstack image, things got worse. Talos gets stuck in the booting phase due to platform config conflicting with my config:

[   80.935841] [talos] found config disk (config-drive) at /dev/sr0
[   80.936517] [talos] fetching meta config from: config-drive/openstack/latest/meta_data.json
[   80.938226] [talos] fetching network config from: config-drive/openstack/latest/network_data.json
[   80.939446] [talos] fetching machine config from: config-drive/openstack/latest/user_data
[   80.941966] [talos] volume status {"component": "controller-runtime", "controller": "block.VolumeManagerController", "volume": "platform/config-2/config", "phase": "ready -> closed", "location": "/dev/sr0"}
[   80.944603] [talos] controller failed {"component": "controller-runtime", "controller": "config.AcquireController", "error": "failed to validate config acquired via platform openstack: 1 error occurred:\n\t* static hostname is already set in v1alpha1 config\n\n"}

So it looks like in this version it is relying on platform metadata even more and is treating it as primary source over the inserted config.

Question
I'm not sure how I should proceed in this situation. I tried using metal ISO, but the problem with this approach is that it doesn't read machineconfig from OpenStack's config drive. And I can't apply this config until I manually configure static IPs on each node which defeats the point of automating this process. But if I use openstack image, OpenStack metadata seems to take precedence over my configuration which is also not what I need.

Is there a way to make openstack image variant ignore or deprioritize OpenStack metadata in favor of my config passed through config drive?
Or could I boot a metal image that would ignore OpenStack metadata, but still consume the config drive?

[smira]

Talos v1.10 is not supported anymore, so the question is not relevant.

Talos v1.12:

[ 80.944603] [talos] controller failed {"component": "controller-runtime", "controller": "config.AcquireController", "error": "failed to validate config acquired via platform openstack: 1 error occurred:\n\t* static hostname is already set in v1alpha1 config\n\n"}

You have incorrect machine config, this has nothing to do with OpenStack.

You're generating machine config for 1.12, which produces HostnameConfig document, but you're patching in .machine.hostname.

See https://github.com/siderolabs/talos/tree/main/pkg/machinery/config/generate/stdpatches for programmatic solution.

See also:

• https://docs.siderolabs.com/talos/v1.12/getting-started/what's-new-in-talos#network-configuration
• https://docs.siderolabs.com/talos/v1.12/networking/configuration/hostname

[biased-badger]

Thank you, that was indeed causing my issue!