- Sources: primary, discussion
- Summary: The post by Matthew Sanabria, dated 2026-08-13, names three provisioning paths shipped, a Rancher node driver merged from a customer pull request, an Omni infrastructure provider built with Sidero Labs, and Cluster API Provider Oxide. The Omni work turned up a Talos bug where the filesystem probe read only an ISO 9660 superblock from the NoCloud disk while Oxide uses FAT12, so Talos never read the user-data needed to join Omni, with the stated workaround padding user-data with comments until it crossed into an ISO 9660 superblock. LoadBalancer services run on floating IPs because Oxide has no native load balancer, and because Oxide translates the destination address before the guest sees it,
kubectl get service shows two addresses of which only one is reachable, while the CSI plugin is not shipped at all because Oxide requires an instance to be stopped before attaching or detaching a disk and Kubernetes expects attachment to a running worker. - Why it matters: The post names the gaps that are still open, and disk hot-plug through the hypervisor and the API is the constraint holding back persistent volumes on this hardware.
- Follow-up: Track whether disk hot-plug lands and whether the CSI plugin ships after it.
send feedback on this story