After you commit changes to an image under the same name and tag, orka3 imagecache add reports it as already cached on some nodes and skips re-caching | 3.6.4 | Fixed in 4.0.0. On 3.6, remove and re-add the cached image, or deploy a VM from it once. See 3.6.4. |
A VM deployment fails with a generic error while orka3 imagecache remove runs for the same image on the same node | 3.6.2 and later | Don’t remove a cached image while VMs are deploying from it on that node. See 3.6.2. |
VMs deployed from an image on one node fail with Unable to deploy the VM as the underlying VM process has crashed, while the same image works on other nodes. The node ran out of disk space while importing the image, for example during two imports at once, and kept the incomplete image. | 3.6.4 and earlier | Fixed in 4.0.0. On 3.6, contact MacStadium support to remove the incomplete cached image, so the node downloads it again in full on next use. |
orka3 vm save and image pulls hang on some nodes, while VMs from already-cached images still deploy and the nodes can still show Ready. This can start after a node has trouble reaching the Kubernetes API, for example during maintenance, and the hang can start weeks later. | 3.6.4 | Fixed in 4.0.0. There’s no 3.6 fix yet, so Intel clusters, which stay on 3.6, can’t take it. On 3.6.4, stop retrying the save or pull, because each retry stays stuck too, and contact MacStadium support to recover the affected nodes. On Orka Anywhere, don’t restart only the virtual kubelet service on an Apple silicon node: it loses its permissions and can’t remount NFS. Recovery is a host reboot. |
New VMs can’t start on an Apple silicon node, and deployments to it time out or fail, while VMs already running on it keep running and the node still shows Ready. macOS stopped the node’s window server, which ended the node’s GUI session, and Orka needs that session to start VMs. Seen on macOS 15.7.1 and 26.6.2 hosts. | 3.6.4 and earlier | Fixed in 4.0.0, which restores the session before starting a VM. On MacStadium-hosted clusters, the upgrade to 4.0 also moves hosts on macOS 15.7.1 to 26.6.2. On 3.6, contact MacStadium support to restore the node. On Orka Anywhere, log in to the node over Screen Sharing as the user Orka runs as. |
An Apple silicon node keeps showing Ready while it can’t start VMs, so new VMs keep being scheduled to it and fail, for example with timed out after 10 minutes or Unable to deploy the VM as the underlying VM process has crashed. VMs already running on the node keep running. A node’s Ready status shows that the virtual kubelet is running, not that the node can start VMs. | All versions | If deployments keep failing while other nodes work, contact MacStadium support to check and restore the node. On Orka Anywhere, restarting the node’s host has cleared it. Restarting stops any VMs running on that node. Two known causes are fixed in 4.0.0: a lost GUI session and a hang in the virtual kubelet’s node status. |