PSA: VMs may be shut down when upgrading to Harvester v1.9.0 When upgrading a multi-node cluster from v1.8.x to v1.9.0, live migration may fail during the node upgrade phase. A VM whose migration fails is shut down, stays off after the upgrade, and must be started manually. Only VMs with a single vCPU and a common CPU model configured migrate successfully. Two known issues cause this: - #11615: VMs with cores * threads > 1 (in practice, any VM with 2 or more vCPUs) fail to migrate with guest CPU doesn’t match specification: extra features: ht. This is a libvirt bug on the new QEMU 10 stack. No workaround; the fix needs a new KubeVirt image and is planned for v1.9.1. - #11356: VMs using the default host-model CPU can fail to migrate between the old and new QEMU versions. Setting a common CPU model (for example, IvyBridge) across the cluster before upgrading avoids this. Recommendation: wait for v1.9.1 (tentatively planned for October). If you must upgrade now, set a common CPU model, gracefully shut down VMs with 2 or more vCPUs in a maintenance window, upgrade, then start them again.

Hey, is your hardware different across the nodes?

plug for having the option to prevent vm shutdown during upgrade github.com/harvester/harvester/issues/8212