# KubeVirt v1.6.3

2025-11-11 · [Original release notes](https://github.com/kubevirt/kubevirt/releases/tag/v1.6.3) · [Full summary](https://ratatosk.io/en/releases/kubevirt/v1.6.3)

Breaking changes 2 · Important fixes 6 · Enhancements 1

## Breaking changes (2)

### Migration-time `memoryOvercommit` recalculation reverted

- Affects: deployments with `memoryOvercommit` below 10
- Action required: Deployments with `memoryOvercommit` below 10 must raise it to 10 or higher before upgrading.

> - [PR #15894][jean-edouard] REVERT: Memory overcommit is now recalculated on migration.

### `virt-api` replica scaling now counts schedulable nodes

- Affects: KubeVirt environments where the number of nodes labeled `kubevirt.io/schedulable=true` differs from the total node count (Inferred)

Previously, the replica calculation counted nodes without filtering by this label, so the resulting `virt-api` replica count can differ when some nodes lack it.

> - [PR #15824][kubevirt-bot] Replicas of `virt-api` are now scaled depending on the number of nodes within the environment with the `kubevirt.io/schedulable=true` label.

## Important fixes (6)

### Incorrect `RestartRequired` conditions for hotpluggable DataVolumes fixed

- Affects: VirtualMachine resources with hotpluggable DataVolumes

Adding a hotpluggable DataVolume no longer sets `RestartRequired`; non-hotpluggable volume changes still set it.

> - [PR #15890][mhenriks] Fix RestartRequired handling for hotplug volumes

### Expired certificates no longer displace valid certificates in `MergeCABundle`

- Affects: CA bundles processed by `MergeCABundle` (Inferred)

When a bundle contained expired certificates within the 50-entry limit, truncating it before filtering could leave fewer valid certificates. Expired certificates are now filtered first, and the remaining valid certificates are capped at 50.

> - [PR #15842][kubevirt-bot] fix: prioritize expired cert removal over 50-cert limit in MergeCABundle

### Restored naked PVC ownership fixed

- Affects: naked PVCs in VMSnapshot restores using the VM restore policy

A naked PVC originally owned by the source VM could be restored without its VM owner reference. The restore now assigns that reference only to PVCs originally owned by the source VM, leaving PVCs owned by something else unchanged.

> - [PR #15805][awels] BugFix: Restoring naked PVCs from a VMSnapshot are now properly owned by the VM if the restore policy is set to VM

### Migration target-pod lookup optimized

- Affects: clusters handling thousands of migrations

With thousands of migrations, active migrations could fail. The optimized lookup prevents failures under this condition.

> - [PR #15903][kubevirt-bot] Bug fix: Thousands of migrations should not cause failures of active migrations

### Migration nil-pointer failure during upgrades fixed

- Affects: KubeVirt migrations

If `virt-handler` was updated before the migration controller, a migration could encounter a missing `AllowWorkloadDisruption` value and trigger a nil-pointer failure. When that value is nil, it now uses `AllowPostCopy` instead.

> - [PR #15944][kubevirt-bot] Fix possible nil pointer caused by migration during kv upgrade

### CX-series hotplug CPU over-allocation fixed

- Affects: CX-series instance types in the `common-instancetypes` bundles, which require CPU manager and node huge pages

When KubeVirt's default hotplug behavior provided more CPUs than a node had available, a CX-series VM could fail. The updated bundle sets `spec.cpu.maxSockets` to the guest CPU count, preventing extra hotplug CPUs from exceeding node capacity.

> - [PR #15925][kubevirt-bot] Updated common-instancetypes bundles to v1.4.1

## Enhancements (1)

- VNC session preservation option (To use: `virtctl vnc --preserve-session` flag or `preserveSession` query option)

---

Summaries are produced by AI, so they can be wrong, and they are provided without warranty.
