# KubeVirt v1.7.0

2025-11-27 · [Original release notes](https://github.com/kubevirt/kubevirt/releases/tag/v1.7.0) · [Full summary](https://ratatosk.io/en/releases/kubevirt/v1.7.0)

Security 5 (high 1) · Breaking changes 20 · Deprecations 1 · Important fixes 31 · Enhancements 33 · Dependency updates 7 · Other changes 22

## Security (5)

### [high] HostDisk ownership vulnerability fixed

- Affects: `HostDisk` `DiskOrCreate` setups with an existing disk file
- [CVE-2025-64324](https://www.cve.org/CVERecord?id=CVE-2025-64324), [GHSA-46xp-26xh-hpqh](https://github.com/kubevirt/kubevirt/security/advisories/GHSA-46xp-26xh-hpqh)

Previously, KubeVirt could change an existing host file’s ownership to `107:107`, letting a VM read and modify it. It now leaves existing HostDisk files’ ownership and permissions unchanged.

> - [PR #15037][jean-edouard] HostDisk: KubeVirt no longer performs chown/chmod to compensate for storage that doesn't support fsGroup

Security advisory:

> Arbitrary Host File Read and Write

https://github.com/kubevirt/kubevirt/security/advisories/GHSA-46xp-26xh-hpqh

### [medium] NodeRestriction blocks updates to other nodes

- Affects: node updates evaluated by KubeVirt's `NodeRestriction` validating admission policy (Inferred)
- [CVE-2025-64436](https://www.cve.org/CVERecord?id=CVE-2025-64436), [GHSA-7xgm-5prm-v5gc](https://github.com/kubevirt/kubevirt/security/advisories/GHSA-7xgm-5prm-v5gc)

The advisory says excessive `virt-handler` permissions could let an attacker who compromised that account make VMIs appear terminated and have them recreated on an attacker-controlled node.

> - [PR #16076][kubevirt-bot] NodeRestriction: Source of node update is now verified

Security advisory:

> Excessive Role Permissions Could Enable Unauthorized VMI Migrations Between Nodes

https://github.com/kubevirt/kubevirt/security/advisories/GHSA-7xgm-5prm-v5gc

### [medium] Aggregated API client certificate Common Name check enforced

- Affects: aggregated API requests authenticated with client certificates
- [CVE-2025-64432](https://www.cve.org/CVERecord?id=CVE-2025-64432), [GHSA-38jw-g2qx-4286](https://github.com/kubevirt/kubevirt/security/advisories/GHSA-38jw-g2qx-4286)

A trusted-CA certificate with a Common Name outside Kubernetes’s allowed list could be used to impersonate the aggregator and bypass RBAC; exploitation also requires a valid front-end proxy certificate and network access to `virt-api`. Now, certificates with an unlisted Common Name are rejected, while any authority-validated certificate is allowed if Kubernetes provides no Common Names.

> - [PR #15314][xpivarc] Common Names are now enforce for aggregated API

Security advisory:

> Authentication Bypass in Kubernetes Aggregation Layer

https://github.com/kubevirt/kubevirt/security/advisories/GHSA-38jw-g2qx-4286

### [medium] virt-handler host-file ownership flaw fixed

- Affects: Not stated in the release notes
- [CVE-2025-64437](https://www.cve.org/CVERecord?id=CVE-2025-64437), [GHSA-2r4r-5x78-mvqf](https://github.com/kubevirt/kubevirt/security/advisories/GHSA-2r4r-5x78-mvqf), [GHSA-9m94-w2vq-hcf9](https://github.com/advisories/GHSA-9m94-w2vq-hcf9)

When an attacker controls a `virt-launcher` pod’s filesystem, can access the host filesystem as `qemu` (UID 107), and can create unprivileged user namespaces, `virt-handler` could mistake `launcher-sock` for the pod’s root mount and change arbitrary host files’ ownership to UID 107. The attacker could then modify those files; the demonstrated exploit races VMI creation or update.

Security advisory:

> It is possible to trick the `virt-handler` component into changing the ownership of arbitrary files on the host node to the unprivileged user with UID `107` due to mishandling of symlinks when determining the root mount of a `virt-launcher` pod.
> In the current implementation, the `virt-handler` does not verify whether the `launcher-sock` is a symlink or a regular file. This oversight can be exploited, for example, to change the ownership of arbitrary files on the host node to the unprivileged user with UID `107` (the same user used by `virt-launcher`) thus, compromising the CIA (Confidentiality, Integrity and Availability) of data on the host.
> To successfully exploit this vulnerability, an attacker should be in control of the file system of the `virt-launcher` pod.
> It is assumed that an attacker has access to a `virt-launcher` pod's file system (for example, [obtained using another vulnerability](https://github.com/kubevirt/kubevirt/security/advisories/GHSA-9m94-w2vq-hcf9)) and also has access to the host file system with the privileges of the `qemu` user (`UID=107`). It is also assumed that they can create unprivileged user namespaces:
> Upon successful exploitation, `virt-launcher` connects to the attacker controlled socket, misinterprets the root mount and changes the permissions of the host's `/etc/passwd` file:
> The attacker controlling an unprivileged user can now update the contents of the file.
> This oversight can be exploited, for example, to change the ownership of arbitrary files on the host node to the unprivileged user with UID `107` (the same user used by `virt-launcher`) thus, compromising the CIA (Confidentiality, Integrity and Availability) of data on the host.

https://github.com/kubevirt/kubevirt/security/advisories/GHSA-2r4r-5x78-mvqf

Security advisory:

> Isolation Detection Flaw Allows Arbitrary File Permission Changes

https://github.com/kubevirt/kubevirt/security/advisories/GHSA-2r4r-5x78-mvqf

### [medium] PVC disk symlink file disclosure fixed

- Affects: Not stated in the release notes
- [CVE-2025-64433](https://www.cve.org/CVERecord?id=CVE-2025-64433), [GHSA-qw6q-3pgr-5cwq](https://github.com/kubevirt/kubevirt/security/advisories/GHSA-qw6q-3pgr-5cwq)

When someone has enough control of a PVC’s contents to create a symlink to a file in the `virt-launcher` pod filesystem, the guest VM could read arbitrary files from that filesystem and from the mounted PVC, exposing storage data across the container-to-VM boundary.

Security advisory:

> Mounting a user-controlled PVC disk within a VM allows an attacker to read any file present in the `virt-launcher` pod. This is due to erroneous handling of symlinks defined within a PVC.
> A vulnerability was discovered that allows a VM to read arbitrary files from the `virt-launcher` pod's file system. This issue stems from improper symlink handling when mounting PVC disks into a VM. Specifically, if a malicious user has full or partial control over the contents of a PVC, they can create a symbolic link that points to a file within the `virt-launcher` pod's file system. Since `libvirt` can treat regular files as block devices, any file on the pod's file system that is symlinked in this way can be mounted into the VM and subsequently read.
> Although a security mechanism exists where VMs are executed as an unprivileged user with UID `107` inside the `virt-launcher` container, limiting the scope of accessible resources, this restriction is bypassed due to a second vulnerability (TODO: put link here). The latter causes the ownership of any file intended for mounting to be changed to the unprivileged user with UID `107` prior to mounting. As a result, an attacker can gain access to and read arbitrary files located within the `virt-launcher` pod's file system or on a mounted PVC from within the guest VM.
> The two PVCs will be mounted as volumes in "filesystem" mode:
> This vulnerability breaches the container-to-VM isolation boundary, compromising the confidentiality of storage data.

https://github.com/kubevirt/kubevirt/security/advisories/GHSA-qw6q-3pgr-5cwq

Security advisory:

> Arbitrary Container File Read

https://github.com/kubevirt/kubevirt/security/advisories/GHSA-qw6q-3pgr-5cwq

## Breaking changes (20)

### Decentralized live migration limits now constrain target-pod creation

- Affects: decentralized live migrations subject to per-node or cluster migration limits (Inferred)

Before this fix, migration limits were not accurately applied to decentralized live migrations. Tests cover pending migrations when either limit is occupied and expect no target pod to be created.

> - [PR #16011][kubevirt-bot] BugFix: The migration limit was not accurately being used with decentralized live migrations

### `VirtualMachinePool` updates now default to proactive restarts

- Affects: `VirtualMachinePool` resources (Inferred)

Previously, pools had no update-strategy API and only basic in-place update behavior. An explicitly configured opportunistic strategy updates the VM while the VMI changes naturally on restart; unmanaged makes no automatic VM updates, and proactive selection uses Random by default when no selection policy is configured.

Evidence: “"description": "SelectionPolicy defines the priority in which VM instances are selected for proactive update Defaults to \"Random\" base policy when no SelectionPolicy is configured"” https://github.com/kubevirt/kubevirt/pull/15123

> - [PR #15123][Sreeja1725] VMpool: Add UpdateStrategy support with Proactive, Opportunistic modes and Selection policies

### VMI graceful shutdown sends one event per request

- Affects: VMIs undergoing graceful shutdown

Previously, retries emitted repeated `VMIGracefulShutdown` events; with a large `TerminationGracePeriodSeconds`, these could exceed the event recorder's burst limit of 25 and cause later events, including the eventual shutdown event, to be dropped. Now only the initial request emits that event; retries remain logged in virt-handler.

> - [PR #15759][lyarwood] Only a single `Signaled Graceful Shutdown` event is now sent to avoid spamming the event recorder during long graceful shutdown attempts

### VMpool cleanup finalizer added

- Affects: setups using VirtualMachinePools

When a pod is deleted manually or the VMpool replica count is scaled down while the pool is configured to preserve VM and PVC state, the VM could be deleted before reconciliation runs, putting PVC preservation at risk. The finalizer lets the controller perform cleanup first and helps preserve the PVC before VM deletion.

Evidence: “Example usecase - When a pod is deleted manually or vmpool replica count is scaled down, and the vmpool is configured to preserve the state of the VM and PVC, it might be possible that the vm might have deleted by the time vmpool reconcile is triggered. Adding a finalizer will help in preserving the pvc before deleting the VM.” https://github.com/kubevirt/kubevirt/pull/15406

> - [PR #15406][Sreeja1725] Add VMpool finalizer to ensure proper cleanup

### VM container memory metrics renamed as request margins

- Affects: Prometheus queries or rules using either VM container memory metric (Inferred)

Their descriptions now define the values as requested memory minus RSS or working set, which can be negative when usage exceeds the request; previously, they described current available memory.

> - [PR #15504][sradco] Update metric kubevirt_vm_container_free_memory_bytes_based_on_rss and kubevirt_vm_container_free_memory_bytes_based_on_working_set_bytes names to kubevirt_vm_container_memory_request_margin_based_on_rss_bytes and kubevirt_vm_container_memory_request_margin_based_on_working_set_bytes so they will be clearer

### `ppc64le` architecture configuration support removed

- Affects: installations configured for the `ppc64le` architecture

Although the project already did not support or maintain `ppc64le`, its configuration and handling code existed before. This release removes that handling while leaving the schema field marked as deprecated.

Evidence: “This PR **removes the `ppc64le` architecture configuration completely** and also removes associated code references and handling logic related to the `ppc64le` architecture.” https://github.com/kubevirt/kubevirt/pull/14976

> - [PR #14976][dasionov] remove ppc64le architecture configuration support

### Export server excludes `lost+found`

- Affects: data volumes with `lost+found` in their mounted contents, exported by the export server (Inferred)

When the export server lacked permission to handle files not created by KubeVirt or CDI, it could encounter permission issues with `lost+found`. Exports now omit that directory, and the export server excludes it from directory-permission checks.

Evidence: “In some environments the export server might lack permissions to handle files that aren't created by Kubevirt or CDI, such as the lost+found directory.” https://github.com/kubevirt/kubevirt/pull/15509

> - [PR #15509][alromeros] Bugfix: Exclude lost+found from export server

### virtctl SSH and SCP now use host clients

- Affects: virtctl SSH or SCP usage, including invocations with `--local-ssh` (Inferred)

Before this release, virtctl included native SSH and SCP clients and the deprecated `--local-ssh` flag.

> - [PR #15478][0xFelix] virtctl: The --local-ssh flag and native ssh and scp clients are removed from virtctl. From now on the local ssh and scp clients on a host are always wrapped by virtctl ssh and scp.

### Migration memory-transfer metric renamed

- Affects: monitoring and automation consuming the migration memory-transfer metric (Inferred)

It was previously exposed as `kubevirt_vmi_migration_disk_transfer_rate_bytes` despite reporting memory transfer; it is now exposed as `kubevirt_vmi_migration_memory_transfer_rate_bytes`.

Evidence: “Before this PR: The metric name is wrong and should be renamed to `kubevirt_vmi_migration_memory_transfer_rate_bytes`” https://github.com/kubevirt/kubevirt/pull/13500

> - [PR #13500][brandboat] Fix incorrect metric name kubevirt_vmi_migration_disk_transfer_rate_bytes to kubevirt_vmi_migration_memory_transfer_rate_bytes

### Descheduler `eviction-in-progress` annotation follows VMI eviction status

- Affects: setups using the descheduler's `eviction-in-progress` annotation for VMI evictions

When a migration attempt failed, the annotation could be removed even though eviction would be retried, falsely signaling completion. It now reflects whether the VMI is marked for eviction and, during an active migration, is added only to the source pod; a stuck, uncanceled migration can stall a descheduler with a concurrent-eviction limit until an eviction completes or is canceled.

Evidence: “Note: If migrations are stuck and evictions are never canceled (KubeVirt currently has no such mechanism), and the descheduler is configured with a limit on concurrent evictions, the descheduler may stall until some evictions complete or are manually canceled.” https://github.com/kubevirt/kubevirt/pull/15423

> - [PR #15423][tiraboschi] Derive eviction-in-progress annotation from VMI eviction status

### Legacy virtctl dot-based target parsing removed

- Affects: virtctl `portforward`, `ssh`, or `scp` targets using `type/name.namespace` syntax

Previously, the part after the last dot was treated as the namespace; now a dotted target is parsed as a resource name with no namespace unless a namespace is supplied using supported syntax. Using the old syntax could prompt host-key verification again and might require a manual `known_hosts` update.

> - [PR #15475][0xFelix] virtctl (portfoward|ssh|scp): Drop support for legacy dot syntax. In case the old dot syntax was used virtctl could ask for verification of the host key again. In some cases the known_hosts file might need to be updated manually.

### VMI foreground-delete finalizer now uses a domain-qualified name

- Affects: VirtualMachineInstances using the KubeVirt foreground-delete finalizer (Inferred)
- Removed in: a future release

The old non-domain-qualified finalizer could trigger a Kubernetes warning that finalizer names should be domain-qualified. The controller now uses `kubevirt.io/foregroundDeleteVirtualMachine` and removes either the old or new finalizer when the VMI is finished or deleted; automation matching the old name may stop matching without an error.

> - [PR #15096][lyarwood] The `foregroundDeleteVirtualMachine` has been deprecated and replaced with the domain-qualified `kubevirt.io/foregroundDeleteVirtualMachine`.

### virt-operator scheduling restricted to control-plane nodes

- Affects: KubeVirt installations with non-control-plane nodes available for virt-operator scheduling (Inferred)

Previously, virt-operator could be scheduled on non-control-plane nodes. Avoiding multiple replicas on one node remains a preference, so replicas may still share a control-plane node when there are more replicas than control-plane nodes.

Evidence: “Avoiding multiple replicas on the same node is still only a preference, so multiple replicas will end up on the same node if there are more replicas than control-plane nodes.” https://github.com/kubevirt/kubevirt/pull/15157

> - [PR #15157][jean-edouard] virt-operator won't schedule on worker nodes

### KubeVirt synchronization status switches to `synchronizationAddresses`

- Affects: clusters with `DecentralizedLiveMigration` enabled, especially those with a migration network or integrations reading KubeVirt CR synchronization-address status

With decentralized live migration enabled and a migration network configured, the operator previously could not advertise the migration-network IP and instead reported a pod-network IP in the singular status field. The status now includes addresses with the configured port; when several are available, the first is also reported in `synchronizationAddress`.

Evidence: “The synchronization addresses one can use to connect to the synchronization controller, includes the port, if multiple addresses are available, the first one is reported in the synchronizationAddress field.” https://github.com/kubevirt/kubevirt/pull/15022

> - [PR #15022][awels] The synchronization controller migration network IP address is advertised by the KubeVirt CR

### Corrected `virtctl expose` Service selector

- Affects: VMIs exposed with `virtctl expose`
- Action required: `virtctl expose` from `virtctl` v1.7 and above does not work with older KubeVirt versions.

Previously, selecting by the full VMI name could miss the launcher pod when a hostname was set, and the launcher label could be truncated for long VMI names. Services now select by `vmi.kubevirt.io/id`.

> - [PR #15783][orelmisan] Specify correct label selection when creating a service via virtctl expose. The expose command on virtctl v1.7 and above will not work with older KubeVirt versions.

### Support for deprecated `instancetype.kubevirt.io/v1alpha{1,2}` APIs removed

- Affects: instancetype or preference resources, manifests, or clients using `instancetype.kubevirt.io/v1alpha{1,2}` (Inferred)

These API versions and their CRDs had remained supported despite being deprecated since v1.0.0.

> - [PR #15400][lyarwood] The deprecated `instancetype.kubevirt.io/v1alpha{1,2}` API and CRDs have been removed

### Minimum for `memoryOvercommit` raised to 10

- Affects: deployments that set `memoryOvercommit` below 10
- Action required: `memoryOvercommit` raised to 10 or higher before upgrading

Evidence: “Important: deployments that set a memoryOvercommit value below 10 need to bump to 10+ before upgrading.” https://github.com/kubevirt/kubevirt/pull/15681

> - [PR #15681][jean-edouard] Memory overcommit is now recalculated on migration.

### VM unpause requests rejected during snapshots

- Affects: VMs with `vm.Status.SnapshotInProgress` set

Previously, a VM could be unpaused while its snapshot was in progress; now the unpause request returns HTTP 409 Conflict.

Evidence: “Make sure unpause will be rejected if `vm.Status.SnapshotInProgress` is not nil.” https://github.com/kubevirt/kubevirt/pull/15001

> - [PR #15001][noamasu] bugfix: Enable vmsnapshot for paused VMs

### Eviction protection for virt-launcher-owned hotplug pods during live migration

- Affects: VMIs with virt-launcher-owned hotplug pods that can live migrate

Previously, a hotplug pod could be terminated during an ongoing live migration, potentially causing inconsistencies or failures before migration completed. The eviction webhook now protects the pod from premature eviction during migration.

Evidence: “This PR extends the same protection to hotplug pods owned by the virt-launcher, ensuring they are not prematurely terminated during an ongoing live migration. This prevents potential inconsistencies or failures caused by the hotplug pod shutting down before the migration completes.” https://github.com/kubevirt/kubevirt/pull/15093

> - [PR #15093][Acedus] bugfix: volume hotplug pod is no longer evicted when associated VM can live migrate.

### `virt-operator` removes `*_SHASUM` image selection

- Affects: `virt-operator` setups that use `*_SHASUM` environment variables for image selection
- Action required: Image selection previously specified with `*_SHASUM` configured through the remaining `*_IMAGE` variables, using a tag, digest, or both.

Previously, `*_SHASUM` variables selected image checksums; they no longer select image versions. The remaining `*_IMAGE` variables select images by tag, digest, or both.

> - [PR #15061][lyarwood] Support for all `*_SHASUM` environment variables has been removed from the `virt-operator` component. Users should instead use the remaining `*_IMAGE` environment variables to request a specific image version using a tag, digest or both.

## Deprecations (1)

### `indications` field deprecated in favor of `sourceIndications`

- Affects: `VirtualMachineSnapshot` resources and clients reading snapshot status (Inferred)
- Removed in: not stated

For online snapshots, status includes the existing indication values in `indications` and corresponding `sourceIndications` entries with messages. Offline snapshots have neither field populated by this update.

> - [PR #16033][noamasu] VMSnapshot: add SourceIndications status field to list snapshot indications with descriptions for clearer meaning.

## Important fixes (31)

### Hotplug volume status updates skip attachment pods marked for deletion

- Affects: setups performing hotplug volume status updates (Inferred)

> - [PR #16199][kubevirt-bot] Don't use attachment pods marked for deletion for hotplug volume status updates.

### Running VM CD-ROM detachment now sets `RestartRequired`

- Affects: running VMs with CD-ROM disks

> - [PR #16129][kubevirt-bot] Add RestartRequired when detaching CD-ROMs from a running VM

### VM disk physical block-size detection corrected

- Affects: VM disks whose physical block size is detected from a block device (Inferred)

Previously, `BLKBSZGET` could report a value that did not reflect the actual physical block size. Detection now uses `BLKPBSZGET` for physical size and `BLKSSZGET` for logical size.

> - [PR #15698][Acedus] It is now possible to configure discard_granularity for VM disks.

### CA bundle certificate filtering fixed in `MergeCABundle`

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

Before, `MergeCABundle` truncated bundles to 50 certificates before removing expired ones, so valid new certificates could be dropped while expired ones remained. It now removes expired certificates first and limits the remaining valid certificates to 50.

> - [PR #15706][ksimon1] fix: prioritize expired cert removal over 50-cert limit in MergeCABundle

### Expanded Object Graph coverage for `NetworkAttachmentDefinitions` and `ServiceAccounts`

- Affects: Object Graph setups with `NetworkAttachmentDefinitions` or `ServiceAccounts` as dependencies (Inferred)

`ServiceAccounts` were unintentionally omitted from the Object Graph, and `NetworkAttachmentDefinitions` were not included.

> - [PR #15615][alromeros] Object Graph: Include NADs and ServiceAccounts

### virt-operator bases virt-api replicas on schedulable nodes

- Affects: clusters where not all nodes are labelled `kubevirt.io/schedulable=true` (Inferred)

In larger environments where KubeVirt is not deployed on every node, the previous calculation could spawn too many replicas.

> - [PR #15690][lyarwood] Replicas of `virt-api` are now scaled depending on the number of nodes within the environment with the `kubevirt.io/schedulable=true` label.

### VM ownership for restored naked PVCs fixed

- Affects: VMSnapshot restores containing naked PVCs with restore policy `VM`

Previously, restoring a regular PVC not owned by a DataVolume could leave it without the VM owner reference, so it remained orphaned after VM deletion. With restore policy `VM`, the restored PVC now gets VM ownership and is removed with the VM.

> - [PR #15692][awels] BugFix: Restoring naked PVCs from a VMSnapshot are now properly owned by the VM if the restore policy is set to VM

### Restored volume ownership assigned to the VM

- Affects: VM restores

When the VM did not exist before the restore and was created after its volumes, the volumes previously ended up without VM ownership. In that sequence, restored volumes now receive ownership by the VM.

> - [PR #15344][SkalaNetworks] Added VolumeOwnershipPolicy to decide how volumes are owned once they are restored.

### Descheduler opt-out annotation propagation corrected

- Affects: VM templates or VMIs using `descheduler.beta.kubernetes.io/evict` or other descheduler opt-out annotations propagated to virt-launcher pods

KubeVirt previously propagated `descheduler.beta.kubernetes.io/evict`, which the Descheduler ignored. It now propagates `descheduler.alpha.kubernetes.io/prefer-no-eviction` instead, while continuing to sync `descheduler.alpha.kubernetes.io/evict`.

> - [PR #15365][tiraboschi] Aligning descheduler opt-out annotation name

### Installation with `KubeVirt.spec.imagetag` works again

- Affects: installations configured with `KubeVirt.spec.imagetag`

Before this release, installations using `KubeVirt.spec.imagetag` were not working.

> - [PR #16198][kubevirt-bot] Bug fix: KubeVirt.spec.imagetag installation is working again

### KSM enabled within three minutes under node pressure

- Affects: Nodes eligible for KSM handling that experience node pressure (Inferred)

The KSM handler checks and handles eligible nodes on a three-minute ticker.

> - [PR #16150][kubevirt-bot] fix: KSM is enabled in case of node pressure within 3 minutes

### Decentralized live migration failure propagation fixed

- Affects: setups performing decentralized live migrations

Failure state was not properly propagated during decentralized live migrations. When the source VMI fails, its migration state is marked failed and completed; when a scheduled target VMI has no pod, its migration state is also marked failed and completed, and the VMI is set to `WaitingForSync`.

> - [PR #16089][kubevirt-bot] A decentralized live migration failure is properly propagates between source and target

### virt-handler migration panic fixed for older controller configurations

- Affects: upgrades where virt-handler is updated before the migration controller

A migration using the older controller’s configuration could raise a nil-pointer exception when `AllowWorkloadDisruption` was omitted. virt-handler now uses `AllowPostCopy` when `AllowWorkloadDisruption` is nil, allowing the migration to proceed without that panic.

> - [PR #15008][fossedihelm] Fix possible nil pointer caused by migration during kv upgrade

### Migration controller fix for active migrations under heavy load

- Affects: deployments handling thousands of migrations

At a scale of thousands of migrations, active migrations could fail. This release changes the migration controller so that this load should no longer cause those failures.

> - [PR #15867][xpivarc] Bug fix: Thousands of migrations should not cause failures of active migrations

### Hotplug volume `RestartRequired` handling fixed

- Affects: setups using hotplug volumes

The regression surfaced after LiveUpdate became the default rollout strategy: `DeepEqual()` comparisons could fail when ordering changed. `RestartRequired` is now handled correctly in those cases.

> - [PR #15788][mhenriks] Fix RestartRequired handling for hotplug volumes

### virt-launcher process reaping fixed

- Affects: virt-launcher processes monitored by virt-launcher-monitor (Inferred)

When multiple signals arrive while a signal is blocked, only one may be delivered; the monitor's one `Wait4` per signal could miss a terminated process, leaving virt-launcher running indefinitely. This was observed after successful migration with both source and target Pods still running; the monitor now tries to reap as many processes as possible for each signal, addressing the missed cleanup.

> - [PR #15676][xpivarc] Bug fix, virt-launcher is properly reaped

### KubeVirt Istio virtual-interface annotation compatibility fixed

- Affects: KubeVirt pods using Istio virtual-interface annotations

The old `traffic.sidecar.istio.io/kubevirtInterfaces` annotation was broken with Istio v1.25 and later. KubeVirt now adds both it and `istio.io/reroute-virtual-interfaces`, so the new annotation provides the working behavior on those versions while the old one remains for v1.24.4 compatibility.

> - [PR #15661][nirdothan] Support Istio versions 1.25 and above.

### Decentralized migration cancellation and cleanup fixed

- Affects: decentralized live migrations

Deleting either migration resource during a migration did not properly cancel it and could leave migration resources that required manually removing finalizers. Now cancellation propagates, the source returns to running, and both migration resources are cleaned up; the target VM is not cleaned up by this change.

> - [PR #15605][awels] BugFix: Able to cancel in flight decentralized live migrations properly

### Default VirtAPI client rate limits raised

- Affects: KubeVirt setups using the default rate limits for individual webhook clients

Before this release, the `DefaultVirtAPI` client defaults were QPS 5 and burst 10, lower than the authorizer-client defaults; during mass eviction events, authorized requests could saturate the webhook client and go unserved. The defaults are now QPS 200 and burst 400.

> - [PR #15422][lyarwood] The `DefaultVirtWebhookClient{QPS,Burst}` values are aligned with `DefaultVirtWebhookClient{QPS,Burst}` to help avoid saturating the webhook client with requests it is unable to serve during mass eviction events

### Decentralized L3 live migration now passes the IP directly

- Affects: decentralized live migration on L3 networks

Previously, KubeVirt looked up a DNS name for the IP, and the resulting FQDN caused issues on the other side when its DNS record was unavailable there.

> - [PR #15630][awels] Allow decentralized live migration on L3 networks

### Migration controller prioritizes active migrations over pending work

- Affects: KubeVirt installations running the migration controller (Inferred)

When informer enqueues raised migration priority to 0, active migrations could be starved. The controller now keeps pending work at -100 and gives active migration work a priority above 0.

> - [PR #15513][jean-edouard] Fixed priority escalation bug in migration controller

### REST lifecycle handlers close launcher clients

- Affects: setups using the virt-handler REST lifecycle handlers (Inferred)

The handlers created `cmdclient.LauncherClient` clients without closing them, causing a leak.

> - [PR #15557][fossedihelm] Fix: grpc client in handler rest requests are properly closed

### VMI shutdown grace-period handling fixed

- Affects: VMIs configured with `spec.TerminationGracePeriodSeconds`

For unresponsive VMs that do not respond to ACPI signals, a force-stop grace-period update such as `virtctl stop vm --force --grace-period=0` could be ignored because virt-handler's cached grace period was stale, making shutdown appear to hang. Shutdown now honors the updated VMI grace period, including for unresponsive VMs.

> - [PR #15170][dasionov] bugfix: ensure grace period metadata cache is synced in virt-launcher

### Repeated `VMSnapshotContent` updates with unchanged errors fixed

- Affects: KubeVirt VM snapshot handling

After a `volumesnapshot` was deleted, `VMSnapshotContent` could be repeatedly updated with the same error and a new timestamp, contributing to ETCD bloat and possible performance degradation. It is no longer updated when the error message is unchanged, but still updates when the message changes.

> - [PR #15397][ShellyKa13] bugfix: prevent VMSnapshotContent repeated update with the same error message

### Snapshot restore failures after live volume updates fixed

- Affects: VM snapshots after live volume updates that change `DataVolumeTemplates`

After a live volume update such as storage volume migration, a VM revision could retain stale `DataVolumeTemplates`, causing restores from snapshots based on it to fail. Snapshot source construction now uses the VM's current `DataVolumeTemplates` alongside the overridden volumes and disks, allowing those restores to work.

> - [PR #15305][Acedus] bugfix: snapshot and restore now works correctly for VMs after a storage volume migration

### Dotted-secret `VMExport` export failures fixed

- Affects: VM exports

Exports could fail when the `VMExport` secret name contained dots.

> - [PR #15182][akalenyu] BugFix: export fails when VMExport has dots in secret

### virt-handler skips `ParallelMigrationThreads` for postcopy migrations

- Affects: postcopy live migrations

During upgrades when virt-handler and virt-launcher ran different versions, postcopy migrations could fail with `Postcopy is not yet compatible with multifd`.

> - [PR #15016][fossedihelm] Fix postcopy multifd compatibility during upgrade

### `VMExport` Service names now replace dots with hyphens

- Affects: `VMExport` resources with dots in their names

Previously, these exports could fail indefinitely when a valid Service name could not be crafted. Dots in the name are now replaced with hyphens to produce a Service name that conforms to DNS1035 label rules.

> - [PR #15099][akalenyu] BugFix: export fails when VMExport has dots in name

### StorageProfile label added to image-upload PVCs

- Affects: PVCs created with `virtctl image-upload`

With CDI's `WebhookPvcRendering` enabled, CDI can apply the associated `StorageProfile`, including a configured minimum supported PVC size for provisioners that do not support smaller sizes.

> - [PR #15067][alromeros] Bugfix: Label upload PVCs to support CDI WebhookPvcRendering

### Synchronization controller connection cleanup fixed for completed migrations

- Affects: clusters using decentralized live migration (Inferred)

When the synchronization controller handled completed migrations, it could try to establish connections, causing excess traffic and other issues. It now cleans up those connections instead.

> - [PR #14882][awels] Decentralized live migration is available to allow migration across namespaces and clusters

### TCP connection retention for `passt` VM migration

- Affects: VMs using the `passt` network binding

Previously, TCP connections for these VMs disconnected during migration. With `PasstIPStackMigrationEnabled` enabled and the target virt-launcher pod assigned the same IP as the source, connections are retained; migration still proceeds, but connections disconnect if the repair process errors or the IP is not retained.

> - [PR #14875][nirdothan] Support seamless TCP migration with passt (alpha)

## Enhancements (33)

- Migration support for known host CPU models after libvirt upgrades
- VMI evacuation cancellation command and subresource
- `pool.kubevirt.io/v1beta1` VirtualMachinePool API introduced
- VMPool auto-healing for failing VMs (To use: Configure `spec.autohealing` on the VMPool; auto-healing is opt-in.)
- VMPool gains unmanaged and opportunistic scale-in strategies
- Configurable VMPool scale-in state preservation
- Live migration for changed ImageVolume container-disk images
- Generalized priority categories for migrations (To use: `MigrationPriorityQueue` feature gate)
- Alpha Intel TDX VM support (To use: `WorkloadEncryptionTDX` feature gate)
- Experimental AMD SEV-SNP guest support (To use: `WorkloadEncryptionSEV` feature gate)
- VMI eviction-requested condition added
- Custom VM-template annotations and labels sync to virt-launcher pods (To use: Configure `virt-controller` with `additional-launcher-annotations-sync` and/or `additional-launcher-labels-sync` keys; both flags default to empty lists.)
- Instance type API adds `ioThreads` support
- Required and preferred architecture values in VirtualMachine preferences
- Label-selector-based Changed Block Tracking for VMs (To use: A matching namespace or VM label selector in `changedBlockTrackingLabelSelectors`, and a VM restart for CBT changes to take effect.)
- KubeVirt screenshots without VNC
- libdv builder option for DataVolume Upload sources
- VM labels exposed through the `kubevirt_vm_labels` gauge
- Configurable PVC ownership in `VirtualMachineRestore`
- Alerts for unhealthy and transitional VM states added
- VNC connections can preserve ongoing sessions (To use: The `--preserve-session` flag on `virtctl vnc` or the `preserveSession` VNC endpoint query parameter)
- virt-handler flag for controlling the Node Labeller controller (To use: The `enable-node-labeller` flag, which defaults to `true`; set it to `false` to disable the controller.)
- Guest CPU queue recording rules and alerts
- Guest-agent snapshots now handle paused VM domains
- Guest-agent CPU load metrics reported
- Bundled common-instancetypes updated to v1.4.0 with new preferences
- VM firmware serial validation removed
- Non-conflicting memory fields allowed alongside VMI instance types
- Architecture-specific `s390x` cluster configuration
- Synchronization address added to `VirtualMachineInstanceMigration` status (To use: The decentralized live migration feature gate must be enabled.)
- GPU and `HostDevices` support through Kubernetes DRA (To use: Enable the corresponding alpha feature gate: `GPUsWithDRAGate` for GPUs or `HostDevicesWithDRAGate` for HostDevices.)
- Decentralized live migration now works across namespaces and clusters (To use: `DecentralizedLiveMigration` feature gate)
- Additional CA trust for KubeVirt infrastructure components

## Dependency updates (7)

7 more, none mentioning security

- `- [PR #16026][fossedihelm] support v0.32.5 code generator`
- `- [PR #15718][Vicente-Cheng] Bump k8s v1.33`
- `- [PR #15936][kubevirt-bot] Updated common-instancetypes bundles to v1.5.1`
- `- [PR #15784][brianmcarey] Build KubeVirt with go v1.24.7`
- `- [PR #15531][Yu-Jack] bump prometheus operator to 0.80.1`
- `- [PR #15098][dominikholler] Update dependecy golang.org/x/oauth2 to v0.27.0`
- `- [PR #15100][dominikholler] Update dependecy golang.org/x/net to v0.38.0`

## Other changes (22)

- Dedicated migration mTLS certificate for virt-handler
- `KubevirtSeccompProfile` feature gate promoted to Beta
- `SecureExecution` feature gate promoted to Beta
- ImageVolume feature gate promoted to Beta
- `VideoConfig` promoted from Alpha to Beta
- v1.6.0 performance and scale benchmarks added to documentation and graphs
- `PanicDevices` feature gate promoted to Beta
- Configurable discard granularity for VM disk block I/O
- Higher default `virt-handler` rate limits
- Memory overcommit recalculated during VM migration
- Build tooling updated to Bazel v6.5.0 and `rules_oci`
- Interface-status normalization clears `IP` and `IPs` in hotplug tests
- VMPool unit tests use a fake client
- Migration PVC creation expectations keyed to migration resources
- Long-name volume migration labels use VMI UIDs
- virt-launcher memory delta added to monitoring metrics
- VM firmware serials now default to UUIDs
- Source VM/VMI removal unblocked after failed decentralized migrations
- Incorrect snapshot safety warning fixed for paused VMs
- virt-launcher agent poller skips guest-agent commands while VMs are paused
- Consistently named CPU and memory setters in `libvmi`
- `NodeRestriction` feature gate advances to Beta

---

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