LoginSign Up
← Back to all sparks
K

KubeVirt

INFRA · APIS
Velocity5.0

Virtual machine management add-on that runs VMs alongside Kubernetes containers

kubevirt.io ↗

KubeVirt's 1.9 cycle is a bet on GPUs and cross-architecture VMs

kubernetesvirtualizationgpu-passthroughdracross-architecturelive-migration
◆Current state
Everything in the window belongs to one long v1.9.0 pre-release cycle — alpha.0 in May through rc.2 in late July — totaling roughly 1,650 changes from over 100 contributors. The release notes are cumulative, so each RC restates the prior list and appends what landed since. There is no stable v1.9.0 tag yet.
◆Where it's heading
The cycle's center of gravity is accelerator and device virtualization. NVIDIA Grace GPU passthrough arrives behind a feature gate with SMMUv3/IOMMUFD, ACPI Generic Initiator NUMA topology and PCI 64-bit hole sizing; GPUsWithDRA and HostDevicesWithDRA graduate to beta with E2E coverage; SR-IOV vGPU gains display support and a new metric correlates GPU UUIDs with VMIs. A second thread pushes the hardware envelope elsewhere: cross-architecture execution behind an alpha gate, workload SEV encryption to beta, and zstd compression for live-migration streams.
◆Prediction
A stable v1.9.0 should follow shortly given rc.2 landed in late July and the recent commits are bug fixes rather than new gates. The alpha gates introduced here — cross-architecture virtualization and Grace I/O virtualization — are the ones to watch for graduation in the 1.10 cycle.

◆Recent moves

  1. 4d ago

    ARM64 GPU passthrough fix and zstd live migration

    The final RC closes out the accelerator thread by exposing VFIO cdev devices to virt-launcher, which is what makes IOMMUFD-based GPU passthrough actually work on ARM64 with SMMUv3. It also adds zstd compression for live-migration data streams and rejects network interfaces with missing or duplicate bindings at admission instead of failing at runtime.

    View source ↗
  2. 16d ago

    GPU and host-device DRA graduate to beta

    GPUsWithDRA and HostDevicesWithDRA move to beta with E2E test coverage, putting Kubernetes-native dynamic resource allocation for VM accelerators on the path to default-on. Supporting work includes NFS failover no longer restarting VMs and network binding plugin sidecars inferring their registration name from the environment.

    View source ↗
  3. 1mo ago

    Grace GPU passthrough and cross-architecture VMs land

    ⚡ SPARK

    The release that defines the 1.9 cycle: two alpha feature gates open capability surfaces KubeVirt did not previously have, one for NVIDIA Grace accelerators and one for running guests on a foreign CPU architecture. Everything in the later RCs is refinement of what this tag introduced.

    View source ↗
  4. 2mo ago

    1.9.0 beta with virt-handler and vmexport fixes

    The beta tag ahead of the RC series, carrying a gRPC connection leak fix in virt-handler's GetLauncherClient and narrower vmexport manifest handling. Internal correctness work before the feature gates that define this cycle appeared.

    View source ↗
  5. 2mo ago

    DRA claim rendering fixed for GPUs and host devices

    The first tag of the cycle, and the earliest visible point in the DRA thread: virt-controller now preserves per-device claim and request tuples when rendering GPU and HostDevice resources, including shared claim names. A correctness prerequisite for the DRA beta graduation that lands two releases later.

    View source ↗