# VM Scale Sets Flex previews automatic OS image upgrades > Flexible orchestration scale sets can now roll instances to newer OS image versions with health-aware, staged replacement controls. Canonical URL: https://www.devobs.io/news/news-azure-vmss-flex-automatic-os-image-upgrades-preview/ By: Lena Fischer Published: 2026-09-06T11:58:54.638Z Updated: 2026-09-06T11:58:54.638Z Event date: 2026-06-01 Section: Infrastructure Automatic OS image upgrades entered public preview for Virtual Machine Scale Sets using Flexible orchestration. Microsoft's [Compute post](https://techcommunity.microsoft.com/blog/azurecompute/public-preview-automatic-os-image-upgrades-for-vmss-flex/4523067) says the feature can apply newer versions of an image while coordinating upgrades across scale-set instances, extending an established maintenance pattern to Flex. ## Image freshness becomes a platform loop Automatic upgrades reduce the gap between publishing a corrected base image and replacing long-lived instances. That is valuable for fleets whose instances otherwise survive many application releases. It also means an image publisher can initiate change outside the application's normal deployment moment. The platform can sequence instance work, but it cannot know whether a new kernel, package, driver, or startup script is safe for a specific service. Health probes and upgrade policy therefore become part of release correctness. ## Treat the image as a release artifact Pin applications and configuration separately from the OS image, and document which image versions are allowed. In a test scale set, publish a new image and observe batching, instance health, capacity, and rollback behavior. Confirm the probe measures useful service readiness rather than merely an open port. Protect enough healthy capacity for peak load during upgrades. Verify that ephemeral caches rebuild, attached disks remount, identities remain assigned, and zone distribution stays acceptable. Emit image version and instance creation time into inventory so operators can see partial rollout. Because this is preview, keep a manual replacement path and avoid the only production fleet as the first test. Decide how emergency image fixes interact with maintenance windows, paused rollouts, and application deployment freezes. Automation is safest when it exposes progress and failure instead of silently refreshing machines. ## Source references -