Azure previews full caching for Ephemeral OS disks

Azure placed Ephemeral OS disk full caching in public preview. In this mode the complete operating-system disk resides on local VM storage rather than splitting cached and remote-backed portions. Microsoft’s announcement targets stateless, I/O-sensitive VMs and scale sets that can be recreated from images.

Local placement changes failure

Ephemeral OS disks exchange persistence for quick reimage and stateless replacement. Full caching removes the OS disk’s steady-state remote-storage dependency. Current documentation says the selected SKU needs local cache capacity greater than twice the OS disk size plus 1 GiB, and configuration uses enableFullCaching with a recent Compute API.

Anything written only to that disk disappears during destructive lifecycle operations such as reimage, redeploy, or service healing. Microsoft separately documents that OS disk data is preserved during Live Migration, so routine maintenance through live migration is not equivalent to replacement. Logs, generated credentials, queues, and application data still need deliberate external destinations.

Validate capacity and recovery

Check image size against the SKU’s cache allocation and confirm regional availability. Build from code, force a nonproduction reimage, and measure readiness from a clean instance. Verify diagnostics leave the machine before failure.

The documentation labels full caching preview, without an SLA, and advises against production use. Treat it as a design experiment: use immutable images, health probes, replacement automation, and an external data plane. A benchmark is insufficient if the application still assumes its root filesystem survives. Test scale-out and repair as well as steady-state I/O, because recoverability is the defining constraint of this storage model.