Currently, legacy local disks in Compute Node work as follows:

There is a Local Disks Controller from which the node allocates/deallocates disks. The deallocation operation is always instant - it returns the disk to the controller, which then begins to clean it asynchronously. The allocation operation is blocking: the controller first tries to provide clean disks, and if only non-wiped disks are available, it blocks allocation until they are wiped.

Due to this approach, in the vast majority of cases, instances with local disks are created and deleted instantly, and only in a small number of cases do we get prolonged creation times while waiting for disk wiping.

Local disks over NBS work differently:

Creation is always instant because only clean devices are selected. Deletion is always slow because the disk deletion process (from the deletion operation perspective) includes devices cleaning. At this point, by switching to local disks over NBS, we're degrading the user experience. We would like the Disk Registry to have the same creation/deletion logic for Local disks over NBS as the current Local Disks Controller in Compute Node