So I think what you’re hitting here is the improved detection of custom volumes that are being used by running instances which was added in LXD 4.8.
Each storage driver has a RunningQuotaResize property which indicates if it supports in-use resizing.
Of all the storage drivers only ceph and lvm have the property set to RunningQuotaResize=false. This hasn’t changed, but what has changed is the code path that more thoroughly detects instances using custom volumes.
This is because they are block device backed volumes and don’t themselves have their own filesystem. This means that some filesystem resize operations require the volume to be unmounted before being resized and if the volume was in use then this would impact the instances using them (or the resize would fail because it was in use).
However the RunningQuotaResize true/false property is rather coarse and does not fully reflect what is truly possible, for example most filesystems on block devices can be grown when online without impacting the existing mounts.
To address this we already have an active case to remove the coarse grained true/false online resize property and replace it with something more subtle that allows online growth when possible.
Ok, thank you. I will follow the issue.
Meanwhile I think there is not else to do than stop the instance, grow the disk and boot again the container, right?