Hi Ole,
Thanks a lot for reproducing and verifying the behavior and also thanks for tip.
It's unfortunate that there is no sequential fillup option. It takes the least-free-space definition too strictly. In that case I don't see any real use-case scenario for lfs/eplfs create policies, it is basically equivalent to only using a single remote. For file systems on physical hard drives (mergerfs etc.), using a most-free-space (or similar) policy to spread data across all drives is probably a preferred option for improving I/O efficiency and drive's life-time. However, for cloud storages, I think sequential fillup will be a more logical or preferred choice in most cases.
So, I wonder is it possible to implement such an option in rclone-union? even if the policy definition deviated from mergerfs', because physical and cloud storages have different areas of concern.
Sorry, I have no experience with mergerfs and I do not have a Linux system with multiple HDDs to try this out. However, I searched online regarding this and found one person mentioned it here. So, it is likely possible but I am not totally sure.
As I was searching more on this issue, I found a software called Air Cluster, it is a GUI utility only focussed on making union storage of cloud remotes. It doesn't have so much customizability like rclone-union, but it has only two basic options, one of which is sequential/ordered fillup of storage, and you can also set the order of fillup. Similar to what I was looking for. So, sequential fillup is in demand as it seems, and not sure if this can serve as a motivation for rclone to implement such a policy.
Here is a screenshot of it
