is rebasing and copying a snapshot/roofts while a guest is paused acceptable? (inode corruption issue)

[JacobZwang]

I'm wondering if the following flow is valid or of this will lead to memory corruption:

1. start vm from memory snapshot and rootfs
2. use vm which changes memory and rootfs
3. pause vm
4. create diff snapshot (because it's faster than full snapshot)
5. rebase diff snapshot onto original snapshot
6. copy rootfs and snapshot to create a fork of the vm (using cp --reflink=always)
7. resume vm
8. start forked vm

For context, each rootfs is an ext4 file based on the example in the getting started guide. I'm storing the ext4 files in an xfs file system so they can be copy on write.

It seems that occasionally my memory snapshots are getting corrupted in some way and I'm getting inode errors. Restarting the VM without the memory snapshot seems to resolve the issue.

It's unclear what's causing this issue. I haven't been able to reliably reproduce this issue by using the above flow. I've never observed this issue happening instantly (although it could be happening instantly and I just haven't noticed). The only times I've noticed this issue is when returning to a vm (that was never stopped) the next day.

Alternatively, I'm wondering if this issue is being caused by my host xfs disk. I'm using Google Cloud's Hyperdisk Extreme. Is using a network attached block storage device like this a bad idea for memory snapshots, especially those copied using reflink?