So… if you unencrypted zfs send from an encrypted source dataset… and do a concurrent snapshot…, there is a chance that the source pool will show checksum errors?
Which might be fixed with a few scrubs and reboots.
Gee. That may explain some of those curious unexplained checksum errors we see.
(Bug, HW failure, Memory Error… those ones)
Maybe we need to start asking if they’re using ZFS Encryption in future.
Or maybe the ZFS encryption failure reports need to check if they’re using ECC?