LVM Root Snapshot Rollback on CentOS: Where the Recovery Boundary Ends
0 reputation · 23 Jun 2021, 12:32 UTC
Goal and constraint
The intended design uses a pre-upgrade LVM snapshot of the root LV as the rollback path for a failed dnf upgrade on CentOS, with / on LVM and adequate free space in the VG or thin pool. The unresolved point is where that recovery boundary ends.
What a root snapshot does not cover
A root LV snapshot captures only that volume. /boot is usually a separate non-LVM partition, the EFI system partition and firmware boot entries are never included, and changes made after the snapshot — configuration management runs, databases, application state — survive the rollback. A kernel and initramfs installed during the failed upgrade can remain in /boot newer than the restored userspace.
Thick and thin snapshots also diverge: merging a thick snapshot into its origin differs from promoting a thin snapshot, and dnf history undo reverses package transactions rather than filesystem state. Release context matters too, since CentOS Linux 7 and 8 are past end of life and Stream's rolling model changes what an upgrade pulls in.
Open questions
- After a root snapshot is merged or promoted, is the newer kernel left in /boot expected to run against the older userspace, or should the default boot entry be returned to the pre-upgrade kernel first?
- For a thin-pool root LV, is activating the snapshot as the new origin preferable to an in-place merge when downtime must stay bounded?
