The machine didn't come back up. That's the short version. I ran a routine upgrade on the home server on a Friday evening, the kind you do without thinking because it's been fine a hundred times before, rebooted it because a kernel had come down, and got a kernel panic about an initramfs it couldn't assemble. No emergency shell that was any use. Just a wall of text and a blinking cursor at the absolute worst time of the week to be debugging boot.
A couple of years ago that's a weekend gone. Boot a live USB, chroot in, work out which package broke, downgrade it, regenerate the initramfs, swear at it, try again. Instead it was about thirty seconds, because the upgrade had quietly taken a snapshot of the root filesystem before it touched a single file, and rolling back to that snapshot is a thing btrfs does well and fast.
the bit that did the work
I'm on btrfs with the snapshotting wired into the package manager via a hook, so every transaction gets a pre-snapshot and a post-snapshot automatically. I didn't set it up that week. I set it up months ago, forgot about it entirely, and it sat there doing nothing visible until the one night it mattered.
The boot menu had the snapshots in it, because the bootloader config gets regenerated to include them. So from the panic, I rebooted, picked the pre-upgrade snapshot from the menu, and the machine came straight up on the old root. Working system, on the same kernel it had booted happily that morning, in less time than it takes to find the live USB I can never find.
Booting a read-only snapshot just gets you running again. To make it stick I set that snapshot as the new default root and rebooted once more.
# confirm what I booted into
btrfs subvolume get-default /
# list the snapshots, find the pre-upgrade one
snapper list
# set it back as the working root, then reboot
snapper rollback 142
reboot
snapper rollback makes the chosen snapshot the new root subvolume and tucks the broken one aside, so I kept the broken state around to look at later rather than destroying the evidence. Which mattered, because the next morning, coffee in hand and no longer mildly panicking, I could actually diagnose it properly.
what actually broke
The culprit was a mismatch between a kernel update and a third-party module that hadn't rebuilt against the new kernel, so the initramfs was missing a driver it needed to find the root device. Entirely my own fault for having that module pinned the way I did. The point isn't the bug. The point is that the bug was a Saturday-morning curiosity instead of a Friday-night emergency, because the cost of being wrong had dropped to nearly zero.
That's the real value of cheap, automatic snapshots. Not that they prevent mistakes, because they don't. They change what a mistake costs. An upgrade stops being a thing you brace for and becomes a thing you can undo, and once undo is free you stop being precious about when you run upgrades, which means you actually run them, which means the system stays patched. The safety net makes you braver, and braver here is the correct direction.
I do still take real backups, to another machine entirely, because a snapshot on the same disk is not a backup and the disk could have died instead of the kernel. But for "I broke it myself and want it back the way it was five minutes ago", nothing has saved me more weekends than a snapshot I forgot I'd set up.