I wonder how much he gained from entirely restoring the system versus simply reprovisioning (gasp, even manually reinstalling) and restoring needed files a will. I'm not sure there's a lot of value in snapshotting and restoring stuff in a lost ssd situation tthat's also available in mirrors across the world.
I've lost data before, but it felt terrible to lose my context and working memory. While I make sure the most important stuff is in git, there's a bunch of momentum and working memory in my bash history and system configuration. It's also nice to not have to think very hard about a patchwork of backup plans.
It's nice to get a fresh start every now and then, but not under duress. I was in the middle of a multi day project and was gonna lose time either way. It was real nice to boot back into a machine that felt like home.
I do a version of this that doesn't require restore at all, I have three separate physical systems that all share a VPN across the world, and wherever I am at any given time zfs snapshots are syncing across that VPN to those three physical systems depending on which one is the primary I'm using at any given point in time. They also use the VPN layer to check if they have peer status on a faster local network like the wifi or LAN and use that instead for the snapshot transfers if so.
If for any reason any one of these systems either is destroyed or is no longer master, picking up from where I left off is as simple as picking another system up and marking it "master". No restore process, no changes, nothing at all, and it picks up from exactly where I left off when I was working on the other system. Means I can just grab my EDC laptop and stuff it in a bag not knowing how long I'll be out or where I'll be going and also know that it will be completely up to date with my datasets, or I can grab my desktop replacement laptop and its enormous external disk if I am going to be on a different continent for an extended period of time and want full geographic dataset locality. At no point in time does any of the above require the manual running of any process or replication or anything like that.
Reprovision and restore would take a whole lot longer than this, wouldn't give the abilities that it provides, and the above is only possible because of zfs snapshot replication.
I also use a USB C external SSD that is a member of a ZFS mirror and a md raid group that is bootable, so even if my EDC laptop were to spontaneously combust, I could immediately get up and running on any similar laptop with roughly comparable hardware simply by putting that SSD in and booting from it, then adding the SSD on the laptop to the zfs mirror / md raid group.
The biggest time save is in time spent recovering. It's so much faster to restore the entire system than to reinstall the OS, reconfigure the bootloader, resetup disk encryption, reconfigure user accounts, reinstall all software, manually reload configs, etc.
Or put it more directly, full disk backups are a great way to get RTO down.
Agreed. It was a large time investment that happened to pay off. If I ever have to do it again I'll be much faster. I hope that with wider ZFS adoption some of the routine tasks will be automated better in the future. I see no reason why in a couple years this couldn't be a mature fire and forget user experience.
