Backups Aren't Simple (filipovski.net)

When I was a teenager, I was the reason for data loss for my dad, twice. Both times it was because I was re-partitioning a hard drive to install linux.

You would think that taught me a lesson about backups, instead it reminds me every now and again to be grateful for an awesome dad and aspire to handle situations with my kid similarly :)

The first was when the telephone pole outside our house was struck directly by lightning. Not only was it the loudest thing I have ever heard, the current surged through the telephone line, into the internal fax modem, and fries everything within its vicinity. I was 10. I did have backuos, but only only floppy and they didn't cover everything.

The second was storing data in OneDrive - a change to their terms surrounding "lifetime" unlikely noted storage, combined with a client that was unusably slow to download and a deadline for data retrieval meant that I lost most of my files.

The third was SD card failure in digital camera on holiday, the controller chip died catastrophically, leaving the card completely unrecognised. It was a brand new Sony 128GB card, manufactured by Toshiba, and it seemed to be a common issue. I now shoot to two cards simultaneously.

And the fourth time was ... Performing a backup. An errant script deleted the source content, but I'd also deleted the existing backup to free up space for the new backup. I've been weary of using rewritable media for some time now as a consequence, but I think backups themselves are high risk activities.

  We are not in the backup business.  We are in the restoration
  business.

    [1]: https://github.com/jimsalterjrs/sanoid
    [2]: https://du.nkel.dev/blog/2026-05-16_rootless_docker_virtiofs_proxmox/

All my hosts run the same CoreOS setup (https://github.com/ebrahim37/infra-template), where container volumes are placed in one central volumes/ folder and that is the only thing I have to backup.

I plan to implement it like this:

  vps1:
  - restic container with custom sh entrypoint that will backup volumes/ to homelab every 24 hours
  homelab:
  - backrest container, to back up volumes/, do prune/check, replicate repo to offsite
  - rest-server container, will store backups from vps1, homelab, offsite
  offsite:
  - restic container, backs up volumes/ to homelab every 24 hours
  - rest-server container, store copy of backups from homelab

The deduplication is nice, you can have a snapshot for each week of the past year without crazy storage cost

I never though of daylight savings times causing such an issue. Learned something today.

We finally got a nice setup with CloudNativePG + Barman. This allows for point-in-time restores, but there were a lot of lessons to learn along the way.

- The various types of (database) backups (logical, binary, onsite, offsite, snapshots, write-ahead log...) in combination with the various types of data (database, files, cluster configuration...)

- In our earlier approaches, we tried to preserve the old database volume if it was not corrupt, and use that in our restore. This caused so many complications, because you are fighting the recommended approach. So now, when we need to restore, we always restore from backups and the 'live volume' is dropped.

- For a restore, we just spin up a completely new Kubernetes cluster, instead of trying to restore in-cluster. This is a lot easier.

- Many object stores allow for retention periods, which you can put to good use to prevent malicious or accidental removal of backups. HOWEVER, not all of them are really 'locked'. In some services, you can still override the lock with a forced delete; in others, you can still remove the project holding the storage buckets, which will delete the buckets, and so on... so test those things, instead of just blindly depending on a 'retention period' claim.

- We now automatically run a scheduled restore with verifications on a weekly basis. This requirement does shape your environment, so keep that in mind! There is also the question of how you can reliably and automatically verify that the restore restored the latest data (of a live prod environment). Various solutions exist here, but most are not very elegant!

Honestly, this is only worth it if you are already handling sufficient volume. If you are just starting out, then the easier approach is to just go with a hosted database, which will handle backups and point-in-time restores for you.

This is an empirical claim, but is it grounded in reality?

99% of people don't backup, and that's probably the right choice because the risk is low, they can't meaningfully improve their restoration rate themselves and they don't care enough about their data to classify its loss as catastrophic.

When a person hears someone talk about the importance of backups, but also haven't heard friends/family suffer this date, they will rightfully ignore this warning.

Or do we all follow the best practices as it relates to backups, exercise, sleep, nutrition, accounting, house maintenance, ... Etc.

I do not. Backups aren't near the top of that list.

This stuff just isn't very important to most people and that's okay.

Obviously for production systems automation & monitoring is the way. I kind of like this setup on my own machine though.

https://www.jwz.org/doc/backups.html

(maybe copy/paste this into browser instead of following link, since referrer from hn apparently does something)

The best part: it is just files, not some proprietary archive. I have high confidence in this. I can restore with the most rudimentary tools.

This setup has worked for me well on linux, windows and now mac. I migrated to new computer&os by restoring the backup mostly.

https://relax-and-recover.org/

This is two-fold useful, first as it enables boot from backup media for total recovery, second in that it creates a backup.tar.gz that holds everything that was not explitictly ignored in the /etc/rear/local.conf file.

I'm running Oracle databases on these systems (and the local.conf is configured to ignore the datafile directories). I have standby databases (not Dataguard in that I don't have online redolog replication) that allow me to recover them to the primary in a disaster, with some loss of committed transactions.

I finally have rsync configured for some scratch temporary files.

I have dallied with btrfs snapshot replication for home directories (in a loopback mount). My vendor support (via a CSI number) has been on and off, so I don't use btrfs in areas that we need it most.

My replacements do not like the complexity (and I am retiring).

Two weeks later they had a fire in one of their huge printing machines which caused quite a lot of direct damage due to heat and rendered a lot of equipment broken due to smoke/fumes. It was also, as it turns out, a week before the first COVID 19 lockdown in the UK.

Their backups landed on a XFS file system with reflinks enabled. I cloned their repo and fired up NFS with their VMs running on our gear. It took me another hour to spin up another OpenVPN server (CA etc) for them to use and sort out a few other details (coffee doesn't brew itself).

They ran via VPN out of our data centre for most of the pandemic.

That was an utter triumph but I also have some rather less triumphant stories about backups and lack thereof. Let's skip over those 8)

GNU tar has its own incremental index via `--listed-incremental=FILE`. Unlike Borg or Restic, which have their own more complicated repository formats, this leaves me with just one additional file (the `.snar`) alongside a dumb, portable full-disk tar archive.

The nice part is that, unlike repositories that require both read/write access patterns, tar can compute deltas using only the small `.snar` file, while the main `full-disk.tar` can remain buried in write-only Glacier storage. This makes it a much better fit for Glacier's write-once model and 180-day minimum retention.

My current plan is to upload a full-disk tar to an external HDD + Glacier on a weekly cadence, and the `.snar`-based deltas daily. That gives me a pretty simple cloud backup solution for a few bucks a month (after burning through $100+ of free credits).

[1]: https://github.com/rsnapshot/rsnapshot/blob/master/rsnapshot...

The issue is companies love to fire QA people who should be the ones testing these.

Microsoft says we don’t need no SDETs, our products can ship full of issues.

What are the hostages, I mean customers going to do about it?

I have no sympathy for any major corp who suffers data loss. Now individuals who lost grandmas wedding photos, that sucks.

But my recovery strategy for that is simple. After scanning I emailed it out to one cousin who emailed them out to a few more. Plus I have a cloud backup ( I have no idea where the the original files are).

Data is meant to be shared after all

- do several backups (not using the same hardware brand or host if you do it online)

- check your backup (if something was wrong, you have another backup from the first rule)

- do not store your backups at the same place (if your place get flooded or burn, it will be pretty handy)

Using old standardized open source tools means it survives system updates without having to fix anything for years. I got my laptop stolen/lost 3 times over the past 15 years, and I have always been able to restore everything the next day on a new laptop, in the time it takes to transfer the files over the network.

I haven't given much thought, I'm sure there is a realistic scenario where this strategy would fails but I haven't found it yet.

Then switched to Restic - so much better - highly recommend this.