Comment by microtonal | Hacker News Reader

Comment by microtonal | Hacker News Reader

We did this for a while in our house, but with Bittorrent/Resilio Sync on later Pi models, and it was a pretty miserable experience. I/O is terrible on most models (which is somewhat important for storage). IIRC there is no acceleration of crypto primitives, so AES encryption used by Resilio Sync and Syncthing is slow. Finally, we found that even with a proper power supply, the Pi does not really provide the stability you'd want from a storage node.

We ended up replacing the Pi(s) by an Intel NUC, which was just over 100 Euro in 2018 without memory and storage. For relatively low cost, you get SATA speeds (or NVMe if you are willing to spend a bit more) and an AES-NI capable CPU. Our current NUC has been humming along for 4 years.

Ps. Syncthing is not backup, even with a read-only node. You are just one Syncthing bug or sysadmin failure away from erasing your data. Having the files on btrts on a single SSD only makes it worse.

Just get some redundant block storage somewhere that supports object lock (to prevent accidental deletion or ransomware encrypting all your files) and do incremental backups with something like Arq or Restic. There are some good services where 100GB storage costs $1 monthly.

This isn't the only backup I have of the data, it's just another copy of the data. BTRFS snapshots should be enough protection against the ransomware use case as well, assuming I discover it fast enough. And if I don't, I still have offline backups.

With a cloud solution, how are you not one bug or sysadmin failure away from data-loss or leaks?

Depends on your cloud solution, but OneDrive has version history so you can just roll it back.

Btrfs has snapshots too. That doesn't rule out bugs in it.

Making something someone else's problem doesn't magically make the problem go away.

Are we really having a debate on whether cloud storage with proper redundancy, reliable hardware with ECC memory and dedicated sysadmins is as vulnerable to mishaps as a consumer-grade SSD taped to a Raspberry Pi?

I hope we can at least agree that it is 3-4 orders of magnitude more likely that an SSD with btrfs and no RAID attached to a Pi to lose data within a given time period than, say S3?

Unless OneDrive silently corrupts your files upon upload: https://github.com/OneDrive/onedrive-api-docs/issues/1577

You avoid backup software bugs by using object lock. Even if your backup software messes up, it cannot remove older backups. On the cloud-side, sure bad things can happen, but a data loss event through a sysadmin or hardware failure is far less likely to happen on e.g. Amazon S3 or Backblaze B2 than with a home Raspberry Pi with a consumer-grade SSD taped on top of it. And you are probably backing up to at least two different destinations (at different companies) if you care about your data.

> Even if your backup software messes up, it cannot remove older backups.

But what if that has a bug?

Multiple write-only DVDs are clearly the only option per your logic.

You also can't just add on backups indefinitely or your costs will also approach infinity given enough time. There has to be a mechanism for deleting things, be it on DVD or on object lock cloud hype.

Personally, I think a backup is just that: a reserve copy. It should be reliable, but so long as you test your backups regularly, you can be confident that there won't suddenly be a bug when the primary copy fails. I, too, like to have two independent backups instead of one (happened to me before that, due to a niche mechanical failure called little brother, an external backup drive failed very soon after the primary), but saying one shouldn't use normal sync software "because it's one bug/misclick away from erasure" is silly. There can always be bugs and misclicks. They even mentioned using a read-only mode. It's a matter of how certain you want to be, and most people don't have any (automated) backups in the first place. Syncthing or similar software wouldn't be (isn't) my choice of backup software either, but I wouldn't dismiss a simple solution that works fine for them.

You also can't just add on backups indefinitely or your costs will also approach infinity given enough time. There has to be a mechanism for deleting things, be it on DVD or on object lock cloud hype.

Object locks have a configurable expiration date.

But what if that has a bug?

Again, this is yet another typical HN discussion. We are now comparing a consumer grade SSD taped to a Raspberry PI without ECC memory to a theoretical bug that might be in S3's or B2's object lock implementation. They have stored petabytes of data and there are virtually no reports of data loss ever, nor has anyone bypassed object lock, even if it's a high-value target.

Hetzner's storage boxes provide native borgbackup support, I've found it very useful as an offsite backup store.

Can you share an example of what might be a good service.

Re ransomware, how do you prevent it from deleting remote backups?

Write-Once-Read-Many and retention times, for example.

There can still be bugs such that you can write a second time much of the time the write once is just a software driver that won't write a second time so if you can write new firmware of your own design you can erase. Even if the media is physically not erasable, you can still write all ones over the top of what is there, corrupting data which is the same result.

Of course finding the bugs in firmware and writing a custom replacement is not trivial. It is conceptually possible though.

The only safe answer is a device in a vault without power. Of course once you retrieve it from the vault you risk whatever erased your data in the first place returning to get this too.

Good luck, you get to decide how paranoid you want go be.

We are talking about object lock as supported by something like Amazon S3 or Backblaze B2. If you find a way to override them, you probably want to apply for a bug bounty, since that'd be a huge security issue.

It works really nicely, I can't even remove my own data with my own credentials. It's really a nice additional security layer.

> If you find a way to override them, you probably want to apply for a bug bounty, since that'd be a huge security issue.

Big security issue, and honest people will just apply for a bug bounty or otherwise responsible disclosure. Most honest people are not actively looking for such bugs though, so evil people are more likely to find them. An evil person is can make even more money from a successful ransom. (there are also honest people looking, but many wrote the code so they may be too close to the problem to see it)

E.g. Amazon S3 and Backblaze B2 support this.

Re ransomware, how do you prevent it from deleting remote backups?

It can't, locks on objects cannot be removed in compliance mode. They can only expire.

https://www.backblaze.com/b2/docs/file_lock.html

https://docs.aws.amazon.com/AmazonS3/latest/userguide/object...

Reply on news.ycombinator.com