Bidirectional auto file sync is a fundamentally broken pattern and I'm tired of pretending it's not. It's just complete chaos with wrong files constantly getting overridden on both ends.
I have no clue why people still use it and I'd cut my losses if I were you, either backup to the cloud or pull from it, not both at the same time like an absolute tictac.
This is an instance of someone familiar with complex file access patterns not understanding the normal use case for these services.
The people using these bidirectional sync services want last writer wins behavior. The mild and moderately technical people I work with all get it and work with it. They know how to use the UI to look for old versions if someone accidentally overwrites their file.
Your characterization as complete chaos with constant problems does not mesh with the reality of the countless low-tech teams I've seen use Dropbox type services since they were launched.
This would be half OK if it worked, but you can't trust it to. OneDrive, for instance, has an open bug for years now where it will randomly revert some of your files to a revision from several months earlier. You can detect and recover this from the history, but only if you know that it happened and where, which you usually won't because it happens silently. I only noticed because it happened to an append-only text file I use daily.
A specific implementation (OneDrive) doing something dumb doesn't invalidate the entire paradigm though. Things work just fine elsewhere (Dropbox, Google Drive, Nextcloud, and Seafile are all solutions I've had good experiences with).
Agreed, I’ve been using Dropbox for 15 years with minimal issues. The key is to ensure it’s running and syncing with the proper settings on both machines.
What can get things into a weird state is if both machines are editing the same file while only one of them is actively syncing. But for basic backup and sync, this is extremely rare.
Unlimited strings are a problem. People will use it as storage.
No, I'm not joking. We used to allow arbitrary paths in a cloud API I owned. Within about a month someone had figured out that the cost to store a single byte file was effectively zero, and they could encode arbitrary files into the paths of those things. It wasn't too long before there was a library to do it on Github. We had to put limits on it because otherwise people would store their data in the path, not the file.
I remember someone telling me that S3 used to be similarly abused - people were creating empty files and using S3 like a key-value store somehow, so AWS just jacked up the price of S3 head-object API call to push people back to DynamoDB or whatever.
Not sufficient, unfortunately. The strings for file paths are stored in wholly different infrastructure with wholly different optimizations. It probably lives in your database. You really don't want people just stuffing gigabytes into that, payment or no payment. Odds are you didn't plan your control plane around, "what if someone uses our strings as encoded data?"
In the fine print, only to be used against bad actors (w/guarantee that filenames under x chars would never be charged), or that too problematic? building good faith into policy + "hiding" info...
Reason - to not overcomplicate or give appearance of nickel-and-diming
No, just charge for the amount of storage they use on your server. Not the amount of data you think you’re storing. In non-special cases these will be the same number.
Would there be any engineering/management pushback on the customer side? "we have to write a tiny script", "this is non-standard" / "why are you the only ones who charge us for filenames?"
You expect the files to still be accessible using relative paths. What do you expect to happen if your cloud storage file path is 50 characters long and is mounted in a folder which is 4050 characters long when PATH_MAX is 4096?
The sync application itself can handle this using openat(2) or similar and should probably be using that regardless to avoid races.
Ah, I forgot that the maximum path length is usually limited by PATH_MAX, it's the path segment that's usually limited by the filesystem.
Point taken, although I still think it's better for cloud storage services to err on the side of compatibility, i.e. what's the lowest common denominator between Linux, macOS, Android, iOS from 10 years ago and Windows 7?
Oh yeah... I remember Windows behaving weirdly when I tried to copy some files with long names into a deeper directory tree. And it was just weird behaviour - no useful error message.
Windows in particular supports at the API level paths tens of thousands of characters long, much longer than Linux. The problem is applications need to explicitly support such paths using the long path syntax, otherwise they're limited to 255 characters.
Except the GNU stuff, which has as a design principle "no arbitrary limits". Meaning no limits at all, not "no sane limits":
Avoid arbitrary limits on the length or number of any data structure, including filenames, lines, files, and symbols, by allocating all data structures dynamically.
I assume they're relying on the OOM Killer and quotas to prevent DoSes all over the place.
You can build such a system yourself quite trivially by getting an FTP account, mounting it locally with curlftpfs, and then using SVN or CVS on the mounted filesystem. From Windows or Mac, this FTP account could be accessed through built-in software.
1 out of a thousand people might do that, the others will buy the product. That's why people use it, most people don't want to build everything themselves.
But as usual it forgets the "For a Linux user" part.
If we remove the whole linux section and just ask "why not map a folder in Explorer" it's a reasonable question, probably even more reasonable in 2026 than in 2007. The network got faster and more reliable, and the dropbox access got slower.
It’s kind of wild to read through these comments and realize hn is still riffing on the same ideas. Is it e2ee? Does it run on Linux? Who would pay for something you can slap together in a weekend with a few bash scripts?
Really highlights this community’s values, skills, and blind spots.
Also a bit of a bummer that the privacy and open source situations today are even worse in many ways.
The equivalent of this is advice from a friend of mine who likes different teas, "just learn to read hanzi like I did and then you can select the ones you like". Apparently there's one called "government tea" (in the original) which he expected to taste of old leather, musty paperwork, and stale cigarette smoke.
Obvious. Explorer even has support built in for transparent ‘native’ gui support. I’m not even sure why you felt the need to explain it in detail. Next you’ll be explaining how to walk. (/s, I loved it)
I think this is a case of people using bidirectional file sync wrong. The point is to make the most up to date version of a file available across multiple devices, not to act as a backup or for collaboration between multiple users.
It works perfectly fine as long as you keep how it works in mind, and probably most importantly don't have multiple users working directly on the same file at once.
I've been using these systems for over a decade at this point and never had a problem. And if I ever do have one, my real backup solution has me covered.
“Every file is only ever written to from a single client, and will be asynchronously made available to all other clients, and after some period of time has elapsed you can safely switch to always writing to the file from a different client”.
Yep. See e.g. steam cloud saves, which is literally just Dropbox for your video game save files. Bidi sync is a super common pattern if you look for it, I'm surprised at all the hate it's getting here.
> And if I ever do have one, my real backup solution has me covered.
What do you use and how do you test / reconcile to make sure it’s not missing files? I find OneDrive extremely hard to deal with because the backup systems don’t seem to be 100% reliable.
I think there are a lot of solutions these days that error on the side of claiming success.
I agree. I use syncthing for syncing phones and laptops. For data like photos, which aren't really updated. It works very nice. And for documents updated by one user, moving between devices is totally seamless.
That being said i understand how it works at a high level.
It works perfectly fine if you're user that know how it works. I use it with Syncthing and it works coz I know to not edit same file at the same time on 2 devices (my third and fourth device is always on internet so chances propagate reasonably fast even if the 2 devices aren't on on the same time)
Hello, Jim from Backblaze here. I wanted to offer some insight into what happened with backing up cloud-synced folders.
It is true that we recently updated how Backblaze Computer Backup handles cloud-synced folders. This decision was driven by a consistent set of technical issues we were seeing at scale, most of them driven by updates created by third-party sync tools, including unreliable backups and incomplete restores when backing up files managed by third-party sync providers.
To give a bit more context on the “why”: these cloud storage providers now rely heavily on OS-level frameworks to manage sync state. On Windows, for example, files are often represented as reparse points via the Cloud Files API. While they can appear local, they are still system-managed placeholders, which makes it difficult to reliably back them up as standard on-disk files.
Moreover, we built our product in a way to not backup reparse points for two reasons:
1. We wanted the backup client to be light on the system and only back up needed user-generated files.
2. We wanted the service to be unlimited, so following reparse points would lead to us backing up tons of data in the cloud
We’ve made targeted investments where we can, for example, adding support for iCloud Drive by working within Apple’s model and supporting Google Drive, but extending that same level of support to third-party providers like Dropbox or OneDrive is more complex and not included in the current version.
We are currently exploring building an add-on that either follows reparse points or backs up the tagged data in another way.
We also hear you clearly on the communication gap. Both the sync providers and Backblaze should have been more proactive in notifying customers about a change with this level of impact. Please don't hesitate to reach out to me or our support team directly if you have any questions. https://help.backblaze.com/hc/en-us/requests/new
I have no clue why people still use it and I'd cut my losses if I were you, either backup to the cloud or pull from it, not both at the same time like an absolute tictac.