The future of large files in Git is Git
tylercipriani.com
tylercipriani.com
You specify the files you want to store in your storage backend via .gitattributes, and use two separate commands to sync files. I have not touched this code in years but the general implementation should still work.
GitHub launched LFS not too long after I wrote this, so I kind of gave up on the idea thinking that no one would want to use it in lieu of GitHub's solution, but based on the comments I think there's a place for it.
It needs some love but the idea is solid. I wrote a little description on the wiki about the low-level implementation if you want to check it out. [1]
Also, all of the metadata is stored using git notes, so is completely portable and is frontend agnostic—doesn't lock you into anything (except, of course, the storage backend you use).
Earliest tagged git lfs: Jun 25, 2014
Check this out: https://pypi.org/project/git-bigstore/#history
I like this approach. If I could configure my repos to use something like S3, I would switch away from using LFS. S3 seems like a really good synergy for large blobs in a VCS. The intelligent tiering feature can move data into colder tiers of storage as history naturally accumulates and old things are forgotten. I wouldn't mind a historical checkout taking half a day (i.e., restored from a robotic tape library) if I am pulling in stuff from a decade ago.
I'd initially at spinning up an LFS backends, but this solves the main pain point, for now. Github was charging us an arm and a leg for pulling LFS files for CI, because each checkout is fresh, the caching model is non-ideal (max 10GB cache, impossible to share between branches), so we end up pulling a bunch of data that is unfortunately in LFS, every commit, possibly multiple times. Because of this they happily charge us for all that bandwidth, because they don't provide tools to make it easy to reduce bandwidth (let me pay for more cache size, or warm workers with an entire cache disc, or better cache control, or...).
...and if I want to enable this for developers it's relatively easy, just add a new git hook to do the same set of operations locally.
All our builds are on GHA definitions, there’s no way it’s worth it to swap us over to another build system, administer it, etc. Our team is small (two at the time, but hopefully doubling soon!), and there’s barely a dozen people in the whole engineering org. The next hit list item is to move from GH hosted builders to GCE workers to get a warmer docker cache (a bunch of our build time is spent pulling images that haven’t changed) - it will also save a chunk of change (GCE workers are 4x cheaper per minute and the caching will make for faster builds), but the opportunity cost for me tackling that is quite high.
And even if that did work, I’ve found it much more reliable to use the actual docker BuildX disk state than to try and get caching for complex multi stage builds working reliably. I have a case right now where there’s no combination of —cache-to/from flags that will give me a 100% cached rebuild starting from a fresh builder, using only remote cache. I should probably report it to the Docker team, but I don’t have a minimal repro right now and there’s a 10% chance it’s actually my fault.
[1] https://github.com/rwx-cloud/packages/blob/main/git/clone/bi...
Apparently, this is coming in Q3 according to their public roadmap: https://github.com/github/roadmap/issues/1029
I run a small git LFS server because of this and will be happy to switch away the second I can get git to natively support S3.
If you fancy it for your datacenter, big players (Fujitsu, Lenovo, Huawei, HPE) will happily sell you "object storage" systems which also support S3 at very high speeds.
Scality's open source S3 Server also can run in a container.
Edit: Particularly the hash algorithm and the change detection (also when this happens) makes a difference if you have 2 GB files and not only the 25MB file from the OP
There are a couple other projects that bridge S3 and LFS, though I had the most success with this setup.
LFS does break disconnected/offline/sneakernet operations which wasn't mentioned and is not awesome, but those are niche workflows. It sounds like that would also be broken with promisors.
The `git partial clone` examples are cool!
The description of Large Object Promisors makes it sound like they take the client-side complexity in LFS, move it server-side, and then increases the complexity? Instead of the client uploading to a git server and to a LFS server it uploads to a git server which in turn uploads to an object store, but the client will download directly from the object store? Obviously different tradeoffs there. I'm curious how often people will get bit by uploading to public git servers which upload to hidden promisor remotes.
I dunno if their solution is any better but it's fairly unarguable that LFS is bad.
1. It is a separate tool that has to be installed separately from git
2. It works by using git filters and git hooks, which need to be set up locally.
Something built in to git doesn't have those problems.
But GP's point was that there is an entire other category of errors with git-lfs that are eliminated with this more native approach. Git-lfs allows you to get into an inconsistent state e.g. when you interrupt a git action that just doesn't happen with native git.
The architecture does seem to still be in the general framing of "treat large files as special and host them differently." That is the crux of the problem in the first place.
I think it would shock no one to find that the official system also needs to be enabled and also falls back to a mode where it supports fetching and merging pointers without full file content.
I do hope all the UX problems will be fixed. I just don't see them going away naturally and we have to put our trust in the hope that the git maintainers will make enjoyable, seamless and safe commands.
Mostly I did not run into such use case but in general I don’t see any upsides trying to shove some big files together with code within repositories.
That is why I don’t understand why people „need to use GIT”.
You still can make something else like keeping versions and keeping track of those versions in many different ways.
You can store a reference in repo like a link or whatever.
Yes because Git currently is not good at tracking large file. That's not some fundamental property of Git; it can be improved.
Btw it isn't GIT.
Wanting to split up the project into multiple storage spaces is inherently hostile to managing the project. People want it together because it's important that it stays together as a basic function of managing a project of digital files. The need to track and maintain digital version numbers and linking them to release numbers and build plans is just a requirement.
That's what actual, real projects demand. Any projects that involve digital assets is going to involve binary, often large, data files. Any projects that involve large tables of pre-determined or historic data will involve large files that may be text or binary which contain data the project requires. They won't have everything encompassed by the project as a text file. It's weird when that's true for a project. It's a unique situation to the Linux kernel because it, somewhat uniquely, doesn't have graphics or large, predetermined data blocks. Well, not all projects that need to be managed by git share 100% of the attributes of the Linux kernel.
This idea that everything in a git project must be all small text file is incredibly bizarre. Are you making a video game? A website? A web application? A data driven API? Does it have geographic data? Does it required images? Video? Music or sound? Are you providing static documentation that must be included?
So the choices are:
1. Git is useful general purpose VCS for real world projects. 2. Git does not permit binary or large files.
Tracking versioning on large files is not some massively complex problem. Not needing to care about diffing and merging simplifies how those files are managed.
There are other tools for managing projects and better ways to version large files or binary assets.
Git is great at handling text changes and that’s it. It sucks with binary blobs.
I think it is a much bigger barrier than ssh and have seen it be one on short timeline projects where it's getting set up for the first time and they just end up paying github crazy per GB costs, or rat nests of tunnels vpn configurations for different repos to keep remote access with encryption with a whole lot more trouble than just an ssh path.
In other words, if you migrate a repo that has commits A->B->C, and C adds the large files, then commits A & B will gain a `.gitattributes` referring to the large files that do not exist in A & B.
This is because the migration function will carry its ~gitattributes structure backwards as it walks the history, for caching purposes, and not cross-reference it against the current commit.
https://github.com/git-lfs/git-lfs/blob/main/docs/man/git-lf...
Now, granted, usually people run migrate to only convert new local commits, so by nature of the ref include/exclude system it will not touch older commits. But in my case I was converting an entire repo into one using LFS. I hoped it would preserve those commits in a base branch that didn't contain large files, but my disappointment was said .gitattributes pollution.
> In all modes, by default git lfs migrate operates only on the currently checked-out branch, and only on files (of any size and type) added in commits which do not exist on any remote. Multiple options are available to override these defaults.
Were your remotes not configured correctly?
> But in my case I was converting an entire repo into one using LFS.
then check out the section in the manual "INCLUDE AND EXCLUDE REFERENCES"
But anyway, yes, LFS rewrites history if you want to apply it to history. I agree it's sub-par; it's disruptive and risks breaking links to specific githashes.
Yea, I had the same thought. And TBD on large object promisors.
Git annex is somewhat more decentralized as it can track the presence of large files across different remotes. And it can pull large files from filesystem repos such as USB drives. The downside is that it's much more complicated and difficult to use. Some code forges used to support it, but support has since been dropped.
Nowhere is this behavior explicitly stated.
I used to use Git LFS on GitHub to do my company’s study on GitHub statistics because we stored large compressed databases on users and repositories.
Is that true? I used git commercially in five companies, and I never used github commercially (except as a platform for projects we opensourced).
You already depend on github if you host your project there. But you're not locked in, because you can just close your github repo and migrate somewhere else. Do I miss something?
If you used LFS, you have to fork and rewrite your repository to update the .lfsconfig backend URLs to get back to a reasonable working state.
> High vendor lock-in – When GitHub wrote Git LFS, the other large file systems—Git Fat, Git Annex, and Git Media—were agnostic about the server-side. But GitHub locked users to their proprietary server implementation and charged folks to use it.
Is this a current issue?
I used Git LFS with a GitLab instance this week, seemed to work fine.
https://docs.gitlab.com/topics/git/lfs/
I also used Git LFS with my Gitea instance a week before that, it was fine too.
https://docs.gitea.com/administration/git-lfs-setup
At the same time it feels odd to hear mentions of LFS being deprecated in the future, while I’ve seldom seen anyone even use it - people just don’t seem to care and shove images and such into regular Git which puzzles me.
Though I did work hard to remove any vendor lock-in by working directly with Atlassian and Microsoft pretty early in the process. It was a great working relationship, with a lot of help from Atlassian in particular on the file locking API. LFS shipped open source with compatible support in 3 separate git hosts.
While git LFS is just a kludge for now, writing a filter argument during the clone operation is not the long-term solution either.
Git clone is the very first command most people will run when learning how to use git. Emphasized for effect: the very first command.
Will they remember to write the filter? Maybe, if the tutorial to the cool codebase they're trying to access mentions it. Maybe not. What happens if they don't? It may take a long time without any obvious indication. And if they do? The cloned repo might not be compilable/usable since the blobs are missing.
Say they do get it right. Will they understand it? Most likely not. We are exposing the inner workings of git on the very first command they learn. What's a blob? Why do I need to filter on it? Where are blobs stored? It's classic abstraction leakage.
This is a solved problem: Rsync does it. Just port the bloody implementation over. It does mean supporting alternative representations or moving away from blobs altogether, which git maintainers seem unwilling to do.
And yes, you can fix defaults without breaking backwards compatibility.
Not strictly true. They did change the default push behaviour from "matching" to "simple" in Git 2.0.
I agree with GP. The git community is very fond of doing checkbox fixes for team problems that aren’t or can’t be set as defaults and so require constant user intervention to work. See also some of the sparse checkout systems and adding notes to commits after the fact. They only work if you turn every pull and push into a flurry of activity. Which means they will never work from your IDE. Those are non fixes that pollute the space for actual fixes.
I’m not trying to argue that interface doesn’t matter. I use jq enough to be in that unfortunate category where I despise its interface. But it is difficult for me to imagine being similarly incapable in git.
Maybe not, but that's not the only group of people that are affected. It also affects beginners and people that don't want to exhaustively read the manual.
Should they be the target? Obviously yes.
Can you explain what the solution is? I don't mean the details of the rsync algorithm, but rather what it would like like from the users' perspective. What files are on your local filesystem when you do a "git clone"?
This will sound absolutely insane, but maybe the source code for the video should be a script? Then the process of building produces a video which is a release artifact?
For example:
PromotionalDemo.mp4.script
"Make a video 10 seconds long showcasing the video, a voice in off should say 'We can click here if we want to do this, or click there if we want to go there'. 1024*768 resolution. Male voice. Perky attitude"
It already kinda is, but that just means you now need access to all the raw footage, and rendering a video file in high quality & good compression takes a long time.
This is similar to replicability in science, there's some experiments that are inmensely expensive to replicate, like LHC, but it still IS replicable technically.
Silly idea, but it's worth thinking about this stuff in an era where the line between source code and target code is being blurred with prompts.
Only the histories of the blobs are filtered out.
Based on my experience (YMMV), I think it is incorrect, yes, because any time I've performed a shallow clone of a repository, the saving wasn't as much as one would intuitively imagine (in other words: history is stored very efficiently).
The logical model of git is that it stores complete files. The physical model of git is that these complete files are stored as deltas within pack files (except for new files which haven't been packed yet; by default git automatically packs once there are too many of these loose files, and they're always packed in its network protocol when sending or receiving).
It would be nice to have a VCS that could manage these more effectively but most binary formats don't lend themselves to that, even when it might be an additional layer to an image.
I reckon there's still room for better image and video formats that would work better with VCS.
Nothing wrong with "forgetting" to write the filter, and then if it's taking more than 10 minutes, write the filter.
Is this really the best we can do in terms of user experience? No. git need to step up.
A beginner will follow instructions in a README "Run git clone" or "run git clone --depth=1
The first time you try to commit on a new install, git nags you to set your email address and name. I could see something similar happen the first time you clone a repo that hits the default global filter size, with instructions on how to disable it globally.
> The cloned repo might not be compilable/usable since the blobs are missing.
Maybe I misunderstood the article, but isn't the point of the filter to prevent downloading the full history of big files, and instead only check out the required version (like LFS does).
So a filter of 1 byte will always give you a working tree, but trying to checkout a prior commit will require a full download of all files.
Tying it all in with GitLab, Artifactory, CodeSonar, Anchore etc
Release artifacts like a .exe would NOT belong in Git because it is not source code.
Commits become multi-step, as you need to first commit the artifacts to get their artifact IDs to put in the repo. You can automate that via git hooks, but then you're back at where you started: git-lfs.
[1] https://stackoverflow.com/questions/2144406/how-to-make-shal...
With a good build system using a shared cache, it makes for a very pleasant development environment.
What does SVN do differently than git when it comes to large binary files, and why can't git use the same approach?
I also don't quite understand tbh how offloading large files to somewhere else would be fundamentally different than storing all files in one place except complicating everything? Storage is storage, how would a different storage location fix any of the current performance and robustness problems? Offloading just sounds like a solution for public git forges which don't want to deal with big files because it's too costly for them, but increased hosting cost is not the 'large binary file problem' of git.
(edit: apparently git supports proper locking(?) so I removed that section - ps: nvm it looks like the file locking feature is only in git-lfs)
People should use the VCS that's appropriate for their project rather than insist on git everywhere.
I can't imagine living without that feature, but I also do a lot of OSS work so I'm probably biased.
And if you and another developer make conflicting changes while offline? What should happen when you return online?
E.g. with current svn you get the latest changes from the server, open a diff editor, fix the conflicts and then commit.
The only difference here between svn and git is that svn merges the 'commit' and 'push' operations into one, e.g. instead of not being allowed to push, you're not allowed to commit in svn if there are pending conflicts.
This would be the part that would need to change if svn would get a proper 'offline mode', e.g. commits would need to go into some sort of 'local staging queue' until you get internet access back, and conflict resolutions would need to happen on the commits in that staging queue. But I really doubt if that's worth the hassle because how often are you actually without internet while coding?
Also, designing around distribution meant that merges have to be fast and work well -- this is a problem that most centralised systems struggle with because it's not a key part of the workflow. Branching and merging are indispensable tools in version control and I'm shocked how long CVS and SVN survived despite having completely broken systems for one or both. Being able to do both (and do stuff like blames over the entire history) without needing to communicate with a server is something that changes the way you work.
My actual hot take (as a kernel developer) is that email patches are good, actually. I really wish more people were aware of how incredibly powerful they are -- I have yet to see a source forge that can get close to the resiliency and flexibility of email (though SourceHut gets very close). git-send-email has its quirks, but b4 honestly makes it incredibly easy.
(There's also the owning your own data benefits that I think often go overlooked -- reflog and local clones mean that you cannot practically lose data even if the server goes away or you get blocked. I suspect Russian or Iranian developers losing access to their full repo history because of US sanctions wouldn't share your views on centralised services.)
Which is kinda funny because most people use git through Github or Gitlab, e.g. forcing git back into a centralized model ;)
> People should use the VCS that's appropriate for their project rather than insist on git everywhere.
Indeed, but I think that train has left long ago :)
I had to look it up after I wrote that paragraph about locking, but it looks like file locking is supported in Git (just weird that I need to press a button in the Gitlab UI to lock a file:
https://docs.gitlab.com/user/project/file_lock/
...and here it says it's only supported with git lfs (so still weird af):
https://docs.gitlab.com/topics/git/file_management/#file-loc...
...why not simply 'git lock [file]' and 'git push --locks' like it works with tags?
I think we really need more development of format-specific diff and merge tools. A lot of binary formats could absolutely be diffed or merged, but you'd need algorithms and maybe UIs specific to that format - there is no "generic" algorithm like for text-based files. (And even there, generic line-wise diffing if often more "good enough" than really good)
I think it would be awesome if we could get "diff/merge servers" analogous to the "language servers" for IDEs some day.
The alternative of preventing complex merge situations in the first place through file locking is low-tech, easy to implement, and automatically works on all current and future file formats.
The problem was that the scene information was fundamentally visual (assets arranged in 3D space) so even a diffable text format wouldn't help you much. On the other hand, scenes are large enough that you often would want to work on them in parallel with other people.
I believe their first solution to that was the Asset Server that supported locking. But that still doesn't give two people the ability to work on a scene concurrently.
Eventually, some users went and developed a custom diff/merge tool to solve the problem.
https://discussions.unity.com/t/scene-diff-ease-your-sufferi...
I've never done exactly that but I have occasionally decided how information will be represented in a data file with merging in mind.
https://github.com/ewanmellor/git-diff-image/blob/master/REA...
Of course if you’re working with others you will want a central Git server you all synchronize local changes with. GitHub is just one of many server options.
Disagree. I really like the "de-facto standard" that git has become for open source. It means if I want to understand some new project's source code, there is one less hassle for me to deal with: I don't need to learn any new concepts just to access the source code and all the tooling is already right there.
The situation we have with package managers, dependency managers and package managers for package managers is worse enough. I really don't want a world in which every language or every project also comes with its own version control system and remote repo infrastructure.
It's only git which has this fractal feature set which requires expert knowledge to untangle.
If nothing else, you have to install it. There will also be subtle differences between concepts, e.g. git and svn both have versions and branches, but the concepts behave differently. I don't know about Mercurial, but I'm sure they have their own quirks as well.
Also, tooling: I have a VSCode plugin that visualizes the entire graph structure of a git repo really nicely. Right now, I can use that on 99% of all repos to get an overview of the branches, last commits, activity, etc.
If version systems were fragmented, I'd have to look for equivalent tools for every versioning system separately - if they exist at all. More likely, I'd be restricted just to the on-board tools of every system.
They’re similar in the UI but the underlying architecture is vastly different, to accomplish different goals - sometimes what you want is an entirely centralized VCS, decentralized VCS, or a mix of both.
As for the tooling, any decent IDE supports different systems equally well. With IntelliJ I can use Git, SVN, and even CVS through the same UI. But yes, VSCode plugin XYZ doesn’t.
But git is likely to be appropriate almost everywhere. You won’t just use svn just for big file purposes while git is better for everything else in the same project
Well yeah because text files are small. Thinking text files are insignificant to games because they are small is a really dumb perspective.
> Yet still people try to use git for version control in game projects just because it is the popular option elsewhere and git is all they know.
Or perhaps it's because it works really well for text files, which are a significant part of most games, and because the tooling is much better than for other VCS's.
Fact is that code is only one aspect of a game project, and arguably not the most important. Forcing a programmer-centric workflow on artists and designers is an even dumber perspective ;)
> and because the tooling is much better than for other VCS's
...only for text files. For assets like images, 3d models or audio data it's pretty much a wasteland.
In games a lot of the tooling assumes P4 so it's often a better choice, on the whole, but if git and LFS was as widely supported in art tooling it would be the clear choice.
A lot of people don't seem to realise this. I work in game dev and SVN or Perforce are far far better than Git for source control in this space.
In AA game dev a checkout (not the complete history, not the source art files) can easily get to 300GB of binary data. This is really pushing Subversion to it's limits.
In AAA gamedev you are looking at a full checkout of the latest assets (not the complete history, not the source art files) of at least 1TB and 2TB is becoming more and more common. The whole repo can easily come in at 100 TB. At this scale Perforce is really the only game in town (and they know this and charge through the nose for it).
In the movie industry you can multiply AAA gamedev by ~10.
Git has no hope of working at this scale as much as I'd like it to.
The above should work. But does git support multiple filters for a file? For example first the above asset split filter and then store the files in LFS which is another filter.
I hope this "new" system works but I think Perforce is safe for now.
Github/gitlab is miles ahead of anything you can get with Perforce. People are not just pushing for git because they ux of it, they're pushing git so they can use the ecosystem.
There’s nothing technically worse about perforce with regard to CI/CD, if that’s what you’re talking about, except that of course there are more options for git where you just click a button and enter your credentials and away you go. But that’s more a function of the fact that there aren’t as many options for perforce hosting and companies are more likely to host it internally themselves.
If companies are using perforce but aren’t running any CI/CD that’s because they’re lazy/small/don’t see the value.
Swarm feels like a minimal effort to me and just isn't as frictionless as GitHub. Id rather have Swarm than not, but it's not great.
Storage is not storage as you can store things as copies or diffs (and a million other ways). For code, diffs are efficient but for binaries, diffs approach double the size of the original files, simply sending/storing the full file is better.
These differences have big effects on how git operates and many design choices assumed diffable text assets only.
If you do a little research, there's plenty of information on the topic.
Prolly trees are very similar to Merkle trees or the rsync algorithm, but they support mutation and version history retention with some nice properties. For example: you always obtain exactly the same tree (with the same root hash) irrespective of the order of incremental edit operations used to get to the same state.
In other words, two users could edit a subset of a 1 TB file, both could merge their edits, and both will then agree on the root hash without having to re-hash or even download the entire file!
Another major advantage on modern many-core CPUs is that Prolly trees can be constructed in parallel instead of having to be streamed sequentially on one thread.
Then the really big brained move is to store the entire SCM repo as a single Prolly tree for efficient incremental downloads, merges, or whatever. I.e.: a repo fork could share storage with the original not just up to the point-in-time of the fork, but all future changes too.
If there’s a new algorithm out there that warrants a look…
We have also talked about doing something similar for tree objects in order to better support very large directories (to reduce the amount of data we need to transfer for them) and very deep directories (to reduce the number of roundtrips to the server). I think we have only talked about that on Discord so far (https://discord.com/channels/968932220549103686/969291218347...). It would not be compatible with Git repos, so it would only really be useful to teams outside Google once there's an external jj-native forge that decides to support it (if our rough design is even realistic).
Maybe we can fork something like codeberg's ui but use jj or maybe the jujutsu team can work with codeberg itself? I am pretty sure that codeberg team is really nice and this could be an experimental feature, which if it really needs to can be crowdfunded by community.
I will chip in the first dollar.
And yes you can represent a whole repo as a giant tar file, but because the boundaries between hash segments won't line up with your file boundaries you get an efficiency hit with very little benefit. Unless you make it file-aware in which case it ends up even closer to what git already does.
Git knows how to store deltas between files. Making that mechanism more reliable is probably able to achieve more with less.
Merging would require support from the DB engine, however.
One of our projects has a UI editor with a 60MB file for nearly everything except images, and people work on different UI flows at the same time.
For your UI editor I'd need to understand the format more.
Alternatively, apps could detect and support SCM storage distinct from ordinary file storage.
As a trivial example: Office documents are just zip files, so compatible apps could just save them without compression when saving to SCM-managed folders.
Or better yet, compression, but restart it for each contained file to enable efficient binary diffs.
I was able to convince them to put in line endings though, so it diffs and compresses well, but GitHub will kick it out in another 40MB or so.
Then the usual settings would be to shallow clone the latest content as well as fetch the full history and maybe the text file historical content. Ideally you could prune to the clone depth settings as well.
Why are we still talking about large file pointers? If you fix shallow and partial clones, then any repo can be an efficient file mirror, right?
> if I git clone a repo with many revisions of a noisome 25 MB PNG file
FYI ‘noisome’ is not a synonym for ‘noisy’ - it’s more of a synonym for ‘noxious’; it means something smells bad.
Does anyone have feedback about personally using DVC vs LFS?
[0]: https://github.com/datalad/datalad/blob/maint/datalad/suppor...
For me, the deciding factor was that with LFS, if you want to delete objects from storage, you have to rewrite git history. At least, that's what both the Github and Gitlab docs specify.
DVC adds a layer of indirection, so that its structure is not directly tied to git. If I change my mind and delete the objects from S3, dvc might stop working, but git will be fine.
Some extra pluses about DVC: - It can point to versioned S3 objects that you might already have as part of existing data pipelines. - It integrates with the Python fsspec library to read the files on demand using paths like "dvc://path/to/file.parquet". This feels nicer than needing to download all the files up front.
Switched to https://github.com/kevin-hanselman/dud and I have been happy since ..
https://github.com/Oxen-AI/Oxen
or check out the performance numbers https://docs.oxen.ai/features/performance
Any ideas why it isn’t more popular and more well known?
While git-annex works very well on Unix-style systems with Unix-style filesystems, it heavily depends on symbolic links, which do not exist on filesystems like exFAT, and are problematic on Windows (AFAIK, you have to be an administrator, or enable an obscure group policy). It has a degraded mode for these filesystems, but uses twice the disk space in that mode, and AFAIK loses some features.
But it's not intended for or good at (without forcing a square peg into a round hole) the sort of thing LFS and promisors are for, which is a public project with binary assets.
git-annex is really for (and shines at) a private backup solution where you'd like to have N copies of some data around on various storage devices, track the history of each copy, ensure that you have at least N copies etc.
Each repository gets a UUID, and each tracked file has a SHA-256 hash. There's a branch which has a timestamp and repo UUID to SHA-256 mapping, if you have 10 repos that file will have (at least) 10 entries.
You can "trust" different repositories to different degrees, e.g. if you're storing a file on both some RAID'd storage server, or an old portable HD you're keeping in a desk drawer.
This really doesn't scale for a public project. E.g. I have a repository that I back up my photos and videos in, that repository has ~700 commits, and ~6000 commits to the metadata "git-annex" branch, pretty close to a 1:10 ratio.
There's an exhaustive history of every file movement that's ever occurred on the 10 storage devices I've ever used for that repository. Now imagine doing all that on a project used by more than one person.
All other solutions to tracking large files along with a git repository forgo all this complexity in favor of basically saying "just get the rest where you cloned me from, they'll have it!".
We have an open source CLI and server that mirrors git, but handles large files and mono repos with millions of files in a much more performant manner. Would love feedback if you want to check it out!
To put it another way, regardless of what max size you give to --filter, you will end up with a complete git checkout, no missing files.
At some point, what you're asking for may become a configuration option, likely tied to git gc. It's not there yet because it's all experimental and nobody is paying developers to work on the feature.
The old repo will still be pointed to whatever the LFS config was at that time. If that service is still up, it should continue to work.
In my case, with a 25GB repo, it was really detrimental to performance
These new features are pretty awesome too. Especially separate large object remotes. They will probably enable git to be used for even more things than it's already being used for. They will enable new ways to work with git.
This doesn’t seem to solve the issue of the long download time / big repo size. Or is the author describing the use of promisors remotes in conjunction with partial cloning? Unless I missed something it is not stated explicitly.
I’ve never used it for anything serious but my understanding is that Mercurial handles binary files better? Like it supports binary diffs if I understand correctly.
Any reason Git couldn’t get that?
Are there many binaries that people would store in git where this would actually help? I assume most files end up with compression or some other form of randomization between revisions making deduplication futile.
https://xethub.com/blog/benchmarking-the-modern-development-...
Editing the ID3 tag of an MP3 file or changing the rating metadata of an image will give a big advantage to block level deduplication. Only a few such cases are needed to more than compensate for that worse than nothing inefficiencies of binary diffs when there's nothing to deduplicate.
(It's been even longer since i used svn in anger, but maybe it could work too. It has file locking, and local storage cost is proportional to size of head revision. It was manageable enough with a 2 GB head revision. Metadata access speed was always terrible though, which was tedious.)
All deltas between versions are binary diffs.
Git has always handled large (including large binary) files just fine.
What it doesn't like is files where a conceptually minor change changes the entire file, for example compressed or encrypted files.
The only somewhat valid complaint is that if someone once committed a large file and then it was later deleted (maybe minutes later, maybe years later) then it is in the repo and in everyone's checkouts forever. Which applies equally to small and to large files, but large ones have more impact.
That's the whole point of a version control system. To preserve the history, allowing earlier versions to be recreated.
The better solution would be to have better review of changes pushed to the master repo, including having unreviewed changes in separate, potentially sacrificial, repos until approved.
But a good segmentation is only good for better compression and nicer diff, git could do byte wise diffs with no issues, so I wonder why doesn't git use customizable segmentation strategies where it calls external tools based on file type (eg a rust thingy for rust file etc, or a PNG thingy for PNG files).
At worst the tool would return either a single segment for the entire file or the byte wise split which would work anyway
At the moment I'm using my own git server + git lfs deduplication using btrfs to efficiently handle the large files.
If large objects are just embedded in various packfiles this approach would no longer work, so I hope that such a behaviour can be controlled.
Then you can clone without checking out all the unnecessary large files to get a working build, This also helps on the legal side to correctly license your repos.
I'm struggling to see how this is a problem with git and not just antipatterns that arise from badly organized projects.
Write your code/tools as if they will be used at 2:00 am while the server room is on fire. Because sooner or later they will be.
A lot of our processes are used like emergency procedures. Emergency procedures are meant to be brainless as much as possible. So you can reserve the rest of your capacity for the actual problem. My version essentially calls out Kernighan’s Law.
Git does the responsible thing and lets the user determine how to proceed with the mess they've made.
I must say I'm increasingly suspicious of the hate that git receives these days.
I just want my files to match what's expected when I pull a commit, that doesn't require some literal "commit build system" and "pull build system". Coming from perforce and SVN, I can't comprehend why git it so popular, beyond cargo cult. It's completely nonsensical to think that software is just source.
The point of a clone without checkout is so you can pick the files you really need. This is not a trivial detail that should be automated away. If the user doesn't know which files matter that's an organizational and project management problem, not a git problem. Developers should absolutely care about such details.
The usual workflow is that you develop a feature branch and only push when you have final changes ready to merge and described by a single or just a few commits so that the history isn't a mess. Someone should be reviewing your changes after all. Git is for versioning, not nightly backups. It's also generally preferred that you commit your project flies for your tools, not the binaries from the build output.
The source is always more important because the goal is to understand and review the changes to the project, not just dump everything in there and make managing the project a nightmare. Your use case sounds unusual and likely caused by many binary file changes being committed frequently. You might be correct that git isn't the right tool for that, but it never claimed to be. This doesn't mean git is a bad tool, but that your workflow doesn't seem to care about reviewing every detail before merge.
Yes, this is the required inevitable hack to get around it, which involves rewriting the history in various, creative, ways, which (last I checked) changes all the git commits SHAs, invalidating all the issue tracker links, and everyone doing a bunch of manual work to get things sane on their local sides, which are now unrelated history.
Git was not build for any of this.
Commit IDs are based on a number of factors about the commit, including the actual contents and the commit ID of the parent commit. Any fully cloned git repository can theoretically be audited to make sure that all its commit IDs are correct. Nobody does this (although perhaps git does automatically?), but it's possible.
But now, picture a git repository that has a one petabyte file in one of its early commits (and deleted again later). Pretty much nobody is going to have the space required to download this, so many people will not even bother to do so. As such, what's to stop the server from just claiming any commit ID it wanted for this particular commit? Who's going to check?
(Bonus: For that matter, is the one petabyte file even real? Or just a faked size in the metadata?)
To be clear, I assume people have already thought about these issues. I'm just curious what the answers are.
Of course, remote files can be used to sneak things in, but those things still have to get approved the same as any other commit content. You should not approve PRs etc. that reference remote files you haven't verified. And, while remote storage could be vulnerable to collision attacks in a way that git itself mostly isn't, git-lfs for example already uses SHA-256.
There's a bunch of binary files that change a lot on small changes due to compression or how the data is serialised, so the problem doesn't go away completely. One could conceivably start handling that, but there are lots of file formats out there and the sum oc complexity tends to be bugs and security issues.
Another alternative would be storing the chunks as blobs so that you reconstruct the full binary and only have to store the changed chunks. However that doesn't work with compressed binaries.
The real problem is that Git wants you to have a full copy of all files that have ever existed in the repo. As soon as you add a large file to a repo it's there forever and can basically never be removed. If you keep editing it you'll build up lots more permanent data in the repo.
Git is really missing:
1. A way to delete old data.
2. A way for the repo to indicate which data is probably not needed (old large binaries).
3. A way to serve large files efficiently (from a CDN).
Some of these can sort of be done, but it's super janky. You have to proactively add confusing flags etc.
Yes, the initial commit is a permanent pain in the ass for initial clones, but after that, the deltas should be efficient.
Better to fix that than try to exclude the large files in the first place.
It's also a question of what is a "large" file.
We (I'm on the xet team at HF) are open sourcing our spec + protocol in the coming months. So, with the spec and protocol open-sourced, anyone can create xet clients and implement the protocol to build a xet backend.
The specific implementation of our xet backend is deeply integrated into HF backend so open-sourcing it directly wouldn't be very helpful. Once we get the spec + protocol released it should be easy to generate a compatible backend.
I am wondering since a lot of people have said that for things like gamedev etc., git isn't best because of lfs issues and xet fixes it, could I use huggingface as a repository if I am gamedev or would that go against since huggingface is for AI stuff?
Also thanks for open sourcing it, I hope that it would be in permissive license and open sourcing the spec + protocol might actually push the git community to be better which is really nice!
That was actually an initial selling point of git: you have the full history locally. You can work from the plane/train/deserted island just fine.
These large files will persist in the repo forever. So people look for options to segregate large files out so that they only get downloaded on demand (aka "lazily").
All the existing options (submodules, LFS, partial clones) are different answers to "how do we make certain files only download on demand"
Don't forget sparse checkouts!
Nice to see some Microsoft and Google emails contributing.
Google has android and chromium as well as Git+Gerrit-on-Borg https://opensource.google/documentation/reference/releasing/...
(only half-trolling)
A good version control system would support petabyte scale history and terabyte scale clones via sparse virtual filesystem.
Git’s design is just bad for almost all projects that aren’t Linux.
(I know this will get downvoted. But most modern programmers have never used anything but Git and so they don’t realize their tool is actually quite bad! It’s a shame.)
This is only possible because git is decentralized. Claiming that git is centralized is complete falsehood.
But I encourage everyone to try out a few alternatives (and adopt their workflows at least for a while). I have no idea if you have or not.
But fine has never used the alternatives, one doesn’t really know just how nice things can be. Or, even if you still find fit to be your preferred can, having an alternative experience can open you to other possibilities and ways of working.
Just like everyone should try a couple of different programming languages or editors or anything else for size. You may not end up choosing it, but seeing the possibilities and different ways of thinking is a very good thing.
If you use Git it's not possible to "have everything" because Git is not capable of storing everything that needs to be version controlled. Most devs their version control is actually a combination of Git, docker images, and lord knows how many other data sources. It's a miserable nightmare. Good luck and God speed to anyone who tries to build something from just a few years ago, never mind a 10+ year old project!
> All the features you think are associated with the D in DVCS are perfectly accessible to a more centralized tool.
You moved the goal posts a little bit. You are correct that if you want to work in a cave with no internet then you do need every bit of data to perform at least some operation. That’s not what the D in DVCS means though.
For me connecting to the Internet is very much a choice. Doesn't mean I wouldn't have access if I didn't want to, but why would I need it most of the time?
I like this idea in principle but I always wonder what that would look in practice, outside a FAANG company: How do you ensure the virtual file system works equally well on all platforms, without root access, possibly even inside containers? How do you ensure it's fast? What do you do in case of network errors?
Tom Lyon: NFS Must Die! From NLUUG 2024:
https://www.youtube.com/watch?v=ZVF_djcccKc
>Why NFS must die, and how to get Beyond FIle Sharing in the cloud.
Slides:
https://nluug.nl/bestanden/presentaties/2024-05-21-tom-lyon-...
Eminent Sun alumnus says NFS must die:
https://blocksandfiles.com/2024/06/17/eminent-sun-alumnus-sa...
Re network errors. How many things break when GitHub is down? Quite a lot! This isn’t particularly special. Prefetch and clone are the same operation.
But most people don't need most of its features and many people need features it doesn't have.
If you look up git worktrees, you'll find a lot of blog articles referring to worktrees as a "secret weapon" or similar. So git's secret weapon is a mode that lets you work around the ugliness of branches. This suggests that many people would be better suited by an SCM that isn't branch-based.
It's nice having the full history offline. But the scaling problems force people to adopt a workflow where they have a large number of small git repos instead of keeping the history of related things together. I think there are better designs out there for the typical open source project.
In my experience, branches are totally awesome. Worktrees make branches even more awesome because they let me check out multiple branches at once to separate directories.
The only way it could get better is if it somehow gains the ability to check out the same branch to multiple different directories at once.
So you want shared object storage, but separate branch metadata. That's git clone with hardlinks, which is what Git does locally by default.
shared worktree, shared branch/index data, shared object storage -> single repo, single worktree
separate worktree, shared branch/index data, shared object storage -> repo, with worktrees
separate worktree, separate branch/index data, shared object storage -> git clone
separate worktree, separate branch/index data, separate object storage -> git clone --no-hardlinks
What I tried to do was have two copies of the same branch checked out to a different directory: one pristine, another with uncommitted changes. The idea was to run make on both directories, profile the results and then decide whether to commit or discard the changes. Now I see the solution is to just make another branch.
Developers have their own drama
Still the approach is to put code and data in a folder and call it a day. Slap a "_FINAL" at the folder name and you are golden.
We found that we could move the large files to Artifactory as it has Git LFS support.
But the problem was the entire history that did not have Artifactory pointers. Every clone included the large files (for some reason the filter functionality wouldn’t work for us - it was a large repo and it it had hundreds of users amongst other problems)
Anyways what we ended up doing was closing that repo and opening a new one with the large files stripped.
Nitpick in the authors page:
“ Nowadays, there’s a free tier, but you’re dependent on the whims of GitHub to set pricing. Today, a 50GB repo on GitHub will cost $40/year for storage”
This is not true as you don’t need GitHub to get LFS support