Show HN: CallFS – S3-style object store in one Go binary (MIT)

63•ebogdum•7h ago

We started CallFS after yet another late-night “why did the uploads vanish?” incident. Our small team had stitched together rsync, a fragile NFS mount, and an S3 bucket—none of it observable, all of it waiting to bite us.

So we wrote a single-process file service in Go that: • Speaks the S3 API (so existing tooling works). • Stores hot data on local disks for speed; cold data can sit in any S3-compatible bucket. • Exposes Prometheus metrics and JSON logs by default, because “what happened?” shouldn’t be guesswork. • Ships as a ~25 MB static binary—no external deps, MIT license.

Today it’s stable for single-node or side-by-side deployments. Clustering is on the roadmap, replication will follow, but we wanted to share the code early and hear real-world pain points. If storage glue code ever ruined your weekend, we’d love feedback and PRs.

Comments

cowboyscott•6h ago

I was wanting this to exist today :) Thanks for sharing the work

mt42or•6h ago

https://github.com/juicedata/juicefs

ebogdum•6h ago

SO COOL. This is way more than what my scope was, maybe I can learn some stuff.

I do have some future plans on replication and mutex and specific files from specific instances.

Thanks for sharing this.

evil-olive•4h ago

https://github.com/seaweedfs/seaweedfs is excellent as well

ebogdum•6h ago

Well, glad to be of service and I really hope it gives you what you need. If not, just open an issue and we will see what we can do. :)

Thanks for sharing your thoughts.

mdaniel•5h ago

I can't quite square up your "speaks the S3 API" with https://github.com/ebogdum/callfs#file-operations seeming to do its own thing, which for sure would not leverage existing tooling

And it would have been better if in your "static binary" sentence you had included its dependence upon both PostgreSQL and Redis

ebogdum•5h ago

Very fair points, thank you very much for taking the time. I will make the changes later today to mention the dependencies.

As for the "speaks the S3 API", the idea is that this has two backends right now, Local File System and any S3 compatible API, like AWS S3, MinIO, DigitalOcean Object Storage ... and so on. In other words you can use them all at the same time, provided you have an instance for each.

But you did give me an idea, as to the fact that I need to add some directory/bucket discovery for existing files.

Thank you very much for your feedback.

jasonjmcghee•5h ago

I had the same question, and started looking more deeply at the project.

Would be interested to hear if this was built by OP or vibe-coded.

ebogdum•5h ago

A bit of both, as you can imagine, there are so many tedious tasks that consume time for so little in return, so yeah.

moritonal•4h ago

So you're trying to get buy-in for a tool, when you yourself don't see the point in confirming it works? Those "tedious tasks" are software development.

ebogdum•4h ago

What, you mean writing documentations and readme?

Of course I'm using this. I have a homelab with 12 PI's, this tool helps me play with files between all of them, any way I want. Might not be the world changing usage you were hopping for, but for me it's enough.

evil-olive•4h ago

> I have a homelab with 12 PI's, this tool helps me play with files between all of them

did you have your LLM write the Show HN description too?

because you've gone from "We started CallFS" and "our small team" to "I play with it in my homelab"

if this is a homelab-level project, there's absolutely nothing wrong with that. but you should be careful not to mislead people, even unintentionally, about the stability/maturity of the project.

especially when the project involves data storage. vibe-coding a game is one thing, if it has bugs then you might miss a power-up or get stuck on a level with no way out. when a vibe-coded storage system has bugs, you're potentially losing or silently corrupting user data.

ebogdum•3h ago

You're asking me if I use it, I'm telling you where and how I use it. Then you complain how I use it.

Listen, I can totally respect anyones opinion, but this discussion isn't bringing any benefits, it doesn't seem to be constructive at all.

I get it, you're dissatisfied with this project or whatever, and usually I do try to be very accommodating, as you can see I do engage, but at the same time, please understand, this, is an open-source MIT licensed project, it's not the next "save the earth project" and still, you're acting like you're the VC and already lost money on the investment. Like I have to prove some worth or something.

To answer your question, yeah, there were two of us working at this, for a while, then one lost interest, I thought it would still be fair to say "our small team".

As for the losing data or any other such things, great idea, I will add that there is always a possibility of losing data, just like with NTFS, with AppleFS, or any other FS no matter who developed it or supports it. But it is a good point.

As far as this comment thread goes, I will personally refrain from future comments, as I believe it doesn't serve any good purpose other than nit picking.

Thank you.

MOARDONGZPLZ•1h ago

They’re clearly not complaining at how you use it, they’re calling out your misrepresentations. These are ethical issues and you should examine your own moral framework.

moritonal•2h ago

But you brought it here to get feedback? Our feedback is things like, the docs (https://github.com/ebogdum/callfs/blob/main/docs_markdown/02...) have a pile of references to things that just don't exist in your code? You have a change-log that isn't used? Your main file references a site that doesn't exist "https://github.com/ebogdum/callfs/blob/main/cmd/main.go".

MOARDONGZPLZ•1h ago

Well, to be fair writing documentation is tedious (as are changelogs, and writing code). So they outsourced it to an LLM.

evil-olive•4h ago

> if this was built by OP or vibe-coded

I hope we start to see some etiquette / conventions developed around this, especially for open source projects.

if I showed up to a dinner party with homemade cookies, my friends would be appreciative.

if I showed up with store-bought cookies, my friends would probably still be appreciative.

but if I showed up with store-bought cookies and tried to claim they were homemade, my friends would probably feel insulted. even if the cookies were amazing, they'd be overshadowed by my dishonesty about them.

the problem, of course, is that if we had some standard section in a readme for expressing the spectrum between "0% LLM output, entirely whittled by hand as a labor of love" vs "100% LLM output, human eyes have never actually looked at any of it" we'd have LLMs hallucinate that readme section, saying they were human-coded.

ebogdum•4h ago

Will not argue, pro or against, everyone is entitled to their own opinion. But this reminds me of the early 2000 interviews when every one was asking if you used Google search, and now we use "google it" as a verb.

There was a meme at one point with a dude at an interview answering that with something like this: If you google a recipe, it doesn't mean the food is done, you still need someone to cook it.

As for the "xx% LLM" or not, I would argue the same for "xx% Googled", "xx% Stackoverflow" "xx% Learned in school vs copied from the interned", "xx% actual development and engineering vs library importing and glueing".

thomascountz•3h ago

Likewise, if I show up with homemade cookies, after your friend asks "who are you and why are you in my house," if they then assume my cookies must be store bought and I probably wouldn't be able to have made them myself, I'd think them rather judgmental.

pjc50•2h ago

Then the more serious one of "do these cookies contain allergen X?"

Any of the answers "yes, no, I don't know" is OK, although the latter is irritating, but if you claim "no" when the allergen actually is present it's a disaster. Whether that's because you bought them and didn't look at the ingredients or made them yourself in a careless shared environment.

For code from the Internet there's the related question "how likely is this to contain security holes or data-loss bugs?" And being LLM-written currently makes that a lot more likely.

techn00•5h ago

I am so sick of AI generated README's, they follow the samn damn format.

ebogdum•5h ago

As you can imagine, "story telling" isn't something everyone has mastered, so it does make the job easier. But I do get your point.

Oxodao•4h ago

We don't want story telling. We just want a clear, few lines long paragraph that tells everything you need to know about the software

dawnerd•5h ago

I'm so sick of these vibe coded apps. They all end up having security issues. Definitely wouldn't trust it.

ebogdum•4h ago

You have a very good point.

Definitely, no one would force you to use something you don't find as trustworthy or valuable.

MOARDONGZPLZ•1h ago

KEY FEATURES :key-emoji:

CONTRIBUTING :handshake-emoji:

CORE FEATURES :apple-core-emoji:

TheDong•1h ago

> Speaks the S3 API

Let's test that lol.

First, I looked at the docker compose file, seems like an easy way to run it right? Wrong, the docker compose file just runs minio, a real s3 compatible thing: https://github.com/ebogdum/callfs/blob/16e1096095c809f31aa93...

Why check in a docker-compose that only runs someone else's code, not your own project?

Okay, fine, let's run it normally:

    ./callfs server -c ./config.yaml.example
    failed to load config file ./config.yaml.example: file provider does not support this method

Okay, fine, your example config doesn't work with an inscrutable error.

But running it without a config flag works, so let's do that:

    {"level":"info","ts":1752571569.821955,"caller":"cmd/main.go:251","msg":"Starting HTTPS server","addr":":8443"}

Let's see if it's really s3 compatible now:

    AWS_CA_BUNDLE=./server.crt aws --endpoint-url https://localhost:8443 s3 mb s3://foo

    make_bucket failed: s3://foo An error occurred (404) when calling the CreateBucket operation: Not Found

"404" is not the response AWS gives to a make bucket request.

Reading the actual docs, this is _not_ s3 compatible, not even close. So much for "Speaks the S3 API".

Just use minio, or one of the other actually functioning things in this space.

MOARDONGZPLZ•1h ago

This is a vibe coded project. It doesn’t work very well at all, not going to rehash what others have found, but these projects have been a major plague to the Go community over the last six months. They’re essentially the code version of littering.

LLM Inevitabilism

Code highlighting with Cursor AI used for $500k theft

Apple's MLX adding CUDA support

Kiro: A new agentic IDE

LIGO detects most massive black hole merger to date

Dog Walk: Blender Studio's official game project

The JPEG XL Image Coding History, Features, Coding Tools, Design Rationale

RFC: PHP license update

VHS, VCDs, and Laserdiscs in Southeast Asia

When Sigterm Does Nothing: A Postgres Mystery

SQLite async connection pool for high-performance

Clashes between web and X11 colors in the CSS color scheme

Context Rot: How increasing input tokens impacts LLM performance

I used Suno AI to cover my own demo album

Show HN: Bedrock – An 8-bit computing system for running programs anywhere

Show HN: VS Code extension to edit the filesystem like a text buffer

C++ Coroutines Advanced: Converting std:future to asio:awaitable

I Solved the Century-Old Mystery of a Miraculous Shipwreck Survivor

Martin (YC S23) Is Hiring Founding Engineers to Build a Better Siri

Happy 20th Birthday, Django

VMware Workstation: Bringing Virtualization to the x86 Architecture [pdf] (2012)

NeuralOS: An operating system powered by neural networks

Replicube: 3D shader puzzle game, online demo

DEWLine Museum – The Distant Early Warning Radar Line

Doge Denizen Marko Elez Leaked API Key for XAI

Strategies for Fast Lexers

Cache Benchmarks

Building Modular Rails Applications: A Deep Dive into Rails Engines

Cognition (Devin AI) to Acquire Windsurf

Embedding user-defined indexes in Apache Parquet

LLM Inevitabilism

Code highlighting with Cursor AI used for $500k theft

Apple's MLX adding CUDA support

Kiro: A new agentic IDE

LIGO detects most massive black hole merger to date

Dog Walk: Blender Studio's official game project

The JPEG XL Image Coding History, Features, Coding Tools, Design Rationale

RFC: PHP license update

VHS, VCDs, and Laserdiscs in Southeast Asia

When Sigterm Does Nothing: A Postgres Mystery

SQLite async connection pool for high-performance

Clashes between web and X11 colors in the CSS color scheme

Context Rot: How increasing input tokens impacts LLM performance

I used Suno AI to cover my own demo album

Show HN: Bedrock – An 8-bit computing system for running programs anywhere

Show HN: VS Code extension to edit the filesystem like a text buffer

C++ Coroutines Advanced: Converting std:future to asio:awaitable

I Solved the Century-Old Mystery of a Miraculous Shipwreck Survivor

Martin (YC S23) Is Hiring Founding Engineers to Build a Better Siri

Happy 20th Birthday, Django

VMware Workstation: Bringing Virtualization to the x86 Architecture [pdf] (2012)

NeuralOS: An operating system powered by neural networks

Replicube: 3D shader puzzle game, online demo

DEWLine Museum – The Distant Early Warning Radar Line

Doge Denizen Marko Elez Leaked API Key for XAI

Strategies for Fast Lexers

Cache Benchmarks

Building Modular Rails Applications: A Deep Dive into Rails Engines

Cognition (Devin AI) to Acquire Windsurf

Embedding user-defined indexes in Apache Parquet

Show HN: CallFS – S3-style object store in one Go binary (MIT)

Comments