Garage – An S3 object store so reliable you can run it outside datacenters

Posted by ibobev 1 day ago

Comments

Comment by adamcharnock 1 day ago

Copy/paste from a previous thread [0]:

We’ve done some fairly extensive testing internally recently and found that Garage is somewhat easier to deploy in comparison to our existing use of MinIO, but is not as performant at high speeds. IIRC we could push about 5 gigabits of (not small) GET requests out of it, but something blocked it from reaching the 20-25 gigabits (on a 25g NIC) that MinIO could reach (also 50k STAT requests/s, over 10 nodes)

I don’t begrudge it that. I get the impression that Garage isn’t necessarily focussed on this kind of use case.

---

In addition:

Next time we come to this we are going to look at RustFS [1], as well as Ceph/Rook [2].

We can see we're going to have to move away from MinIO in the foreseeable future. My hope is that the alternatives get a boost of interest given the direction MinIO is now taking.

[0]: https://news.ycombinator.com/item?id=46140342

[1]: https://rustfs.com/

[2]: https://rook.io/

Comment by johncolanduoni 19 hours ago

Somewhat unrelated, but I just looked at the RustFS docs intro[1] after seeing it here. It has this statement:

> RustFS is a high-performance, distributed object storage software developed using Rust, the world's most popular memory-safe language.

I’m actually something of a Rust booster, and have used it professionally more than once (including working on a primarily Rust codebase for a while). But it’s hard to take a project’s docs seriously when it describes Rust as “the world’s most popular memory-safe language”. Java, JavaScript, Python, even C# - these all blow it out of the water in popularity and are unambiguously memory safe. I’ve had a lot more segfaults in Rust dependencies than I have in Java dependencies (though both are minuscule in comparison to e.g. C++ dependencies).

[1]: https://docs.rustfs.com/installation/

Comment by riedel 2 hours ago

>Secure: RustFS is written in Rust, a memory-safe language, so it is 100% secure

[0]

qed

[0] https://docs.rustfs.com/installation/

Comment by tormeh 36 minutes ago

Oh wow, they really wrote that. I love Rust but this is clown shit.

Comment by woodruffw 3 hours ago

I agree, although I’m guessing they’re measuring “most popular” as in “most beloved” and not as in “most used.” That’s the metric that StackOverflow puts out each year.

Comment by PunchyHamster 12 hours ago

The docs of it and the marketing is a bit of a mess, tho I'm just gonna blame that on culture barrier as the devs are chinese

Comment by teiferer 16 hours ago

It's hard to take a project seriously if it focuses so much on the language it's written in. As a user, I don't care. Show me the results (bug tracker with low rate of issues), that's what I care about. Whether you program in Rust or C or Java or assembly or PHP.

Comment by limagnolia 5 hours ago

As a potential user of an open source project, I care a fair bit what language it is implemented in. As an open source project, I preffer projects in languages and ecosystems I am familair and comfortable with. I may need to fix bugs, add features, or otherwise make contributions back to the project, and thus I am more likely to pick a solution in a language I am comfortable with than in a language I am not as comfortable with, given my other needs and priorities are met.

Comment by b112 18 hours ago

I feel dumbfounded. All I've ever heard from rust users, is the equivalent of football fans running up, waving pendants in my face and screaming. So much so, that everything else said seems like the wild fantasies of "our team gonna win".

Then things like this appear:

https://www.phoronix.com/news/First-Linux-Rust-CVE

And I'm all warm and feeling schadenfreude.

To hear "yes, it's safer" and yet not "everyone on the planet not using rust is a moron!!!", is a nice change.

Frankly, the whole cargo side of rust has the same issues that node has, and that's silly beyond comprehension. Memory safe is almost a non-concern, compared to installing random, unvetted stuff. Cargo vet seems barely helpful here.

I'd want any language caring about security and code safety, to have a human audit every single diff, on every single package, and host those specific crates on locked down servers.

No, I don't care about "but that will slow down development and change!". Security needs to be first and front.

And until the Rust community addresses this, and its requirement for 234234 packages, it's a toy.

And yes, it can be done. And no, it doesn't require money. Debian's been doing just this very thing for decades, on a far, far, far larger scale. Debian developers gatekeep. They package. They test and take bug reports on specific packages. This is a solved problem.

Caring about 'memory safe!' is grand, but ignoring the rest of the ecosystem is absurd.

Comment by necovek 16 hours ago

Debian has been doing this for decades, yes, but it is largely a volunteer effort, and it's become a meme how slow Debian is to release things.

I've long desired this approach (backporting security fixes) to be commercialized instead of the always-up-to-date-even-if-incompatible push, and on top of Red Hat, Suse, Canonical (with LTS), nobody has been doing it for product teams until recently (Chainguard seems to be doing this).

But, if you ignore speed, you also fail: others will build less secure products and conquer the market, and your product has no future.

The real engineering trick is to be fast and build new things, which is why we need supply chain commoditized stewards (for a fee) that will solve this problem for you and others at scale!

Comment by PunchyHamster 12 hours ago

> Debian has been doing this for decades, yes, but it is largely a volunteer effort, and it's become a meme how slow Debian is to release things.

which is a bit silly considering that if you want fast, most packages land in testing/unstable pretty quickly.

Comment by necovek 11 hours ago

But then you as a consumer/user of Debian packages need to stay on top of things when they change in backwards-incompatible ways.

I believe the sweet spot is Debian-like stable as the base platform to build on top of, and then commercial-support in a similar way for any dependencies you must have more recent versions on top.

Comment by PunchyHamster 10 hours ago

> But then you as a consumer/user of Debian packages need to stay on top of things when they change in backwards-incompatible ways.

If you need latest packages, you have to do it anyway.

> I believe the sweet spot is Debian-like stable as the base platform to build on top of, and then commercial-support in a similar way for any dependencies you must have more recent versions on top.

That if the company can build packages properly. Also too old OS deps sometimes do throw wrench in the works.

Tho frankly "latest Debian Testing" have far smaller chance breaking something than "latest piece of software that couldn't figure out how to upstream to Debian"

Comment by necovek 7 hours ago

The difference is between staying on stable and cherry-picking the latest for what you really do need, and being on everything latest.

The latter has a huge maintenance burden, the former is the, as I said already, sweet spot. (And let's not talk about combining stable/testing, any machine I tried that on got into an non-upgradeable mess quickly)

I am not saying it is easy, which is exactly why I think it should be a commercial service that you pay for for it to actually survive.

Comment by dotancohen 9 hours ago

  > supply chain commoditized stewards (for a fee)

I agree with this, but the open source licenses allow anyone who purchases a stewarded implementation to distribute it freely.

I would love to see a software distribution model in which we could pay for vetted libraries, from bodies that we trust, which would become FOSS after a time period - even a month would be fine.

There are flaws in my argument, but it is a safer option than the current normal practices.

Comment by sporkland 17 hours ago

Not dismissing your point, but Looking at the article, it looks like it's in rust unsafe code. Which seems to me to be a point that the rest of the rust code is fine but the place where they turned off the static safety the language provides they got bit.

Comment by b112 17 hours ago

Hey! Can't I just enjoy my schadenfreude in peace?

I guess the takeaway is that, doubly so, trusting rust code to be memory safe, simply because it is rust isn't sensible. All its protections can simple be invalidated, and an end user would never know.

Comment by teiferer 16 hours ago

Um I doubt Debian maintainers look at every single line of code in the packages they maintain.

Comment by SEJeff 15 hours ago

One might even call the rust community a “cargo cult”

Comment by nine_k 1 day ago

They explicitly say that top performance is not a goal: «high performances constrain a lot the design and the infrastructure; we seek performances through minimalism only» (https://garagehq.deuxfleurs.fr/documentation/design/goals/)

But it might be interesting to see where the time is spent. I suspect they may be doing fewer things in parallel than MinIO, but maybe it's something entirely different.

Comment by PunchyHamster 12 hours ago

My small adventure with rustfs is that it is somewhat underbaked at the moment.

And also it is already rigged for a rug-pull

https://github.com/rustfs/rustfs/blob/main/rustfs/src/licens...

Comment by evil-olive 5 hours ago

yeah, their docs look pretty comprehensive, but there's a disturbing number of 404s that scream "not ready for prime-time" to me.

from https://rustfs.com/ if you click Documentation, it takes you to their main docs site. there's a nav header at the top, if you click Docs there...it 404s.

"Single Node Multiple Disk Installation" is a 404. ditto "Terminology Explanation". and "Troubleshooting > Node Failure". and "RustFS Performance Comparison".

on the 404 page, there's a "take me home" button...which also leads to a 404.

Comment by __turbobrew__ 1 day ago

I wouldn’t use rook if you solely want S3. It is a massively complex system which you really need to invest in understanding or else your cluster will croak at some point and you will have no idea on how to fix it.

Comment by breakingcups 1 day ago

IS there a better solution for self-healing S3 storage that you could recommend? I'm also curious what will make a rook cluster croak after some time and what kind of maintenance is required in your experience.

Comment by __turbobrew__ 19 hours ago

I have unfortunately got a ceph cluster in a bad enough state that I just had to delete the pools and start from scratch. It was due to improper sequencing when removing OSDs, but that is kindof the point is you have to know what you are doing to know how to do things safely. For the most part I have so far learned by blundering things and learning hard lessons. Ceph clusters when mistreated can get into death spirals that only an experienced practitioner can advert through very carefully modifying cluster state through things like upmaps. You also need to make sure you understand your failure domains and how to spread mons and osds across the domains to properly handle failure. Lots of people don’t think about this and then one day a rack goes poof and you didn’t replicate your data across racks and you have data loss. Same thing with mons, you should be deploying mons across at least 3 failure domains (ideally 3 different datacenters) to maintain quorum during an outage.

Comment by adamcharnock 1 day ago

Not used it yet, but RustFS sounds like it has self healing

https://docs.rustfs.com/troubleshooting/healing.html

Comment by adastra22 1 day ago

ceph?

Comment by yupyupyups 21 hours ago

Rook is ceph.

Comment by hardwaresofton 1 day ago

Please also consider including SeaweedFS in the testing.

Comment by Roark66 13 hours ago

Having just finished a "hobby size" setup of Rook-Ceph on 3 n100 mini pcs, with every service to fit in a couple hundred MB of ram (one service needs up to 3Gb when starting, but then runs around 250MB) I'd ask why not ceph?

At work I'm typically a consumer of such services from large cloud providers. I read in few places how "difficult" it is, how you need "4GB minimum RAM for most services" and how "friends do not let friends run Ceph below 10Gb".

But this setup runs on a non dedicated 2.5Gb interface (there is VLAN segmentation and careful QoSing).

My benchmarks show I'm primarily network latency and bandwidth limited. By the very definition you can't get better than that.

There were many factors why I chose Ceph and not Garage, Seaweed or MinIo. (One of the biggest is that ceph does 2 birds with one stone for me - block and object).

Comment by PunchyHamster 12 hours ago

Ceph is far higher on RAM usage and complexity. Yeah if you need block storage in addition it's a good choice, but for anything smaller than half a rack of devices it's kinda overkill

Also from our experience the docs outright lie about ceph's OSD memory usage and we've seen double or more than what docs claim (8-10GB instead of 4)

Comment by NL807 23 hours ago

>I get the impression that Garage isn’t necessarily focussed on this kind of use case.

I wouldn't be surprised if this will be fixed sometime in the future.

Comment by throwaway894345 20 hours ago

> We can see we're going to have to move away from MinIO in the foreseeable future.

My favorite thing about all of this is that I had just invested a ton of time in understanding MinIO and its Kubernetes operator and got everything into a state that I felt good about. I was nearly ready to deploy it to production when the announcement was released that they would not be supporting it.

I’m somewhat surprised that no one is forking it (or I haven’t heard about any organizations of consequence stepping up anyway) instead of all of these projects to rebuild it from scratch.

Comment by Emjayen 18 hours ago

Those rates are peanuts considering that a decade ago saturating 40G, per core, was more than reasonable via standard userspace networking, with atleast a few copies in the datapath.

Comment by PunchyHamster 12 hours ago

passing blocks of memory around vs referencing filesystem/database, ACLs, authentication and SSL

Comment by fabian2k 1 day ago

Looks interesting for something like local development. I don't intend to run production object storage myself, but some of the stuff in the guide to the production setup (https://garagehq.deuxfleurs.fr/documentation/cookbook/real-w...) would scare me a bit:

> For the metadata storage, Garage does not do checksumming and integrity verification on its own, so it is better to use a robust filesystem such as BTRFS or ZFS. Users have reported that when using the LMDB database engine (the default), database files have a tendency of becoming corrupted after an unclean shutdown (e.g. a power outage), so you should take regular snapshots to be able to recover from such a situation.

It seems like you can also use SQLite, but a default database that isn't robust against power failure or crashes seems suprising to me.

Comment by lxpz 1 day ago

If you know of an embedded key-value store that supports transactions, is fast, has good Rust bindings, and does checksumming/integrity verification by default such that it almost never corrupts upon power loss (or at least, is always able to recover to a valid state), please tell me, and we will integrate it into Garage immediately.

Comment by __padding 48 minutes ago

I’ve not looked at it in a while but sled/rio were interesting up and coming options https://github.com/spacejam/sled

Comment by agavra 1 day ago

Sounds like a perfect fit for https://slatedb.io/ -- it's just that (an embedded, rust, KV store that supports transactions).

It's built specifically to run on object storage, currently relies on the `object_store` crate but we're consdering OpenDAL instead so if Garage works with those crates (I assume it does if its S3 compatible) it should just work OOTB.

Comment by evil-olive 5 hours ago

for Garage's particular use case I think SlateDB's "backed by object storage" would be an anti-feature. their usage of LMDB/SQLite is for the metadata of the object store itself - trying to host that metadata within the object store runs into a circular dependency problem.

Comment by johncolanduoni 19 hours ago

I’ve used RocksDB for this kind of thing in the past with good results. It’s very thorough from a data corruption detection/rollback perspective (this is naturally much easier to get right with LSMs than B+ trees). The Rust bindings are fine.

It’s worth noting too that B+ tree databases are not a fantastic match for ZFS - they usually require extra tuning (block sizes, other stuff like how WAL commits work) to get performance comparable to XFS/ext4. LSMs on the other hand naturally fit ZFS’s CoW internals like a glove.

Comment by fabian2k 1 day ago

I don't really know enough about the specifics here. But my main points isn't about checksums, but more something like WAL in Postgres. For an embedded KV store this is probably not the solution, but my understanding is that there are data structures like LSM that would result in similar robustness. But I don't actually understand this topic well enough.

Checksumming detects corruption after it happened. A database like Postgres will simply notice it was not cleanly shut down and put the DB into a consistent state by replaying the write ahead log on startup. So that is kind of my default expectation for any DB that handles data that isn't ephemeral or easily regenerated.

But I also likely have the wrong mental model of what Garage does with the metadata, as I wouldn't have expected that to be ever limited by Sqlite.

Comment by lxpz 1 day ago

So the thing is, different KV stores have different trade-offs, and for now we haven't yet found one that has the best of all worlds.

We do recommend SQLite in our quick-start guide to setup a single-node deployment for small/moderate workloads, and it works fine. The "real world deployment" guide recommends LMDB because it gives much better performance (with the current status of Garage, not to say that this couldn't be improved), and the risk of critical data loss is mitigated by the fact that such a deployment would use multi-node replication, meaning that the data can always be recovered from another replica if one node is corrupted and no snapshot is available. Maybe this should be worded better, I can see that the alarmist wording of the deployment guide is creating quite a debate so we probably need to make these facts clearer.

We are also experimenting Fjall as an alternate KV engine based on LSM, as it theoretically has good speed and crash resilience, which would make it the best option. We are just not recommending it by default yet, as we don't have much data to confirm that it works up to these expectations.

Comment by BeefySwain 1 day ago

(genuinely asking) why not SQLite by default?

Comment by lxpz 1 day ago

We were not able to get good enough performance compared to LMDB. We will work on this more though, there are probably many ways performance can be increased by reducing load on the KV store.

Comment by srcreigh 1 day ago

Did you try WITHOUT ROWID? Your sqlite implementation[1] uses a BLOB primary key. In SQLite, this means each operation requires 2 b-tree traversals: The BLOB->rowid tree and the rowid->data tree.

If you use WITHOUT ROWID, you traverse only the BLOB->data tree.

Looking up lexicographically similar keys gets a huge performance boost since sqlite can scan a B-Tree node and the data is contiguous. Your current implementation is chasing pointers to random locations in a different b-tree.

I'm not sure exactly whether on disk size would get smaller or larger. It probably depends on the key size and value size compared to the 64 bit rowids. This is probably a well studied question you could find the answer to.

[1]: https://git.deuxfleurs.fr/Deuxfleurs/garage/src/commit/4efc8...

Comment by asa400 2 hours ago

Other than knowing this about SQLite beforehand, is there any way one could discover that this is happening through tracing?

Comment by lxpz 1 day ago

Very interesting, thank you. It would probably make sense for most tables but not all of them because some are holding large CRDT values.

Comment by tensor 1 day ago

Keep in mind that write safety comes with performance penalties. You can turn off write protections and many databases will be super fast, but easily corrupt.

Comment by rapnie 1 day ago

I learned that Turso apparently have plans for a rewrite of libsql [0] in Rust, and create a more 'hackable' SQLite alternative altogether. It was apparently discussed in this Developer Voices [1] video, which I haven't yet watched.

[0] https://github.com/tursodatabase/libsql

[1] https://www.youtube.com/watch?v=1JHOY0zqNBY

Comment by skrtskrt 1 day ago

Could you use something like Fly's Corrosion to shard and distribute the SQLite data? It uses a CRDT reconciliation, which is familiar for Garage.

Comment by lxpz 1 day ago

Garage already shards data by itself if you add more nodes, and it is indeed a viable path to increasing throughput.

Comment by __turbobrew__ 1 day ago

RocksDB possibly. Used in high throughput systems like Ceph OSDs.

Comment by patmorgan23 1 day ago

Valkey?

Comment by VerifiedReports 19 hours ago

It's "key/value store", FYI

Comment by kqr 18 hours ago

It's not a store of "keys or values", no. It's a store of key-value pairs.

Comment by VerifiedReports 7 hours ago

A key-value store would be a store of one thing: key values. A hyphen combines two words to make an adjective, which describes the word that follows:

  A used-car lot

  A value-added tax

  A key-based access system

When you have two exclusive options, two sides to a situation, or separate things; you separate them with a slash:

  An on/off switch

  A win/win situation

  A master/slave arrangement

Therefore a key-value store and a key/value store are quite different.

Comment by kqr 6 hours ago

All of your slash examples represent either–or situations. A swich turns it on or off, the situation is a win in the first outcome or a win in the second outcome, etc.

It's true that key–value store shouldn't be written with a hyphen. It should be written with an en dash, which is used "to contrast values or illustrate a relationship between two things [... e.g.] Mother–daughter relationship"

https://en.wikipedia.org/wiki/Dash#En_dash

I just didn't want to bother with typography at that level of pedanticism.

Comment by VerifiedReports 5 hours ago

No, they don't. A master/slave configuration (of hard drives, for example) involves two things. I specifically included it to head off the exact objection you're raising.

"...the slash is now used to represent division and fractions, as a date separator, in between multiple alternative or related terms"

-Wikipedia

And what is a key/value store? A store of related terms.

And if you had a system that only allowed a finite collection of key values, where might you put them? A key-value store.

Comment by kqr 5 hours ago

The hard drives are either master or slave. A hard drive is not a master-and-slave.

Comment by abustamam 19 hours ago

Wikipedia seems to find "key-value store" an appropriate term.

https://en.wikipedia.org/wiki/Key%E2%80%93value_database

Comment by VerifiedReports 7 hours ago

See above.

Comment by DonHopkins 9 hours ago

Which is infinite of value is zero.

Comment by yupyupyups 22 hours ago

Depending on the underlying storage being reliable is far from unique to garage. This is what most other services do too, unless we're talking about something like Ceph which manages the physical storage itself.

Standard filesystems such as ext4 and xfs don't have data checksumming, so you'll have to rely on another layer to provide integrity. Regardless, that's not garage's job imo. It's good that they're keeping their design simple and focus their resources on implementing the S3 spec.

Comment by igor47 1 day ago

I've been using minio for local dev but that version is unmaintained now. However, I was put off by the minimum requirements for garage listed on the page -- does it really need a gig of RAM?

Comment by dsvf 1 day ago

I always understood this requirement as "garage will run fine on hardware with 1GB RAM total" - meaning the 1GB includes the RAM used by the OS and other processes. I think that most current consumer hardware that is a, potential garage host, even on the low end, has at least 1GB total RAM.

Comment by archon810 1 day ago

The current latest Minio release that is working for us for local development is now almost a year old and soon enough we will have to upgrade. Curious what others have replaced it with that is as easy to set up and has a management UI.

Comment by mbreese 1 day ago

I think that's part of the pitch here... swapping out Minio for Garage. Both scale a lot more than for just local development, but local dev certainly seems like a good use-case here.

Comment by lxpz 1 day ago

It does not, at least not for a small local dev server. I believe RAM usage should be around 50-100MB, increasing if you have many requests with large objects.

Comment by nijave 21 hours ago

The assumption is nodes are in different fault domains so it'd be highly unlikely to ruin the whole cluster.

LMDB mode also runs with flush/syncing disabled

Comment by moffkalast 1 day ago

That's not something you can do reliably in software, datacenter grade NVMe drives come with power loss protection and additional capacitors to handle that gracefully. If power is cut at the wrong moment the partition may not be mountable afterwards otherwise.

If you really live somewhere with frequent outages, buy an industrial drive that has a PLP rating. Or get a UPS, they tend to be cheaper.

Comment by crote 1 day ago

Isn't that the entire point of write-ahead logs, journaling file systems, and fsync in general? A roll-back or roll-forward due to a power loss causing a partial write is completely expected, but surely consumer SSDs wouldn't just completely ignore fsync and blatantly lie that the data has been persisted?

As I understood it, the capacitors on datacenter-grade drives are to give it more flexibility, as it allows the drive to issue a successful write response for cached data: the capacitor guarantees that even with a power loss the write will still finish, so for all intents and purposes it has been persisted, so an fsync can return without having to wait on the actual flash itself, which greatly increases performance. Have I just completely misunderstood this?

Comment by unsnap_biceps 1 day ago

you actually don't need capacitors for rotating media, Western Digital has a feature called "ArmorCache" that uses the rotational energy in the platters to power the drive long enough to sync the volatile cache to a non volatile storage.

https://documents.westerndigital.com/content/dam/doc-library...

Comment by toomuchtodo 1 day ago

Very cool, like the ram air turbine that deploys on aircraft in the event of a power loss.

Comment by patmorgan23 1 day ago

Good I love engineers

Comment by Aerolfos 1 day ago

> but surely consumer SSDs wouldn't just completely ignore fsync and blatantly lie that the data has been persisted?

That doesn't even help if fsync() doesn't do what developers expect: https://danluu.com/fsyncgate/

I think this was the blog post that had a bunch more stuff that can go wrong too: https://danluu.com/deconstruct-files/

But basically fsync itself (sometimes) has dubious behaviour, then OS on top of kernel handles it dubiously, and then even on top of that most databases can ignore fsync erroring (and lie that the data was written properly)

So... yes.

Comment by Nextgrid 1 day ago

> ignore fsync and blatantly lie that the data has been persisted

Unfortunately they do: https://news.ycombinator.com/item?id=38371307

Comment by btown 1 day ago

If the drives continue to have power, but the OS has crashed, will the drives persist the data once a certain amount of time has passed? Are datacenters set up to take advantage of this?

Comment by Nextgrid 1 day ago

> will the drives persist the data once a certain amount of time has passed

Yes, otherwise those drives wouldn't work at all and would have a 100% warranty return rate. The reason they get away with it is that the misbehavior is only a problem in a specific edge-case (forgetting data written shortly before a power loss).

Comment by unsnap_biceps 1 day ago

Yes, the drives are unaware of the OS state.

Comment by SomaticPirate 1 day ago

Seeing a ton of adoption of this after the Minio debacle

https://www.repoflow.io/blog/benchmarking-self-hosted-s3-com... was useful.

RustFS also looks interesting but for entirely non-technical reasons we had to exclude it.

Anyone have any advice for swapping this in for Minio?

Comment by dpedu 1 day ago

I have not tried either myself, but I wanted to mention that Versity S3 Gateway looks good too.

https://github.com/versity/versitygw

I am also curious how Ceph S3 gateway compares to all of these.

Comment by skrtskrt 1 day ago

When I was there, DigitalOcean was writing a complete replacement for the Ceph S3 gateway because its performance under high concurrency was awful.

They just completely swapped out the whole service from the stack and wrote one in Go because of how much better the concurrency management was, and Ceph's team and codebase C++ was too resistant to change.

Comment by jiqiren 1 day ago

Unrelated, but one of the more annoying aspects of whatever software they use now is lack of IPv6 for the CDN layer of DigitalOcean Spaces. It means I need to proxy requests myself. :(

Comment by zipzad 1 day ago

I'd be curious to know how versitygw compares to rclone serve S3.

Comment by chrislusf 22 hours ago

Disclaim: I work on SeaweedFS.

Why skipping SeaweedFS? It rank #1 on all benchmarks, and has a lot of features.

Comment by meotimdihia 21 hours ago

I confirm this, I used SeaweedFS to serve 1M users daily with 56 million images / ~100TB with 2 servers + HDD only, while Minio can't do this. Seaweedfs performance is much better than Minio's. The only problem is that SeaweedFS documentation is hard to understand.

Comment by magicalhippo 15 hours ago

SeaweedFS is also so optimized for small objects, it can't store larger objects (max 32GiB[1]).

Not a concern for many use-cases, just something to be aware of as it's not a universal solution.

[1]: https://github.com/seaweedfs/seaweedfs?tab=readme-ov-file#st...

Comment by chrislusf 3 hours ago

Not correct. The files are chunked into smaller pieces and spread to all volume servers.

Comment by magicalhippo 2 hours ago

Well, then I suggest updating the incorrect readme. It's why I've ignored SeaweedFS.

Comment by ted_dunning 15 hours ago

SeaweedFS is very nice and takes quite an effort to lose data.

Comment by dionian 22 hours ago

can you link benchmarks

Comment by chrislusf 3 hours ago

It is in the parent comment.

Comment by Implicated 1 day ago

> but for entirely non-technical reasons we had to exclude it

Able/willing to expand on this at all? Just curious.

Comment by misnome 14 hours ago

They seem to have gone all-in on AI, for commits and ticket management. Not interested in interacting with that.

Otherwise, the built in admin on one-executable was nice, and support for tiered storage, but single node parallel write performance was pretty unimpressive and started throwing strange errors (investigating of which led to the AI ticket discovery).

Comment by NitpickLawyer 1 day ago

Not the same person you asked, but my guess would be that it is seen as a chinese product.

Comment by lima 1 day ago

RustFS appears to be very early-stage with no real distributed systems architecture: https://github.com/rustfs/rustfs/pull/884

I'm not sure if it even has any sort of cluster consensus algorithm? I can't imagine it not eating committed writes in a multi-node deployment.

Garage and Ceph (well, radosgw) are the only open source S3-compatible object storage which have undergone serious durability/correctness testing. Anything else will most likely eat your data.

Comment by dewey 1 day ago

What is this based on, honest question as from the landing page I don't get that impression. Are many committers China-based?

Comment by NitpickLawyer 1 day ago

https://rustfs.com.cn/

> Beijing Address: Area C, North Territory, Zhongguancun Dongsheng Science Park, No. 66 Xixiaokou Road, Haidian District, Beijing

> Beijing ICP Registration No. 2024061305-1

Comment by dewey 1 day ago

Oh, I misread the initial comment and thought they had to exclude Garage. Thanks!

Comment by scottydelta 1 day ago

From what I have seen in the previous discussions here (since and before Minio debacle) and at work, Garage is a solid replacement.

Comment by 13 hours ago

Comment by klooney 1 day ago

Seaweed looks good in those benchmarks, I haven't heard much about it for a while.

Comment by thhck 1 day ago

BTW https://deuxfleurs.fr/ is one of the most beautiful website I have ever seen

Comment by codethief 1 day ago

It's beautiful from an artistic point of view but also rather hard to read and probably not very accessible (haven't checked it, though, since I'm on my phone).

Comment by isoprophlex 1 day ago

Works perfectly on an iphone. I can't attest to the accessibility features, but the aesthetic is absolutely wonderful. Something I love, and went for on my own portfolio/company website... this is executed 100x better tho, clearly a labor of love and not 30 minutes of shitting around in vi.

Comment by self_awareness 17 hours ago

Well it's ASCII-themed but it's completely unreadable in terminal links/lynx.

Comment by PunchyHamster 12 hours ago

For someone recently migrating from minio, caveats

* no lifecycle management of any kind - if you're using it for backups you can't set "don't delete versions for 3 months", so if anyone takes hold of your key, you backups are gone. I relied on minio's lifecycle management for that but it's feature missing in garage (and to be fair, most other) S3

* no automatic mirroring (if you want to have second copy in something other than garage or just don't want to have a cluster but rather have more independent nodes)

* ACLs for access are VERY limited - can't make a key access only sub-path, can't make a "master key" (AFAIK, couldn't find an option) that can access all the buckets so the previous point is also harder - I can't easily use rclone to mirror entire instance somewhere else unless I write scrip iterating over buckets and adding them bucket by bucket to key ACK

* Web hosting features are extremely limited so you won't be able to say set CORS headers for the bucket

* No ability to set keys - you can only generate on inside garage or import garage-formatted one - which means you can't just migrate storage itself, you have to re-generate every key. It also makes automating it harder, in case of minio you can pre-generate key and pass then fed it to clients and to the minio key command, here you have to do the dance of "generate with tool" -> "scrape and put into DB" -> put onto clients.

Overall I like the software a lot but if you have setup that uses those features, beware.

Comment by coldtea 12 hours ago

>no lifecycle management of any kind - if you're using it for backups you can't set "don't delete versions for 3 months", so if anyone takes hold of your key, you backups are gone

If someone gets a hold of your key, can't they also just change your backup deletion policy, even if it supported one?

Comment by PunchyHamster 10 hours ago

> If someone gets a hold of your key, can't they also just change your backup deletion policy, even if it supported one?

Minio have full on ACLs so you can just create a key that can only write/read but not change any settings like that.

So you just need to keep the "master key" that you use for setup away from potentially vulnerable devices, the "backup key" doesn't need those permissions.

Comment by topspin 1 day ago

No tags on objects.

Garage looks really nice: I've evaluated it with test code and benchmarks and it looks like a winner. Also, very straightforward deployment (self contained executable) and good docs.

But no tags on objects is a pretty big gap, and I had to shelve it. If Garage folk see this: please think on this. You obviously have the talent to make a killer application, but tags are table stakes in the "cloud" API world.

Comment by lxpz 1 day ago

Thank you for your feedback, we will take it into account.

Comment by topspin 1 day ago

Great, and thank you.

I really, really appreciate that Garage accommodates running as a single node without work-arounds and special configuration to yield some kind of degraded state. Despite the single minded focus on distributed operation you no doubt hear endlessly (as seen among some comments here,) there are, in fact, traditional use cases where someone will be attracted to Garage only for the API compatibility, and where they will achieve availability in production sufficient to their needs by means other than clustering.

Comment by VerifiedReports 19 hours ago

What are "tags on objects?"

Comment by topspin 18 hours ago

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

Arbitrary name+value pairs attached to S3 objects and buckets, and readily available via the S3 API. Metadata, basically. AWS has some tie-ins with permissions and other features, but tags can be used for any purpose. You might encode video multiple times at different bitrates, and store the rate in a tag on each object, for example. Tags are an affordance used by many applications for countless purposes.

Comment by VerifiedReports 6 hours ago

Thanks! I understand what tags are, but not what an "object" was in this context. Your example of multiple encodings of the same video seems very good.

Comment by ai-christianson 1 day ago

I love garage. I think it has applications beyond the standard self host s3 alternative.

It's a really cool system for hyper converged architecture where storage requests can pull data from the local machine and only hit the network when needed.

Comment by singpolyma3 1 day ago

I'd love to hear what configuration you are using for this

Comment by Powdering7082 1 day ago

No erasure coding seems like a pretty big loss in terms of how much resources do you need to get good resiliency & efficiency

Comment by munro 1 day ago

I was looking at using this on an LTO tape library, it seems the only resiliency is through replication, but this was my main concern with this project, what happens with HW goes bad

Comment by lxpz 1 day ago

If you have replication, you can lose one of the replica, that's the point. This is what Garage was designed for, and it works.

Erasure coding is another debate, for now we have chosen not to implement it, but I would personally be open to have it supported by Garage if someone codes it up.

Comment by hathawsh 1 day ago

Erasure coding is an interesting topic for me. I've run some calculations on the theoretical longevity of digital storage. If you assume that today's technology is close to what we'll be using for a long time, then cross-device erasure coding wins, statistically. However, if you factor in the current exponential rate of technological development, simply making lots of copies and hoping for price reductions over the next few years turns out to be a winning strategy, as long as you don't have vendor lock-in. In other words, I think you're making great choices.

Comment by Dylan16807 1 day ago

I question that math. Erasure coding needs less than half as much space as replication, and imposes pretty small costs itself. Maybe we can say the difference is irrelevant if storage prices will drop 4x over the next five years? But looking at pricing trends right now... that's not likely. Hard drives and SSDs are about the same price they were 5 years ago. The 5 years before that SSDs were seeing good advancements, but hard drive prices only advanced 2x.

Comment by faizshah 1 day ago

One really useful usecase for Garage for me has been data engineering scripts. I can just use the S3 integration that every tool has to dump to garage and then I can more easily scale up to cloud later.

Comment by supernes 1 day ago

I tried it recently. Uploaded around 300 documents (1GB) and then went to delete them. Maybe my client was buggy, because the S3 service inside the container crashed and couldn't recover - I had to restart it. It's a really cool project, but I wouldn't really call it "reliable" from my experience.

Comment by JonChesterfield 1 day ago

Corrupts data on power loss according to their own docs. Like what you get outside of data centers. Not reliable then.

Comment by lxpz 1 day ago

Losing a node is a regular occurrence, and a scenario for which Garage has been designed.

The assumption Garage makes, which is well-documented, is that of 3 replica nodes, only 1 will be in a crash-like situation at any time. With 1 crashed node, the cluster is still fully functional. With 2 crashed nodes, the cluster is unavailable until at least one additional node is recovered, but no data is lost.

In other words, Garage makes a very precise promise to its users, which is fully respected. Database corruption upon power loss enters in the definition of a "crash state", similarly to a node just being offline due to an internet connection loss. We recommend making metadata snapshots so that recovery of a crashed node is faster and simpler, but it's not required per se: Garage can always start over from an empty database and recover data from the remaining copies in the cluster.

To talk more about concrete scenarios: if you have 3 replicas in 3 different physical locations, the assumption of at-most one crashed node is pretty reasonable, it's quite unlikely that 2 of the 3 locations will be offline at the same time. Concerning data corruption on a power loss, the probability to lose power at 3 distant sites at the exact same time with the same data in the write buffers is extremely low, so I'd say in practice it's not a problem.

Of course, this all implies a Garage cluster running with 3-way replication, which everyone should do.

Comment by JonChesterfield 1 day ago

That is a much stronger guarantee than your documentation currently claims. One site falling over and being rebuilt without loss is great. One site losing power, corrupting the local state, then propagating that corruption to the rest of the cluster would not be fine. Different behaviours.

Comment by lxpz 1 day ago

Fair enough, we will work on making the documentation clearer.

Comment by jiggawatts 1 day ago

So if you put a 3-way cluster in the same building and they lose power together, then what? Is your data toast?

Comment by lxpz 1 day ago

If I make certain assumptions and you respect them, I will give you certain guarantees. If you don't respect them, I won't guarantee anything. I won't guarantee that your data will be toast either.

Comment by Dylan16807 1 day ago

If you can't guarantee anything for all the nodes losing power at the same time, that's really bad.

If it's just the write buffer at risk, that's fine. But the chance of overlapping power loss across multiple sites isn't low enough to risk all the existing data.

Comment by InitialBP 1 day ago

It sounds like that's a possibility, but why on earth would you take the time to setup a 3 node cluster of object storage for reliability and ignore one of the key tenants of what makes it reliable?

Comment by eduardogarza 1 day ago

I use this for booting up S3-compatible buckets for local development and testing -- paired up with s5cmd, I can seed 15GB and over 60,000 items (seed/mock data) in < 60s... have a perfect replica of a staging environment with Docker containers (api, db, cache, objects) all up in less than 2mins. Super simple to set up for my case and been working great.

Previously I used LocalStack S3 but ultimately didn't like the lack of persistance thats not available on the OSS verison. MinIO OSS is apparently no longer maintained? Also looked at SeaweedFS and RustFS but from a quick reading into them this once was the easiest to set up.

Comment by chrislusf 22 hours ago

I work on SeaweedFS. So very biased. :)

Just run "weed sever -s3 -dir=..." to have an object store.

Comment by eduardogarza 20 hours ago

I'll try it!

Comment by awoimbee 1 day ago

How is garage for a simple local dev env ? I recently used seaweedfs since they have a super simple minimal setup compared to garage which seemed to require a config file just to get started.

Comment by ianopolous 13 hours ago

@lxpz It would be great to do a follow up to this blog post with the latest Peergos. All the issues with baseline bandwidth and requests have gone away, even with federation on. The baseline is now 0, and even many locally initiated requests will be served directly from a Peergos cache without touching S3.

https://garagehq.deuxfleurs.fr/blog/2022-ipfs/

Let's talk!

Comment by 1 day ago

Comment by tenacious_tuna 19 hours ago

Anyone know if it's possible to bandwidth-limit the sync operations? I'd love to set up garage instances across my families' houses to act as a distributed backup, but I don't want to hose their (or my) down/uplink during awake hours. Having redundant selfhosted S3like storage would solve many problems for me, but I really need that capability.

Comment by apawloski 1 day ago

Is it the same consistency model as S3? I couldn't see anything about it in their docs.

Comment by lxpz 1 day ago

Read-after-write consistency : yes (after PutObject has finished, the object will be immediately visible in all subsequent requests, including GetObject and ListObjects)

Conditionnal writes : no, we can't do it with CRDTs, which are the core of Garage's design.

Comment by skrtskrt 1 day ago

Does RAMP or CURE offer any possibility of conditional writes with CRDTs? I have had these papers on my list to read for months, specifically wondering if it could be applied to Garage

https://dd.thekkedam.org/assets/documents/publications/Repor... http://www.bailis.org/papers/ramp-sigmod2014.pdf

Comment by lxpz 1 day ago

I had a very rapid look at these two papers, it looks like none of them allow the implementation of compare-and-swap, which is required for if-match / if-none-match support. They have a weaker definition of a "transaction". Which is to be expected as they only implement causal consistency at best and not consensus, whereas consensus is required for compare-and-swap.

Comment by skrtskrt 1 day ago

ack - makes sense, thank you for looking!

Comment by wyattjoh 1 day ago

Wasn't expecting to see it hosted on forgejo. Kind of a breath of fresh air to be honest.

Comment by agwa 1 day ago

Does this support conditional PUT (If-Match / If-None-Match)?

Comment by codethief 1 day ago

https://news.ycombinator.com/item?id=46328218

Comment by 1 day ago

Comment by k__ 23 hours ago

Half-OT:

Does anyone know a good open source S3 alternarive that's easily extendable with custom storage backends?

For example, AWS offers IA and Glacier in addition to the defaults.

Comment by onionjake 23 hours ago

Storj supports arbitrary configured backends each with different erasure coding, node placement, etc.

Comment by yupyupyups 21 hours ago

Garage is amazing! But it would be even more amazing if it had immutable object support. :)

This is used for ransomware resistant backups.

Comment by allanrbo 1 day ago

I use Syncthing a lot. Is Garage only really useful if you specifically want to expose an S3 drop in compatible API, or does it also provide other benefits over syncthing?

Comment by lxpz 1 day ago

They are not solving the same problem.

Syncthing will synchronize a full folder between an arbitrary number of machines, but you still have to access this folder one way or another.

Garage provides an HTTP API for your data, and handles internally the placement of this data among a set of possible replica nodes. But the data is not in the form of files on disk like the ones you upload to the API.

Syncthing is good for, e.g., synchronizing your documents or music collection between computers. Garage is good as a storage service for back-ups with e.g. Restic, for media files stored by a web application, for serving personal (static) web sites to the Internet. Of course, you can always run something like Nextcloud in front of Garage and get folder synchronization between computers somewhat like what you would get with Syncthing.

But to answer your question, yes, Garage only provides a S3-compatible API specifically.

Comment by 1 day ago

Comment by sippeangelo 1 day ago

You use Syncthing for object storage?

Comment by Eikon 1 day ago

Unfortunately, this doesn’t support conditional writes through if-match and if-none-match [0] and thus is not compatible with ZeroFS [1].

[0] https://git.deuxfleurs.fr/Deuxfleurs/garage/issues/1052

[1] https://github.com/Barre/ZeroFS

Comment by chrislusf 22 hours ago

I work on SeaweedFS. It has support for these if conditions, and a lot more.

Comment by ekjhgkejhgk 1 day ago

Anybody understand how this compares with Vast?