Show HN: Apache Fory Rust – 10-20x faster serialization than JSON/Protobuf (fory.apache.org)

67 points by chaokunyang ↗ HN
Serialization framework with some interesting numbers: 10-20x faster on nested objects than json/protobuf.

  Technical approach: compile-time codegen (no reflection), compact binary protocol with meta-packing, little-endian layout optimized for modern CPUs.

  Unique features that other fast serializers don't have:
  - Cross-language without IDL files (Rust ↔ Python/Java/Go)
  - Trait object serialization (Box<dyn Trait>)
  - Automatic circular reference handling
  - Schema evolution without coordination

  Happy to discuss design trade-offs.

  Benchmarks: https://fory.apache.org/docs/benchmarks/rust

33 comments

[ 2.7 ms ] story [ 57.3 ms ] thread
What's the story for JS. I see that there is a javascript directory, but it only mentions nodejs. I don't see an npm package. So does this work in web browsers?
JS support is still experimental, I have not publish it to npm
How does this deal with numeric types like NaN, Infinity...?
I'm wondering how do you share you shared types between languages if there's no schema ?
Why this over serialization free formats like CapnProto and Flatbuffers? If you want it to be compact, send it through zstd (with a custom dictionary).

I do really like that is broad support out of the box and looks easy to use.

For Python I still prefer using dill since it handles code objects.

https://github.com/uqfoundation/dill

Apache Fory is also a drop-in replacement for pickle/cloudpickle, you can use it to serialize code object such as local function/Classes too.

https://github.com/apache/fory/tree/main/python#serialize-lo...

When serializing code objects, pyfory is 3× higher compression ratio compared cloudpickle

And pyfory also provide extra security audit capability to avoid maliciously deserialization data attack.

Regarding design tradeoffs: I am very skeptical that this can be made to work for the long run in a cross-language way without formalizing the on-the-wire contract via IDL or similar.

In my experience, while starting from a language to arrive at the serialization often feels more ergonomic (e.g. RPC style) in the start, it hides too much of what's going on from the users and over time suffers greatly from programming language / runtime changes - the latter multiplied by the number of languages or frameworks supported.

That’s a fair point — with more languages in the mix, having a formal schema can definitely help prevent drift.

The way I think about it is: • Single‑language projects often work best without an IDL — it keeps things simple and avoids extra steps. • Two languages – both IDL and no‑IDL approaches can work, depending on the team’s habits. • Three or more – an IDL can be really useful as a single source of truth and to avoid manually writing struct definitions in every language.

For Apache Fory, my plan is to add optional IDL support, so teams who want that “single truth” can generate definitions automatically, and others can continue with language‑first development. My hope is to give people flexibility to choose what fits their situation best.

even in single language, how do you handle schema evolution (detect breaking changes at compile time) without IDL?
Would love to see how it compares to Flatbuffers - was surprised to not see it in the benchmarks!
Curious about comparisons with Apache Arrow, which uses flatbuffers to avoid memory copying during deserialization, which is well supported by the Pandas ecosystem, and which allows users to serialize arrays as lists of numbers that have hardware support from a GPU (int8-64, float)
Apache Arrow is more of a memory format than a general‑purpose data serialization system. It’s great for in‑memory analytics and GPU‑friendly columnar storage.

Apache Fory, on the other hand, has its own wire‑stream format designed for sending data across processes or networks. Most of the code is focused on efficiently converting in‑memory objects into that stream format (and back) — with features like cross‑language support, circular reference handling, and schema evolution.

Fory also has a row format, which is a memory format, and can complement or compete with Arrow’s columnar format depending on the use case.

These binary protocols generally also try to keep the data size small. Protobuf is essentially compressing its integers (varint or zigzag encoding), for example.

It'd be helpful to see a plot of serialization costs vs data size. If you only display serialization TPS, you're always going to lose to the "do nothing" option of just writing your C structs directly to the wire, which is essentially zero cost.

Fory also compress integers using varint or zigzag encoding. The size are basically same:

| data type | data size | fory | protobuf |

| --------------- | --------- | ------- | -------- |

| simple-struct | small | 21 | 19 |

| simple-struct | medium | 70 | 66 |

| simple-struct | large | 220 | 216 |

| simple-list | small | 36 | 16 |

| simple-list | medium | 802 | 543 |

| simple-list | large | 14512 | 12876 |

| simple-map | small | 33 | 36 |

| simple-map | medium | 795 | 1182 |

| simple-map | large | 17893 | 21746 |

| person | small | 122 | 118 |

| person | medium | 873 | 948 |

| person | large | 7531 | 7865 |

| company | small | 191 | 182 |

| company | medium | 9118 | 9950 |

| company | large | 748105 | 782485 |

| e-commerce-data | small | 750 | 737 |

| e-commerce-data | medium | 53275 | 58025 |

| e-commerce-data | large | 1079358 | 1166878 |

| system-data | small | 311 | 315 |

| system-data | medium | 24301 | 26161 |

| system-data | large | 450031 | 479988 |

I wish we would focus on making tooling better for W3C EXI (Binary XML encoding) instead of inventing new formats. Just being fast isn't enough, I don't see many using Aeron/SBT, it need a ecosystem - which XML does have.
Binary XML encoding (like W3C EXI) is useful in some contexts, but it’s generally not as efficient as modern binary serialization formats. It also can’t naturally express shared or circular reference semantics, which are important for complex object graphs.

Fory’s format was designed from the ground up to handle those cases efficiently, while still enabling cross‑language compatibility and schema evolution.

Are the benchmarks actually fair? See:

https://github.com/apache/fory/blob/fd1d53bd0fbbc5e0ce6d53ef...

It seems if the serialization object is not a "Fory" struct, then it is forced to go through to/from conversion as part of the measured serialization work:

https://github.com/apache/fory/blob/fd1d53bd0fbbc5e0ce6d53ef...

The to/from type of work includes cloning Strings:

https://github.com/apache/fory/blob/fd1d53bd0fbbc5e0ce6d53ef...

reallocating growing arrays with collect:

https://github.com/apache/fory/blob/fd1d53bd0fbbc5e0ce6d53ef...

I'd think that the to/from Fory types is shouldn't be part of the tests.

Also, when used in an actual system tonic would be providing a 8KB buffer to write into, not just a Vec::default() that may need to be resized multiple times:

https://github.com/hyperium/tonic/blob/147c94cd661c0015af2e5...

IMO, not a fair benchmark.

I can see the source of an 10x improvement on an Intel(R) Xeon(R) Gold 6136 CPU @ 3.00GHz, but it drops to 3x improvement when I remove the to/from that clones or collects Vecs, and always allocate an 8K Vec instead of a ::Default for the writable buffer.

If anything, the benches should be updated in a tower service / codec generics style where other formats like protobuf do not use any Fory-related code at all.

Note also that Fory has some writer pool that is utilized during the tests:

https://github.com/apache/fory/blob/fd1d53bd0fbbc5e0ce6d53ef...

Original bench selection for Fory:

    Benchmarking ecommerce_data/fory_serialize/medium: Collecting 100 samples in estimated 5.0494 s (197k it
    ecommerce_data/fory_serialize/medium
                            time:   [25.373 µs 25.605 µs 25.916 µs]
                            change: [-2.0973% -0.9263% +0.2852%] (p = 0.15 > 0.05)
                            No change in performance detected.
    Found 4 outliers among 100 measurements (4.00%)
      2 (2.00%) high mild
      2 (2.00%) high severe
Compared to original bench for Protobuf/Prost:

    Benchmarking ecommerce_data/protobuf_serialize/medium: Collecting 100 samples in estimated 5.0419 s (20k
    ecommerce_data/protobuf_serialize/medium
                            time:   [248.85 µs 251.04 µs 253.86 µs]
    Found 18 outliers among 100 measurements (18.00%)
      8 (8.00%) high mild
      10 (10.00%) high severe
However after allocating 8K instead of ::Default and removing to/from it for an updated protobuf bench:

    fair_ecommerce_data/protobuf_serialize/medium
                            time:   [73.114 µs 73.885 µs 74.911 µs]
                            change: [-1.8410% -0.6702% +0.5190%] (p = 0.30 > 0.05)
                            No change in performance detected.
    Found 14 outliers among 100 measurements (14.00%)
      2 (2.00%) high mild
      12 (12.00%) high severe
Still mad they had to change the name. "Fury" was a really fitting name for fast serialization framework, "fory" is just bogus. Should've renamed it to "foray" or something.
I liked the name “Fury” too — I actually named it myself and was really fond of it, but unfortunately we had to change it.
Probably not for everyone. The current limit of 4096 types could be expanded if there’s a real need — it’s not a hard technical barrier.

I’m curious though: what’s an example scenario you’ve seen that requires so many distinct types? I haven’t personally come across a case with 4,096+ protocol messages defined.

The prevalence of AI slop in the landing page doc does not inspire confidence.
The https://github.com/apache/fory/blob/main/AGENTS.md is a very detailed document only for AI coding, but an excelent reference for development. But you are right, it may introduce concerns, let me remove it from landing pages
That’s not what I meant, I mean it is obvious from many phrases on the landing page that they were written with AI
Is Google guava really needed? I would like it to be taken out.
> endian flag: 1 when data is encoded by little endian, 0 for big endian.

Have we learned nothing? Endian swap on platforms that need it is faster than conditionals, and simpler.

Looks promising but the level of AI in the repo is a real turn-off.