created_at2022-02-26 15:57:13.666042
updated_at2023-04-26 03:13:28.283723
descriptionSome adaptors to stream rkyv Archives over AsyncRead and AsyncWrite
Zyansheep (zyansheep)




docs.rs crates.io

Simple async codec for rkyv. Reuses streaming buffer for maximum speed!

This crate provides a makeshift adaptor for streaming &Archived<Object>s from an AsyncRead using a reusable external buffer, as well as a futures::Sink implementation to serialize Objects to an AsyncWrite. It uses multiformat's unsigned_varint for variable-length length encoding by default, but allows for other kinds of length encoding through the LengthEncoding trait. It also supports directly using bytes::BytesMut and #[no_std] when default features are disabled.


This crate has three examples: chat_client, chat_server & no-std. Run the first two at the same time to see a proof-of-concept Archive tcp echo server in action.

To run:

cargo run --example chat_client

cargo run --example chat_server

Simple usage example:

#[derive(Archive, Deserialize, Serialize, Debug, PartialEq, Clone)]
#[archive(check_bytes)] // check_bytes is required
struct Test {
    int: u8,
    string: String,
    option: Option<Vec<i32>>,
let value = Test {
    int: 42,
    string: "hello world".to_string(),
    option: Some(vec![1, 2, 3, 4]),
// Writing
let writer = Vec::new();
let mut codec = RkyvWriter::<_, VarintLength>::new(writer);
// Reading
let mut reader = &codec.inner()[..];
let mut buffer = AlignedVec::new(); // Aligned streaming buffer for re-use
let data: &Archived<Test> = archive_stream::<_, Test, VarintLength>(&mut reader, &mut buffer).await.unwrap(); // This returns a reference into the passed buffer
let value_received: Test = data.deserialize(&mut Infallible).unwrap();
assert_eq!(value, value_received);

See examples/no-std for an example with no-std support.


These are a set of benchmarks, each benchmark represents 50 test objects being either sent or received. (requires nightly)

test tests::bench_archive_sink_50              ... bench:      10,388 ns/iter (+/- 1,603)
test tests::bench_archive_sink_prearchived_50  ... bench:       2,032 ns/iter (+/- 302)
test tests::bench_archive_stream_50            ... bench:       3,544 ns/iter (+/- 438)
test tests::bench_futures_cbor_sink_50         ... bench:      14,105 ns/iter (+/- 1,160)
test tests::bench_futures_cbor_stream_50       ... bench:       9,632 ns/iter (+/- 1,437)
test tests::bench_rkyv_futures_codec_sink_50   ... bench:       6,671 ns/iter (+/- 521)
test tests::bench_rkyv_futures_codec_stream_50 ... bench:       4,925 ns/iter (+/- 948)
test tests::bench_rkyv_writer_50               ... bench:       3,547 ns/iter (+/- 271)
test tests::bench_u64_length_encoding          ... bench:       4,226 ns/iter (+/- 327)
test tests::bench_varint_length_encoding       ... bench:       4,243 ns/iter (+/- 343)

The fastest real benchmark (full serialization and bytecheck) is using RkyvWriter for writing and archive_stream for reading. This is compared to the slowest benchmark: futures_codec library's CborCodec.

Feel free to contribute your own benchmarks!

Commit count: 49

cargo fmt