A wait-free single-producer single-consumer (SPSC) ring buffer for Rust.
This crate can be used without the standard library (#![no_std]
)
by disabling the std
feature (which is enabled by default),
but the alloc crate is needed nevertheless.
Add this to your Cargo.toml
:
[dependencies]
rtrb = "0.3"
For a list of breaking changes and for instructions how to upgrade between released versions, have a look at the changelog.
Measuring the performance of a data structure for inter-thread communication can be quite brittle and the results depend on many factors. A few performance comparisons between competing crates are shown in issue #39, but like all benchmarks, they are deeply flawed and to be taken with a grain of salt. You should make your own measurements that are relevant to your usage patterns. Feel free to share your own results by commenting on that issue.
Creating the HTML docs (which will be available in target/doc/rtrb/index.html
):
cargo doc
Running the tests:
cargo test
Testing the benchmarks (without actually benchmarking):
cargo test --benches
Running the benchmarks (using the criterion crate;
results will be available in target/criterion/report/index.html
):
cargo bench
Creating flame graphs for the benchmarks; first a few preparations:
cargo install flamegraph
echo -1 | sudo tee /proc/sys/kernel/perf_event_paranoid
export CARGO_PROFILE_BENCH_DEBUG=true
Then, creating the flame graph (which will be saved to flamegraph.svg
),
providing a benchmark (e.g. two_threads
), a desired runtime and optionally
a benchmark function (e.g. large
):
cargo flamegraph --bench two_threads -- --bench --profile-time 10 large
To measure code coverage, nightly Rust is required, as well as a few additional dependencies:
rustup toolchain install nightly
rustup component add llvm-tools-preview
cargo install grcov
Test coverage data can be obtained and analyzed with these commands:
cargo clean
RUSTFLAGS="-Z instrument-coverage" RUSTDOCFLAGS="-Z instrument-coverage -Z unstable-options --persist-doctests target/debug/doctestbins" LLVM_PROFILE_FILE="coverage/%p-%m.profraw" cargo +nightly test
grcov coverage --source-dir . --binary-path target/debug --output-type html --output-path coverage
The last command creates an HTML report in coverage/index.html
.
Testing with Miri also needs nightly Rust:
cargo +nightly miri test
This Miri flag should also be tried:
MIRIFLAGS="-Zmiri-preemption-rate=0" cargo +nightly miri test
Running the tests with ThreadSanitizer requires nightly Rust as well:
RUSTFLAGS="-Z sanitizer=thread" cargo +nightly test --tests -Z build-std --target x86_64-unknown-linux-gnu
You might have to adapt the --target
option to your system (see e.g. rustup show
).
rustc
VersionThis crate's minimum supported rustc
version (MSRV) is 1.38.0
.
The MSRV is not expected to be updated frequently, but if it is,
there will be (at least) a minor version bump.
The initial code has been ripped off of https://github.com/crossbeam-rs/crossbeam/pull/338, with permission of the PR author.
It has been isolated from the rest of crossbeam
with git-filter-repo:
git-filter-repo --subdirectory-filter crossbeam-queue --path src/spsc.rs --path tests/spsc.rs --refs refs/heads/spsc
If you don't like this crate, no problem, there are several alternatives for you to choose from. There are many varieties of ring buffers available, here we limit the selection to wait-free SPSC implementations:
heapless::spsc
)jack::Ringbuffer
)magnetic::spsc
module)npnc::bounded::spsc
module)shmem_ipc::sharedring
and shmem_ipc::ringbuf
modules)There are also implementations in other languages:
SharedArrayBuffer
)If you know more alternatives for this list, please open an issue.
Licensed under either of
at your option.
Note that this crate contains a copy of the file cache_padded.rs
from
https://github.com/crossbeam-rs/crossbeam.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.