cykit is a collection Cython utilities.
cykit requires python 3.9 or greater. It is available on pypi. To install, run:
pip install cykitcyloggeris the initial component of thecykitcollection.
It is a thin wrapper around spdlog.
cylogger is designed to provide a unified logging interface across the entire stack — Python, underlying Cython layers,
and native C/C++ modules can all share the same logger instance and logging pipeline. This makes it easier to maintain consistent
formatting, sinks (console, file, rotating, daily, UDP, TCP, HTTP, SMTP), log levels, and tracing across mixed-language systems.
Detailed examples can be found here: cykit/examples/cylogger
cykit.ipc provides a high-performance shared-memory IPC mechanism built around a ring-buffer design.
It supports multiple producer/consumer topologies while exposing both a native Cython API and a Python-friendly API.
- Shared-memory IPC for inter-process communication.
- Ring-buffer based message transport.
- Supports SPSC, SPMC, MPSC and MPMC modes.
- Native Cython API with an additional Python API.
- Blocking and non-blocking operations.
- Fixed-size and variable-size messages.
- Cross-platform (Linux, Windows and macOS).
Detailed examples can be found here: cykit/examples/ipc
Benchmark configuration:
- 1 Producer / 1 Consumer (separate processes)
- 64-byte fixed-size messages
- 2,000,000 messages
- Ubuntu Linux
- Python 3.12
Results are measured on the author's development machine and are intended for relative comparison between implementations.
| Implementation | msg/sec | MiB/sec | us/msg |
|---|---|---|---|
| cykit.ipc (Cython-Cython) | 12,641,965 | 771.60 | 0.079 |
| cykit.ipc (Python API) | 3,672,439 | 231.15 | 0.272 |
| multiprocessing.Pipe | 690,622 | 42.15 | 1.448 |
| multiprocessing.SimpleQueue | 234,203 | 14.29 | 4.270 |
| multiprocessing.shared_memory (ring) | 232,724 | 14.20 | 4.297 |
| multiprocessing.Queue | 172,548 | 10.53 | 5.796 |
cykit.queue is a in-process Lock-free message queue based on ring buffer. It supports SPSC, SPMC, MPSC, and MPMC communication patterns while preserving broadcast (fan-out) semantics.
- Broadcast (fan-out) semantics
- Supports SPSC, SPMC, MPSC, and MPMC
- Blocking and non-blocking operations
- Fixed-size message storage
Detailed examples can be found here: cykit/examples/queue
The following benchmark measures throughput using 64-byte fixed-size messages with 2,000,000 messages transferred between one producer and one consumer (SPSC push/pop).
Note: Python's standard library queues (
queue.Queue,queue.SimpleQueue, and typicalcollections.dequeimplementations) provide work-queue semantics, where each message is consumed by only one consumer.cykit.queueprovides broadcast (fan-out) semantics, allowing every consumer to receive every published message. To provide the closest comparable scenario, the benchmark below uses the SPSC configuration.
| Implementation | Throughput | Bandwidth | Avg. Latency |
|---|---|---|---|
| cykit.queue (Cython API) | 16.91 M msg/s | 1032.02 MiB/s | 0.059 μs |
queue.SimpleQueue |
11.64 M msg/s | 710.17 MiB/s | 0.086 μs |
| cykit.queue (Python API) | 11.06 M msg/s | 675.29 MiB/s | 0.090 μs |
collections.deque |
0.38 M msg/s | 23.28 MiB/s | 2.622 μs |
queue.Queue |
0.32 M msg/s | 19.71 MiB/s | 3.097 μs |
- Multi-mode message dispatcher to bridge Cython and Python (Both synchronous and asynchronous), built on a lock-free SPSC queue. (detailed documentation coming in a later update)
- TCP, UDP, HTTP Synchronous Clients.
- SMTP client with OAuth2 (XOAUTH2) support.
- Boost.Asio based cross platform signal handler. It register contexts to be notified on SIGINT/SIGTERM, with automatic cleanup and Python KeyboardInterrupt injection. (detailed documentation to follow)
Contributions are welcome! Any kind of help — bug reports / suggestions, feature requests, or pull requests—is appreciated.
This project is licensed under the MIT License and/or Apache License 2.0. See the license files for details.
This project vendors header‑only dependencies (Boost, spdlog, fmtlib) to simplify builds and ensure compatibility across platforms. All vendored code retains its original copyright and license terms.
All third‑party code retains its original copyright and license notices.
The full license texts are included in the NOTICE.md file at the root of this repository.