ZeroPool v0.3.1

What's Changed

🐛 Bug Fixes

• Platform Compatibility: Fixed tokio-uring feature to only compile on Linux systems, resolving CI failures on macOS and Windows when using --all-features

🔧 CI/CD Improvements

• Linting: Updated Clippy configuration to allow unsafe code warnings while maintaining strict safety standards
• Documentation: Fixed HTML tag issues in doc comments for better documentation generation

📚 Documentation

• README: Enhanced clarity on performance optimizations, security, and use cases
• API Docs: Improved formatting and consistency in documentation comments

🏗️ Internal Improvements

• Code Quality: Simplified vector initialization logic for better maintainability
• Feature Gates: Properly gated tokio-uring functionality by both feature flag and target OS

🚀 Features

• tokio-uring Support: Added optional zero-copy I/O support for Linux systems (requires --features tokio-uring)

This release maintains ZeroPool's focus on maximum performance while improving cross-platform compatibility and code quality.

ZeroPool v0.3.0

What's New in v0.3.0

Breaking Changes

• BREAKING: BufferPool::get() now returns PooledBuffer instead of Vec<u8>
  • PooledBuffer implements Deref<Target = Vec<u8>> for transparent access
  • Buffers automatically return to pool when dropped (RAII pattern)
  • No more manual pool.put() calls needed
• BREAKING: BufferPool::put() is now internal-only (pub(crate))

New Features

• New PooledBuffer type with automatic buffer return via Drop trait
• Configurable eviction policies (LIFO, ClockPro) for fine-tuned performance
• Profiling binaries for performance analysis:
  • ml_checkpoint_loader: ML model loading simulation
  • network_server: Multi-threaded network server simulation
  • file_pipeline: File processing pipeline simulation
  • stress_test: High-stress workload testing
• Modularized codebase with separate modules

Improvements

• Removed all unsafe code blocks for improved safety
• Fixed all clippy warnings and formatting issues
• Improved code organization and documentation

Migration Guide

Before (v0.2.x):

let buf = pool.get(1024);
// use buf
pool.put(buf); // Manual return

After (v0.3.0):

let buf = pool.get(1024);
// use buf
// Auto-returns on drop!

// For explicit early return:
drop(buf);

See CHANGELOG.md for full details.

Release v0.2.1

Release v0.2.1

Major Changes

• Modularized codebase: Split the monolithic `lib.rs` into focused modules (`config.rs`, `pool.rs`, `tls.rs`, `utils.rs`) for better maintainability and code organization.

New Features

• Configurable eviction policies: Added support for CLOCK-Pro (default) and LIFO buffer eviction strategies to optimize cache behavior for different workloads.
  • CLOCK-Pro: Intelligent eviction using access counters for better cache locality (~8 bytes overhead per buffer)
  • LIFO: Simple last-in-first-out for minimal overhead and best uniform buffer size performance

Improvements

• Expanded benchmarking suite with new performance tests for allocation patterns, cache behavior, and memory features
• Added profiling support with flamegraph and perf data artifacts
• Updated documentation with eviction policy usage examples

Technical Details

• Improved code idiomaticity and thread-local shard affinity
• Fixed clippy warnings and code formatting
• Enhanced performance testing capabilities

v0.2.0 - Thread-Local Shard Affinity

Release v0.2.0 - Thread-Local Shard Affinity

Major performance improvements through architectural changes:

Performance Gains:
- Multi-threaded throughput: 32 TiB/s at 8 threads (14-38% improvement)
- 2 threads: 14.2 TiB/s (38% faster than v0.1.5)
- 4 threads: 25.0 TiB/s (34% faster than v0.1.5)
- 8 threads: 32.0 TiB/s (14% faster than v0.1.5)
- 8.2x faster than sharded-slab in multi-threaded scenarios

Architecture Changes:
- Thread-local shard affinity replacing round-robin selection
- Each thread consistently uses same shard for better cache locality
- Reduced CPU cache line bouncing between cores
- Hash-based thread-to-shard assignment for balanced distribution

Code Quality Improvements:
- Added debug assertions to all unsafe code blocks
- Removed unnecessary const qualifiers from builder methods
- Simplified conditionals using idiomatic Rust patterns (match, let-else)
- Optimized TLS cache initialization
- Better inline attributes for compiler optimization
- Builder validation with documented panic conditions

Breaking Changes:
- None - fully backward compatible API

Benchmarks run on Intel i9-10900K @ 3.7GHz, Linux 6.14.0

v0.1.5

Fixes

• Fix clippy warning: remove unnecessary cast in pin_buffer function
• Format code with cargo fmt for consistency

Full Changelog: v0.1.4...v0.1.5

v0.1.4

Performance Improvements

• O(1) len() queries: Added atomic counters for instant pool size queries without locking
• Optimized memory pinning: Redesigned pinning strategy for better performance
• Unchecked shard indexing: Eliminated redundant bounds checks using bitmask guarantees
• Improved preallocate(): Batch allocation and distribution with minimal locking

Enhancements

• Add pinned memory benchmarks
• Enhanced safety documentation with detailed explanations of unsafe usage
• Improved Builder API for pinned_memory configuration

Full Changelog: v0.1.3...v0.1.4

v0.1.3

Changes

• Add workflow_dispatch trigger to publish action for manual releases
• Add dry-run option to publish workflow
• Style improvements: remove unnecessary blank line in lib.rs
• Improve test assertion clarity

Full Changelog: https://github.com/ZeroPool/zeropool/compare/v0.1.2...v0.1.3

v0.1.2 - Idiomatic Builder Pattern & Memory Pinning

What's New

Idiomatic Rust Builder Pattern

• Added BufferPool::builder following tokio/rayon style
• Removed non-idiomatic preset methods
• Made PoolConfig private (internal implementation)

Memory Pinning Support

• Added pinned_memory builder method
• Always available (no feature flags needed)
• Pre-allocates and pins buffers for optimal performance
• Graceful fallback if pinning fails

Improved Test Coverage

• 12 comprehensive tests (was 5)
• New tests: builder API, pinned memory, concurrent access, edge cases
• All tests passing, clippy clean

Rust Best Practices

• const fn for builder methods
• Proper must_use annotations
• Modern format strings
• Optimized allocations

API Changes

New:

• BufferPool::builder
• Builder::min_buffer_size
• Builder::num_shards
• Builder::tls_cache_size
• Builder::max_buffers_per_shard
• Builder::pinned_memory

Removed:

• PoolConfig public API (now private)
• PoolConfig preset methods

Migration Guide

Before:
let pool = BufferPool::with_config(PoolConfig::default().with_min_buffer_size(512 * 1024));

After:
let pool = BufferPool::builder().min_buffer_size(512 * 1024).build();