Lock-free log structures in Rust represent a crucial advancement in high-performance system design. These techniques eliminate traditional mutex-based synchronization, reducing contention and improving throughput in concurrent systems.
Atomic Append Operations form the foundation of lock-free logging. They ensure thread-safe writes without blocking. The AtomicLog implementation uses atomic pointers and counters to manage concurrent access:
use std::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
struct AtomicLog {
buffer: Vec<AtomicPtr<Entry>>,
head: AtomicUsize,
capacity: usize,
}
impl AtomicLog {
fn append(&self, entry: Entry) -> Result<(), Entry> {
let current = self.head.load(Ordering::Relaxed);
if current >= self.capacity {
return Err(entry);
}
let entry_ptr = Box::into_raw(Box::new(entry));
self.buffer[current].store(entry_ptr, Ordering::Release);
self.head.fetch_add(1, Ordering::AcqRel);
Ok(())
}
}
Memory-mapped storage provides efficient disk I/O without explicit system calls. This technique leverages the operating system’s virtual memory system for transparent persistence:
use memmap2::MmapMut;
struct MappedLog {
data: MmapMut,
write_pos: AtomicUsize,
}
impl MappedLog {
fn write(&self, bytes: &[u8]) -> Result<usize, io::Error> {
let offset = self.write_pos.fetch_add(bytes.len(), Ordering::AcqRel);
if offset + bytes.len() > self.data.len() {
return Err(io::Error::new(io::ErrorKind::Other, "Log full"));
}
self.data[offset..offset + bytes.len()].copy_from_slice(bytes);
Ok(offset)
}
}
Entry batching improves throughput by reducing the number of atomic operations and I/O calls. The BatchWriter accumulates entries until reaching a threshold:
struct BatchWriter {
entries: Vec<LogEntry>,
max_size: usize,
current_size: usize,
}
impl BatchWriter {
fn add(&mut self, entry: LogEntry) -> Option<Vec<LogEntry>> {
self.entries.push(entry);
self.current_size += entry.size();
if self.current_size >= self.max_size {
let batch = std::mem::take(&mut self.entries);
self.current_size = 0;
Some(batch)
} else {
None
}
}
}
Segmented logs enable efficient log rotation and cleanup. Each segment operates independently, allowing concurrent access and maintenance:
struct LogSegment {
id: u64,
data: Vec<u8>,
active: AtomicBool,
start_offset: u64,
end_offset: AtomicUsize,
}
impl LogSegment {
fn write(&self, data: &[u8]) -> Option<usize> {
let current = self.end_offset.load(Ordering::Acquire);
let new_end = current + data.len();
if new_end > self.data.capacity() {
return None;
}
self.data[current..new_end].copy_from_slice(data);
self.end_offset.store(new_end, Ordering::Release);
Some(current)
}
fn seal(&self) -> bool {
self.active.swap(false, Ordering::AcqRel)
}
}
Zero-copy reading maximizes performance by avoiding unnecessary data copying. The LogReader provides direct access to log entries:
struct LogReader<'a> {
data: &'a [u8],
position: usize,
checksum: Crc32,
}
impl<'a> LogReader<'a> {
fn next_entry(&mut self) -> Option<&'a [u8]> {
if self.position >= self.data.len() {
return None;
}
let header = EntryHeader::parse(&self.data[self.position..])?;
let entry_end = self.position + header.length as usize;
if entry_end > self.data.len() {
return None;
}
let entry = &self.data[self.position..entry_end];
if !self.verify_checksum(entry, header.checksum) {
return None;
}
self.position = entry_end;
Some(&entry[EntryHeader::SIZE..])
}
}
These techniques require careful consideration of memory ordering and atomicity. Proper use of atomic operations ensures thread safety:
struct CommitLog {
segments: Vec<Arc<LogSegment>>,
active_segment: AtomicUsize,
config: LogConfig,
}
impl CommitLog {
fn append(&self, data: &[u8]) -> Result<LogPosition, LogError> {
let segment_idx = self.active_segment.load(Ordering::Acquire);
let segment = &self.segments[segment_idx];
match segment.write(data) {
Some(offset) => Ok(LogPosition {
segment_id: segment.id,
offset: offset as u64,
}),
None => {
self.roll_segment()?;
self.append(data)
}
}
}
fn roll_segment(&self) -> Result<(), LogError> {
let current = self.active_segment.load(Ordering::Acquire);
let new_segment = self.create_segment()?;
self.segments.push(Arc::new(new_segment));
self.active_segment.store(current + 1, Ordering::Release);
Ok(())
}
}
Error handling and recovery mechanisms ensure data integrity:
struct LogRecovery {
segments: Vec<LogSegment>,
last_valid_position: AtomicU64,
}
impl LogRecovery {
fn recover(&self) -> Result<LogPosition, RecoveryError> {
for segment in self.segments.iter() {
let valid_end = self.scan_segment(segment)?;
if valid_end < segment.end_offset.load(Ordering::Acquire) {
segment.end_offset.store(valid_end, Ordering::Release);
}
}
Ok(LogPosition {
segment_id: self.segments.last()?.id,
offset: self.last_valid_position.load(Ordering::Acquire),
})
}
fn scan_segment(&self, segment: &LogSegment) -> Result<usize, RecoveryError> {
let mut reader = LogReader::new(&segment.data);
let mut last_valid = 0;
while let Some(entry) = reader.next_entry() {
last_valid = reader.position;
self.last_valid_position.store(
segment.start_offset + last_valid as u64,
Ordering::Release
);
}
Ok(last_valid)
}
}
The combination of these techniques creates a robust, high-performance logging system suitable for demanding applications. The lock-free design eliminates contention points while maintaining data consistency and durability.
Implementation details require careful attention to memory barriers and ordering constraints. The use of appropriate atomic operations ensures thread safety without compromising performance.
I’ve found these patterns particularly effective in systems requiring high throughput and low latency. The zero-copy approach significantly reduces CPU overhead, while segmented storage enables efficient cleanup and rotation procedures.
Regular testing and monitoring help identify potential issues early. Proper instrumentation and metrics collection provide insights into system behavior and performance characteristics.
Remember to consider your specific use case when implementing these patterns. Different applications may require different trade-offs between consistency, durability, and performance.
