Struct ArenaMvccMemTable 

pub struct ArenaMvccMemTable { /* private fields */ }

Arena-backed MVCC MemTable with optimized key storage

This version uses ArenaKeyHandle instead of Vec<u8> for keys, reducing per-write allocations from 3 to 1:

• Before: 3 × Vec clones per write (dirty_list, ordered_index, data)
• After: 1 × ArenaKeyHandle creation, 3 × O(1) copies (16 bytes each)

Performance

Expected improvement: 20-30% throughput increase on write-heavy workloads by reducing:

• Heap allocations: 3 → 1 per write
• Bytes copied: 3L → L + 48 bytes (where L = key length)

Implementations

impl ArenaMvccMemTable

pub fn new() -> Self

pub fn with_ordered_index(enable_ordered_index: bool) -> Self

pub fn write( &self, key: &[u8], value: Option<Vec<u8>>, txn_id: u64, ) -> Result<()>

Write a key-value pair using arena key handle

Only creates ONE ArenaKeyHandle, then copies it (16 bytes) to each location. This is much cheaper than cloning Vec which requires heap allocation.

pub fn write_batch( &self, writes: &[(&[u8], Option<Vec<u8>>)], txn_id: u64, ) -> Result<()>

Write batch using arena key handles

pub fn read( &self, key: &[u8], snapshot_ts: u64, current_txn_id: Option<u64>, ) -> Option<Vec<u8>>

Read at snapshot timestamp

pub fn commit( &self, txn_id: u64, commit_ts: u64, write_set: &HashSet<InlineKey>, )

Commit transaction

pub fn abort(&self, txn_id: u64)

Abort transaction

pub fn scan_prefix( &self, prefix: &[u8], snapshot_ts: u64, current_txn_id: Option<u64>, ) -> Vec<(Vec<u8>, Vec<u8>)>

Scan prefix

pub fn size(&self) -> u64

Get approximate size

pub fn gc(&self, min_active_ts: u64) -> usize

Garbage collect old versions

Trait Implementations

impl Default for ArenaMvccMemTable

fn default() -> Self

Returns the “default value” for a type.