Struct EntityCache 

pub struct EntityCache { /* private fields */ }

Sharded, bounded LRU cache for UnifiedStore::get_any lookups.

Implementations

impl EntityCache

pub fn new() -> Self

pub fn with_capacity(total_capacity: usize) -> Self

pub fn get(&self, key: u64) -> Option<(String, UnifiedEntity)>

Cache lookup. Updates hits / misses.

Acquires the shard write lock on a hit (LRU touch needs it). If the touch cost ever shows up in profiles we can split into a fast read-only probe + deferred touch, but at shard size ≈ 625 it’s not the bottleneck.

pub fn insert(&self, key: u64, value: (String, UnifiedEntity))

Insert. Counts evictions for observability.

pub fn remove(&self, key: u64) -> bool

Remove a single key. Returns true if the key was present.

Hot path on delete_batch / delete. Probes with a read lock first: most deletes target rows that were never read, so the cache miss case avoids the write-lock acquisition entirely. This is the load-bearing optimisation for #85’s delete_sequential regression — the original code took the write lock unconditionally, even for a 100 % miss workload.

pub fn remove_many(&self, keys: impl IntoIterator<Item = u64>)

Bulk remove. Same fast-path optimisation as remove: skip the shard’s write lock entirely when none of its candidate keys are present.

pub fn retain<F>(&self, keep: F)

where F: FnMut(u64, &(String, UnifiedEntity)) -> bool,

Drop every entry whose (collection, entity) value fails keep. Used by collection-drop paths.

pub fn len(&self) -> usize

Total entries across all shards. Diagnostic only — locks every shard.

pub fn hit_rate(&self) -> Option<f64>

Hit / (hit + miss). Returns None if no lookups have occurred.

pub fn stats(&self) -> EntityCacheStats

Snapshot of the three counters: (hits, misses, evictions).

Trait Implementations

impl Default for EntityCache

fn default() -> Self

Returns the “default value” for a type.