Struct NodeSecondaryIndex 

pub struct NodeSecondaryIndex { /* private fields */ }

Inverted secondary indexes on graph nodes.

• by_type: LabelId → set of node ids — keyed by the registry-assigned identifier so user-defined labels participate in the same fast path the legacy enum used to enjoy.
• by_label: label string → set of node ids (display label, not category)
• label_bloom: fast negative filter over distinct display labels

Implementations

impl NodeSecondaryIndex

pub fn new(expected_labels: usize) -> NodeSecondaryIndex

Create an empty index sized for expected_labels distinct label values (used to size the bloom filter).

pub fn insert(&self, node_id: &str, label_id: LabelId, label: &str)

Record (label_id, label, node_id) in both inverted maps.

Safe to call concurrently — each map takes its own write lock. Duplicate inserts are idempotent (sets).

pub fn remove(&self, node_id: &str, label_id: LabelId, label: &str)

Remove a node from both inverted maps. Does not rebuild the bloom (bloom filters don’t support removal — stale positives are harmless).

pub fn nodes_by_type(&self, label_id: LabelId) -> Vec<String>

Return all node ids of a given label. O(1) lookup + clone.

pub fn nodes_by_label(&self, label: &str) -> Vec<String>

Return all node ids with a given label. Uses the bloom as a pre-check — callers get an immediate empty Vec for definitely-absent labels.

pub fn count_by_type(&self, label_id: LabelId) -> usize

Cardinality of a label bucket (fast stat for the planner).

pub fn label_id_counts(&self) -> Vec<(LabelId, u64)>

Snapshot (label_id, cardinality) for every populated bucket. Cheap enough for stats() to call on demand instead of a full iter_nodes() scan.

pub fn distinct_labels(&self) -> usize

Number of distinct labels tracked.

pub fn distinct_types(&self) -> usize

Number of distinct node types tracked.

pub fn clear(&self)

Reset everything. Used by rebuild_indexes.

impl NodeSecondaryIndex

pub fn may_contain_label(&self, label: &str) -> bool

Public fast-path for label membership. Returns false iff the bloom proves the label was never inserted.

Trait Implementations

impl Default for NodeSecondaryIndex

fn default() -> NodeSecondaryIndex

Returns the “default value” for a type.

impl HasBloom for NodeSecondaryIndex

HasBloom impl pushes the label bloom through the shared trait so the unified query planner can consult it uniformly.

Note: this returns None because the underlying bloom is behind a RwLock. See NodeSecondaryIndex::may_contain_label for the actual fast-path. The impl exists so call-sites that only know dyn HasBloom can still reach the index via IndexBase::definitely_absent.

fn bloom_segment(&self) -> Option<&BloomSegment>

Reference to the bloom filter attached to this segment, if any.

fn definitely_absent(&self, key: &[u8]) -> bool

Fast-path negative check. Returns true iff the bloom is present and reports the key as absent.

impl IndexBase for NodeSecondaryIndex

fn name(&self) -> &str

Human-readable name (e.g. “users.email”, “graph.city_by_node”).

fn kind(&self) -> IndexKind

Index family (btree, hash, bitmap, hnsw, …).

fn stats(&self) -> IndexStats

Current statistics (cardinality, estimated selectivity, memory).

fn bloom(&self) -> Option<&BloomFilter>

Optional bloom filter for fast negative lookups. Cross-structure pruning relies on this.

fn definitely_absent(&self, key_bytes: &[u8]) -> bool

Returns true iff the key is guaranteed to be absent from this index. Default implementation consults IndexBase::bloom and falls back to false when no bloom is available (meaning “don’t know — caller must probe”).