Struct MainEdgeDataset 

pub struct MainEdgeDataset { /* private fields */ }

Main edge dataset for the unified edges table.

This table contains all edges regardless of type, providing:

• Fast ID-based lookups without knowing the edge type
• Unified traversal queries

Implementations

impl MainEdgeDataset

pub fn new(base_uri: &str) -> Self

Create a new MainEdgeDataset.

pub fn compute_edge_uid( src_uid: &UniId, dst_uid: &UniId, edge_type: &str, props: &Properties, ) -> UniId

Compute the content-addressed UID for an edge.

Edge identity is the SHA3-256 of (src_uid, dst_uid, edge_type, sorted_properties) — the same content-addressed pattern as MainVertexDataset::compute_vertex_uid but extended with edge endpoint UIDs and the edge type. This lets fork diff and promote distinguish parallel edges between the same endpoints when their property bags differ (multi-edge support).

Property iteration is sorted by key for deterministic hashing across machines and runs.

pub fn get_arrow_schema() -> Arc<ArrowSchema>

Get the Arrow schema for the main edges table.

pub fn table_name() -> &'static str

Get the table name for the main edges table.

pub fn build_record_batch( edges: &[(Eid, Vid, Vid, String, Properties, bool, u64)], created_at: Option<&HashMap<Eid, i64>>, updated_at: Option<&HashMap<Eid, i64>>, ) -> Result<RecordBatch>

Build a record batch for the main edges table.

Arguments

• edges - List of (eid, src_vid, dst_vid, edge_type, properties, deleted, version) tuples
• created_at - Optional map of Eid -> nanoseconds since epoch
• updated_at - Optional map of Eid -> nanoseconds since epoch

pub async fn write_batch( backend: &dyn StorageBackend, batch: RecordBatch, ) -> Result<()>

Write a batch to the main edges table.

Creates the table if it doesn’t exist, otherwise appends to it. Race-safe under async-flush — see crate::storage::manager::write_batch_with_lance_conflict_retry.

pub async fn ensure_default_indexes(backend: &dyn StorageBackend) -> Result<()>

Ensure default indexes exist on the main edges table.

Checks for existing indexes before creating to avoid expensive full-table rebuilds on every flush (LanceDB replaces indexes on create).

pub async fn exists_by_eid( backend: &dyn StorageBackend, eid: Eid, ) -> Result<bool>

Check whether an edge exists by EID, regardless of deletion status.

Unlike find_props_by_eid, this does NOT filter by _deleted = false, so it returns true for both active and soft-deleted edges. Used by the compaction invariant check to verify dual-writes occurred.

pub async fn find_props_by_eid( backend: &dyn StorageBackend, eid: Eid, version: Option<u64>, ) -> Result<Option<Properties>>

Find properties for an edge by EID in the main edges table.

Returns the props_json parsed into a Properties HashMap if found. This is used as a fallback for unknown/schemaless edge types.

Arguments

• version - Optional version high water mark for snapshot isolation. Mirrors MainVertexDataset::find_props_by_vid; without it a snapshot-pinned reader reads L0 and the delta tier at its snapshot but this L1 fallback at HEAD, so a post-snapshot write becomes visible.

  The bound only bites when the calling PropertyManager was built over pinned storage, which today means UniInner::at_snapshot’s time-travel view. A read-write transaction routes its scans through pinned_at_version but deliberately keeps the live, unbounded PropertyManager so property point-reads honour read-your-writes — see the design note at uni-query’s executor/read.rs. Schemaless and overflow edge properties live only in props_json (never in delta columns), so this path is reached on every such read — not only after compaction.

Errors

Returns an error if the table query fails or JSON parsing fails.

pub async fn find_edges_by_type_names( backend: &dyn StorageBackend, type_names: &[&str], endpoint_filter: Option<(EndpointSide, &[Vid])>, ) -> Result<Vec<(Eid, Vid, Vid, String, Properties)>>

Find edge data (eid, src_vid, dst_vid, edge_type, props) by multiple type names in the main edges table.

Returns all non-deleted edges with any of the given type names. This is used for OR relationship type queries like [:KNOWS|HATES].

endpoint_filter pushes a bounded endpoint set into the scan (review perf #5: a 1-source schemaless traversal used to materialize the whole edge type). None keeps the full-type scan.

Trait Implementations

impl Debug for MainEdgeDataset

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.