Struct EmbeddingRegistry 

pub struct EmbeddingRegistry {
    pub concurrency: usize,
    /* private fields */
}

Generic Qdrant-backed embedding registry.

Owns a QdrantOps instance, a collection name and a UUID namespace for deterministic point IDs (uuid v5). The in-memory hashes map enables O(1) delta detection between syncs.

The cached_dim field caches the collection’s vector dimension after the first successful sync so that search_raw can validate the query vector dimension without an extra Qdrant round-trip on every call. When a mismatch is detected, search_raw returns EmbeddingRegistryError::DimensionProbe instead of silently issuing a gRPC search that would return near-zero cosine scores (Qdrant gRPC behaviour on dim mismatch).

Fields

concurrency: usize

Maximum number of embedding requests dispatched concurrently during a sync.

Implementations

impl EmbeddingRegistry

pub fn new( ops: QdrantOps, collection: impl Into<String>, namespace: Uuid, ) -> Self

Create a registry wrapping an existing QdrantOps connection.

pub async fn sync<T: Embeddable>( &mut self, items: &[T], embedding_model: &str, embed_fn: impl Fn(&str) -> EmbedFuture, on_progress: Option<Box<dyn Fn(usize, usize) + Send>>, ) -> Result<SyncStats, EmbeddingRegistryError>

Sync items into Qdrant, computing a content-hash delta to avoid unnecessary re-embedding. Re-creates the collection when the embedding model changes.

on_progress, when provided, is called after each successful embed+upsert with (completed, total) counts so callers can display progress indicators.

Errors

Returns EmbeddingRegistryError on Qdrant or embedding failures.

pub async fn search_raw( &self, query: &str, limit: usize, embed_fn: impl Fn(&str) -> EmbedFuture, ) -> Result<Vec<ScoredVectorPoint>, EmbeddingRegistryError>

Search the collection, returning raw scored Qdrant points.

Validates that the query vector dimension matches the collection before issuing the gRPC call. Qdrant gRPC silently returns near-zero cosine scores (~0.022) when dimensions mismatch instead of returning an error — this guard prevents that silent failure.

The dimension is checked against the cache populated by the most recent sync call. If no sync has occurred (cache is None) the check is skipped to avoid blocking reads before the first sync.

Consumers map the payloads to their domain types.

Errors

Returns EmbeddingRegistryError::DimensionProbe when the query vector dimension does not match the stored collection dimension. Returns EmbeddingRegistryError::Embedding if the embed function fails, or EmbeddingRegistryError::VectorStore on Qdrant search failure.

Trait Implementations

impl Clone for EmbeddingRegistry

fn clone(&self) -> EmbeddingRegistry

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for EmbeddingRegistry

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

Formats the value using the given formatter.