Struct FederatedSearch 

pub struct FederatedSearch { /* private fields */ }

Concurrent search over multiple AsyncVectorIndex backends.

Queries every backend at once, applies a per-backend timeout so one slow remote cannot stall the whole query, drops failing or timed-out backends with an observable tracing::warn!, and merges the survivors with the configured FusionStrategy. If every backend fails, returns KernelError::Search.

Synchronous backends (e.g. TurbovecIndex) participate via federate_results instead — search them directly and fold the list in.

Implementations

impl FederatedSearch

pub fn new() -> Self

Create an empty federated search (default strategy RRF k=60, 5s timeout).

pub fn with_backend(self, index: Arc<dyn AsyncVectorIndex>, weight: f32) -> Self

Add a backend with a fusion weight (used only by FusionStrategy::WeightedSum; ignored by RRF).

pub fn strategy(self, strategy: FusionStrategy) -> Self

Set the fusion strategy (default FusionStrategy::Rrf k=60).

pub fn timeout(self, timeout: Duration) -> Self

Set the per-backend query timeout (default 5s). A backend that exceeds it is dropped with a warning rather than blocking the query.

pub async fn search( &self, query: &[f32], k_req: usize, ) -> Result<Vec<SearchResult>>

Run query against every backend concurrently, merge survivors.

Each backend is queried for 2 * k results (over-fetch) so RRF rank-credit is preserved for a document that ranks just below k in one backend but near the top in another; the fused list is then truncated to the requested k. A per-backend timeout drops slow or failing backends with an observable tracing::warn! rather than stalling the query.

Returns the fused result list (at most k items). KernelError::Search is returned only when no backend succeeded; one or more survivors yield a partial (but non-empty) merged result.

Trait Implementations

impl Default for FederatedSearch

fn default() -> Self

Returns the “default value” for a type.