4.2.3 · Lesson 7 of 16
Retrieval & Vector Store Performance Tuning
Accelerate vector search latency and optimize retrieval throughput with cluster tuning and metadata pre-filtering.
WHAT
Accelerate vector database search latency and optimize vector throughput.
HOW
- OpenSearch tuning: sharding,
knn.algo_param.ef_search, HNSW ANN indices. - Metadata Pre-Filtering: filter by
tenant_id,category,date_rangebefore distance calculations.
Key points
- OpenSearch sharding
- k-NN / HNSW ANN indices
- ef_search tuning
- Metadata pre-filtering
- Shrink candidate search space
Exam focus
Apply metadata filters before vector distance calculations to cut the candidate set — do not rely on brute-force full-corpus search.