vector-db-benchmark
vector-db-benchmark copied to clipboard
OpenSearch integration improvements
The bellow changes aim to:
- Ease the OpenSearch integration, while also reusing as much as possible the connection creation steps.
- Add support for anonymous auth on opensearch client
- Add support for HTTPS traffic and detect ssl configs from connection details
- Ensure that the cluster is at least at yellow state before querying
- Enables inner product distance (supported for Lucene in OpenSearch version 2.13 and later)
- Enables vector dimensions up to 16K (Source: https://opensearch.org/docs/latest/search-plugins/knn/approximate-knn/)
- Uses a backoff stragegy on search_one method when required (due to 429 errors). Fixes https://github.com/qdrant/vector-db-benchmark/issues/142