zilliztech/VectorDBBench v1.0.21

VDBBench 1.0.21 release

on GitHub

VDBBench 1.0.21 adds Lindorm, PolarDB, Apache Pinot, VectorChord, and Intel SVS support; migrates Milvus benchmarks to MilvusClient; adds concurrent insert support; upgrades to Pydantic v2; improves Streamlit error
surfacing; and refreshes benchmark results for Milvus, ElasticCloud, ZillizCloud, and turbopuffer.

What's Changed

New database and index support

• Added Lindorm benchmark support (#718)
• Added PolarDB benchmark support (#737)
• Added Apache Pinot vector search client (#757)
• Added VectorChord benchmark support (#745)
• Added Intel SVS support (#749)
• Added SQ4U scalar quantization type for Milvus HNSW indexes

Benchmark runner improvements

• Added concurrent insert support for performance cases (#741)
• Fixed concurrent insert runner behavior for non-thread-safe databases (#764)
• Capped default insert workers to min(cpu, 4) to avoid overloading clients by default (#767)
• Added maintenance_work_mem and max_parallel_workers parameters for PgVector IVFFlat (#736)

Milvus and database client updates

• Migrated Milvus integration from PyMilvus ORM to MilvusClient (#738)
• Skipped Milvus compaction when permission errors occur (#753)
• Added label-filter support for PgDiskANN (#724)
• Fixed Qdrant timeout propagation (#725)
• Fixed pgvector index type normalization and ivf_flat → ivfflat mapping (#760, #763)
• Added self-hosted Elasticsearch support through --host / --port options in ElasticCloud commands (#761)
• Improved turbopuffer region handling, SDK compatibility, pickle/ID compatibility, and benchmark results

UI, dependency, and compatibility updates

• Upgraded to Pydantic v2 (#750)
• Restored ujson dependency (#744)
• Upgraded Streamlit and improved Run Test page error surfacing (#766)

Benchmark results and leaderboard updates

• Refreshed benchmark results for Milvus 2.6.14, ElasticCloud 8.17, ZillizCloud, and turbopuffer
• Filled missing Milvus and ZillizCloud build durations (#751, #754)
• Unified result timestamps and run_id values across result files
• Aligned streaming leaderboard labels with vector-search results
• Updated ZillizCloud benchmark results with Cardinal backend results

Docs and CI

• Added missing pgvector reranking CLI options to README (#726)
• Added stale-issue auto-close workflow (#730)
• Added /keep-open comment workflow for stale issues

Full Changelog

v1.0.20...v1.0.21