Qdrant v1.19.2 improves sparse vector upserts and storage support

The Qdrant vector database has released version 1.19.2, focusing on performance improvements and stability. Key updates include faster sparse vector upserts, better thread utilization during HNSW builds, and expanded object storage support for snapshots. The release also addresses several data consistency bugs and security hardening for internal gRPC calls.

Technical improvements in Qdrant 1.19.2

• Sparse vector performance: Upserts no longer rewrite the full sparse posting list every time.
• Storage expansion: Collection snapshots now support Google Cloud Storage (GCS) and Azure Blob Storage.
• Concurrency and locking: Raft snapshots no longer hold collection locks, preventing access blocks; HNSW builds now better saturate assigned threads.
• Runtime stability: Several operations, including BM25 sparse embedding and queue proxy WAL reads, were moved to blocking threads to prevent blocking the Tokio runtime.
• Data integrity: Fixed write tearing in Gridstore storages to prevent potential corruption.

Why it matters

The addition of major cloud object storage providers for snapshots makes Qdrant more viable for enterprise-scale production environments.