Vector Databases in Production
Comparing pgvector self hosted, cloudsql with pgvector extension, and qdrant self hosted
Here are the links I found:
Comprehensive benchmark comparisons: - https://nirantk.com/writing/pgvector-vs-qdrant/ - https://www.timescale.com/blog/pgvector-vs-qdrant (also at https://www.tigerdata.com/blog/pgvector-vs-qdrant) - https://www.timescale.com/blog/pgvector-vs-pinecone (also at https://www.tigerdata.com/blog/pgvector-vs-pinecone) - https://qdrant.tech/benchmarks/ - https://redis.io/blog/benchmarking-results-for-vector-databases/ - https://blueteam.ai/blog/vector-benchmarking/
Methodology and benchmarking tools: - https://ann-benchmarks.com/ - https://weaviate.io/developers/weaviate/benchmarks/ann - https://opensourceconnections.com/blog/2025/02/27/vector-search-navigating-recall-and-performance/ - https://zilliz.com/learn/benchmark-vector-database-performance-techniques-and-insights - https://turbopuffer.com/blog/continuous-recall
Selection guides and comparisons: - https://jkfran.com/selecting-ideal-self-hosted-vector-database/ - https://www.firecrawl.dev/blog/best-vector-databases-2025 - https://www.myscale.com/blog/comprehensive-comparison-pgvector-vs-qdrant-performance-vector-database-benchmarks/ - https://zilliz.com/comparison/qdrant-vs-pgvector - https://data-ai.theodo.com/en/technical-blog/how-to-choose-your-vector-database-in-2023
Setup and getting started: - https://www.sarahglasmacher.com/how-to-pgvector-docker-local-vector-database/ - https://www.yugabyte.com/blog/postgresql-pgvector-getting-started/ - https://github.com/pgvector/pgvector - https://github.com/timescale/pgvectorscale
Production lessons learned: - https://towardsdatascience.com/six-lessons-learned-building-rag-systems-in-production/ - https://blog.abdellatif.io/production-rag-processing-5m-documents - https://medium.com/@kaushalsinh73/rag-architectures-in-production-lessons-learned-the-hard-way-dff06c78a9fa
Amazon RDS for PostgreSQL:
pgvector supported as an extension Straightforward, familiar if you’ve used RDS before Good enough for most production workloads
Amazon Aurora PostgreSQL:
Better performance characteristics than vanilla RDS Supports pgvector Storage auto-scaling, faster failover Higher cost but worth it for production vector search at scale
Google Cloud platform