Jonathan S. Katz

Searching over vector spaces is a well-studied concept in mathematics and computer science, but the problem space has new meaning with the rise of AI/ML systems, particularly generative AI, that output very large vectors. Prior to generative AI, we predominately saw vector search in databases through geospatial, full-text, and bioinformatic/cheminformatic searches, all of which are supported...

An open discussion with Postgres community members around the topic of scaling up for challenging workloads. We will be looking at how PostgreSQL deals with each of these challenges, and discussing possible mitigations, tuning tips or workarounds.

The way to distribute data from PostgreSQL has evolved over the past decade, making it simpler to build systems for high availability, move data to other locations, or manage independent shards. Much of this is due to logical replication, which allows users to define exactly how they want to stream data between different databases.

Vectors are a centuries old, well-studied mathematical concept, yet they pose many challenges around efficient storage and retrieval in database systems. The heightened ease-of-use of AI/ML has lead to a surge of interested of storing vector data alongside application data, leading to some unique challenges.

PgBouncer is an essential tool for protecting your database against rapid spikes and churn of database connections and helping to manage transaction throughput. But how do you know how much PgBouncer processes will accommodate your application needs?