October 2024 Summaries

Vitess 21 is now generally available, focusing on enhancing query compatibility, improving cluster management, and expanding VReplication capabilities. Key features include reference table materialization, multi-metric throttler support, enhanced online DDL functionality, a new mysqlshell engine for backup and restore processes, and detailed execution traces and schema analysis with vexplain. The Vitess Kubernetes Operator introduces horizontal auto-scaling for VTGate pods and Kubernetes 1.31 support, improving overall scalability and deployment flexibility.

PlanetScale has announced the public beta of its vector search and storage feature for MySQL databases. This implementation is based on Microsoft Research's SPANN (Space-Partitioned Approximate Nearest Neighbors) and SPFresh algorithms, which are designed to work well with larger-than-RAM indexes requiring SSD usage. The vector search solution supports pre-filtering and post-filtering, full SQL syntax, and ACID compliance. It is integrated with InnoDB and Vitess for efficient management of vector indexes at terabyte scale. To enable vector support, users can go to their database settings page, click "Beta features," find Vectors, and enroll in the beta program.

The text delves into the design considerations for implementing a database throttler, comparing singular and distributed deployment models. A singular throttler acts as a centralized service that manages requests by accessing database and OS metrics directly, but it raises concerns about high availability and scaling. In contrast, a distributed throttler involves multiple instances, potentially across different availability zones, which can operate independently or collaborate by sharing metrics. The discussion also highlights the complexities introduced by separating metric collection from the throttler itself, such as through an API or agent, which can lead to issues with synchronicity and data staleness. The text explores the concept of reducing the throttler's impact on the system by controlling the frequency of metric collection and communication, suggesting strategies such as metric hibernation and efficient handling of replication lag. The challenges and trade-offs of each approach are examined, with a case study on the Vitess tablet throttler demonstrating a practical application of these principles.