Company
Date Published
Author
Hristo Stoyanov
Word count
2098
Language
English
Hacker News points
None

Summary

The new approach to failover in SingleStoreDB Self-Managed 7.1 aims to improve the availability and performance of the database by distributing data redundantly across multiple leaf nodes, reducing the impact of node failures on query performance. In this new approach, the partition placement algorithm distributes master copies of partitions evenly across leaves, and then distributes replicas needed by any one leaf node across multiple other leaf nodes. This reduces the creation of hot spots that can slow down queries after a node failure. The feature is enabled through a global variable called `leaf_failover_fanout`, which determines whether to use load-balanced or paired failover modes, with load-balanced mode providing better performance and reduced buffer space requirements. The upgrade process for existing databases involves setting the global variable to 'load_balanced' and running REBALANCE operations, which may incur costs such as bandwidth, time, and disk space. Overall, the new approach is designed to provide improved availability, performance, and flexibility in managing database failures.