Home / Companies / ScyllaDB / Blog / Post Details
Content Deep Dive

Seedless NoSQL: Getting Rid of Seed Nodes in ScyllaDB

Blog post from ScyllaDB

Post Details
Company
Date Published
Author
Asias He
Word Count
1,606
Language
English
Hacker News Points
-
Summary

ScyllaDB has traditionally relied on seed nodes to manage cluster discovery and gossip information propagation, but this architecture has introduced complexities and operational challenges due to the need for special roles and configurations. Seed nodes do not participate in the bootstrap process, leading to user confusion and necessitating careful documentation to avoid errors. To address these issues, ScyllaDB aims to eliminate seed nodes in favor of a more symmetric node architecture, thereby simplifying configuration and improving robustness. The new approach involves changes to gossip shadow and normal round communication to make the process more deterministic, along with the introduction of a new synchronous RPC message to streamline communication during the gossip shadow round. These changes ensure all nodes, both seed and non-seed, must bootstrap when joining the cluster, reducing configuration errors and making the system easier to manage. These improvements are part of ScyllaDB's upcoming 4.3 release, marking a shift to a fully seedless architecture.