SingleStore: Storage Metadata for Modern Cloud Databases
summary
SingleStore adds a centralized metadata DB — the bottle service — that records blobs, snapshots, log chunks, compute sessions and per-blob references in a SingleStore cluster with stored procedures and external Golang workers, enabling branching by copying reference rows and enabling cross‑region SmartDR by bi‑directional file replication. Garbage collection runs by cleaning references below a retention LSN and physically deleting unreferenced blobs; system runs in production for ~18 months with occasional hiccups (one run deleted ~1.5B files after an outage).
AI-written summary · about
chapters (8)
- 0:00Introduction and agenda
- 2:04SingleStore storage overview
- 9:21The need for branching
- 12:21The bottle service design
- 21:59Service tables and structure
- 34:16Database branching application
- 38:41Smart DR application
- 44:32Conclusion and Q&A
related talks
46mKEEPING UP IN A DATA-INTENSIVE WORLD WITH SINGLESTORE
31mHow Kurtosys Ditched Its Legacy Database To Drive The Future Of Investment Management
41mThe New Generation Of Data Stores
43mNatively Engineered for AI: The Evolution
33mNext Gen of Datastores: Hot Queries, Cold Storage
19mKeep Storing and Carry On
19mSnowpark & SingleStore: A Seamless Integration Demo
28mPersistent Storage with Kubernetes in Production - Which Solution and Why?