BanyanDB: A Purpose-Built Observability database
summary
BanyanDB is a purpose-built observability store for SkyWalking that implements the RAM conjecture to trade read latency for far lower memory and CPU costs while keeping rich span and log filtering. It uses segmented span buffers, auto-flush to a KV-backed column layout, a block-attached index for sparse trace IDs and horizon-tree compaction for queries. The project targets multi-modal metrics/traces/logs, aims for better cost/performance than Elastic/TSDB combos, and ships a first release at month end.
AI-written summary · about
chapters (14)
- 0:00Intro
- 2:14Table of contents
- 2:48Tracing waterfall
- 3:42Span Inside
- 5:07Store traces into Bigtable
- 6:20Principle of Tracing Storage
- 8:34Multiple Storage Layer
- 15:56RUM Space
- 17:16Examples
- 20:07RUM conjecture
- 20:59RUM aware system
- 26:17Structure
- 30:46Compression
- 35:56Thank You
related talks
12mSpanDB: A Fast, Cost-Effective LSM-tree Based KV Store on Hybrid Storage
45mThe Database: A Sophisticated and Special Data Structure
44mBerkeleyDB
16mScaleDB: A Scalable, Asynchronous In-Memory Database
40mKalDB: A cloud native log search platform
24mSplinterDB: Closing the Bandwidth Gap for NVMe Key-Value Stores
15mSaved by In-Memory NoSQL
25mYugabyteDB Helps Bed Bath and Beyond Improve Their Digital Customer Experience