Distributed Loose Index Scan (aka Skip/Jump/Hybrid Scan): Time Series in YugabyteDB
summary
YugabyteDB implements a storage‑level iterator called a hybrid scan (aka Loose Index Scan) that jumps between index keys so a 31 million row time‑series table can return distinct device IDs after reading ~32 rows and issuing one RPC instead of scanning millions. Implementation sits in the LSM/RocksDB‑derived tablet layer with a new iterator; a full count of 10 million rows still takes ~13s, but distinct and filtered queries run in milliseconds.
AI-written summary · about
chapters (9)
- 0:00Distributed Loose Index Scan intro
- 3:36Time series example 31M rows
- 8:03Skip scan optimizes index access
- 13:43Hybrid scan in YugabyteDB
- 17:39Execution plan analysis
- 20:31Misconceptions about YugabyteDB
- 23:43Architecture advantages
- 25:30SQL data modeling comparison
- 27:19Testing and contributing
related talks
02h 27mDistributed SQL Tuning and Optimization in YugabyteDB
26mEfforts Toward Adoption and Uncovering Hidden Features
01h 27mYugabyteDB YSQL Development
27mSparking our way to YugabyteDB: Our journey from ScyllaDB to
30mPerformance: Tips & Tricks to Optimize the Speed of Your Mission-Critical Database
29mData Modeling for Distributed SQL
27mUnified AI & Relational Power: Scalable Distributed Vector Search with
27mIntroduction to YugabyteDB: Design and Architecture