HTAP made live — from the engineering perspective
summary
Claims a Raft‑based HTAP: TiDB with a multi‑Raft layout plus a columnar replica, TiFlash, that receives Raft logs as a non‑voter/learner and stores column format for OLAP while three row replicas serve OLTP. Engineering specifics: Raft optimizations—pipeline, batching, parallel append/replicate and parallel commit—a split size of 96 MB and merge threshold 20 MB, epoch versioning for safe splits, and a non‑voter Raft replica pattern for near‑real‑time sync. Built and benchmarked (TPC‑H under concurrent OLTP) showing strong OLAP throughput versus Aurora.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:43ted Database 101
- 15:04you think Raft is slow?
- 19:16rld Raft optimizations: Pipeline
- 21:19Raft optimizations: Append & Log Replication (Leader only)
- 37:30ation: Non-Vote Follower
- 40:14for an OLAP system
related talks
29mHTAP vs OLTP/OLAP
01h 01mHTAP Workloads: Challenges and Solutions
36mHow Can HTAP Databases Simplify Your Data Infrastructure?
18mTiFlash: Make TiKV 10x Faster and HTAP-able
49mHTAP with Azure Cosmos DB: Hybrid Transaction & Analytical Processing
11mAdapting TiKV for Cloud Storage
49mThe Paradigm Shift - Understanding the Move To Distributed Databases
29mHow We Make TiKV- a Distributed Storage, More Cost-Effective On the Cloud