TiKV Best Practices
summary
TiKV is a Rust key value store using Raft + RocksDB to give ACID, external consistency and horizontal scaling and it runs under PD which manages region placement and leader distribution. Best practices include keeping three replicas, wait for tombstone before removing a store so replication finishes, identify hotspots by PD hot-region metrics and either split hot regions or migrate leaders, and tune RocksDB block cache, compression type, and TiKV thread pools to fix disk, CPU or apply-thread bottlenecks.
AI-written summary · about
chapters (20)
- 0:00Intro
- 2:08TIKV Overview and History
- 6:56Architecture
- 12:25st Part: Theories
- 13:291 The topology of a TIKV cluster
- 18:503.1 Scale - Transfer leader
- 20:27Ecosystem
- 20:462rd Part: Practices
- 22:421.2 Deployment - Cross IDC
- 23:511.3 Demployment - Configurations
- 24:532.1 Scale - Add new nodes
- 25:442.2 Scale - Remove old nodes
- 28:373.1 Fight with hotspot
- 29:253.2 Fight with hotspots - Good design
- 30:213.4 Fight with hotspots - Find write hotspot
- 31:033.5 Fight with hotspot - Find read hotspot
- 33:223.6 Fight with hotspot - Auto balancing based on statistics
- 35:413.7 Fight with hotspot - Manual balance
- 37:064.1 Performance tuning - Find bottlenecks
- 39:244.3 Performance tuning . Read in RocksDB
related talks
40mTiKV Best Practices
11mAdapting TiKV for Cloud Storage
24mUsing a Distributed Key-Value Store
29mHow We Make TiKV- a Distributed Storage, More Cost-Effective On the Cloud
28mHow Does Geo-Replication Work in TiDB?
29mDistributed SQL How TiDB Modernized Data Management
22mElastic Scheduling with TiKV
49mThe Paradigm Shift - Understanding the Move To Distributed Databases