How Does Geo-Replication Work in TiDB?
summary
TiDB uses Raft-based replication with PD as the metadata router and regions (default region size 96MB) split across TiKV instances; regions default to three replicas, leaders serialize writes into logs and replicate to followers, followers ask leaders for the latest log index before serving reads, and joint consensus handles safe moves. A group commit requires replication to two groups to protect across two ACs, placement rules steer leaders and replicas, and delegates halve cross-AZ bandwidth via local forwarding.
AI-written summary · about
chapters (9)
- 0:00Introduction to TiDB
- 1:07TiDB architecture overview
- 2:14Data storage in regions
- 4:02Raft-based replication
- 5:31Replication and consistency
- 9:33Dynamic configuration changes
- 14:09Commit group and placement
- 17:18Deployment and optimization
- 25:24Q&A and concluding remarks
related talks
40mTiKV Best Practices
40mTiKV Best Practices
29mDistributed SQL: How TiDB Modernized Data Management
29mDistributed SQL How TiDB Modernized Data Management
49mThe Paradigm Shift - Understanding the Move To Distributed Databases
11mAdapting TiKV for Cloud Storage
24mUsing a Distributed Key-Value Store
29mHow We Make TiKV- a Distributed Storage, More Cost-Effective On the Cloud