Fearless global transactions with
summary
CockroachDB runs a shared‑nothing cluster of 512MB ranges with Raft (default 3x replication) and targets a 20ms global write budget by combining three SQL primitives: database regions, declarative survival goals and table locality. It uses non‑voting replicas to decouple quorum size from geo footprint, adds regional by row partitioning (hidden crdb_region column) to home rows per region, and implements global reads via shifted timestamps with a commit wait and future closed timestamp notifications so reads are low‑latency while writes pay the consistency cost.
AI-written summary · about
chapters (17)
- 0:00Start
- 4:27Globally Distributed Databases
- 7:56Horizontal Scaling
- 9:54Overview
- 23:12Non-Voting Replicas
- 24:43Latency Survivability
- 26:10Table's Locality Setting
- 28:17What if I Take an Existing Postgres Fact Application and Move It Over to Cockroach without Configuring any Table Locality Settings
- 32:19Why Do We Need Regional Tables in Home Regions At All
- 35:31Why Is this Column Hidden
- 42:00Locality Optimized Search
- 46:19Closed Timestamp
- 49:37Non-Blocking Transactions
- 51:33How Global Tables Are Implemented under the Hood
- 57:52Ensure Consistency
- 58:51The Read Uncertainty Interval
- 1:02:24Global Tables
related talks
48mBuilding mission-critical applications with CockroachDB
42mThe hows and whys of a distributed SQL database
52mCockroachDB
24mNavigating Distributed Databases: Challenges & Opportunities
22mBuilding CockroachDB for performance at scale
31mFuture-proofing operational data at scale
14mHow CockroachDB can enable Multi-Region Deployments and Data Locality 
37mRunning a Cloud-Native SQL Database on Mesos