Order And Consistency At Scale
summary
Spanner synchronously replicates ranges with Paxos and enforces external consistency so concurrent transactions appear serializable and ordered by real time, enabling lockless consistent snapshot reads. A global TrueTime service provides commit timestamps with an average uncertainty under 1.5 milliseconds, letting read‑only transactions be cheap and nonblocking. Production Spanner handles hundreds of millions of requests per second across 30,000 databases and exabytes of data, and Zanzibar uses Spanner for ACLs at roughly 20 million reads per second with 95% checks under 10 ms.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:09Bigtable to Spanner design
- 3:54Secondary index challenges
- 5:19Strict consistency matters
- 8:27Correctness and performance
- 10:10External consistency with TrueTime
- 13:22Read-only transactions
- 15:47Secondary index timestamps
- 16:09Zanzibar access control
- 18:18Latency vs consistency
- 21:06Intuitive system semantics
related talks
49mSpanner - fully managed horizontally scalable relational DB with ACID transactions
41mSpanner - Google's Distributed Database
47mGoogle Cloud Spanner: Global consistency at scale
01h 08mKeynote: Building Spanner
33mGoogle Cloud Spanner: Global consistency at scale
37mGoogle Cloud Spanner: Going global with relational database
21mZanzibar: Google’s Consistent, Global Authorization System
01h 00mDistributed transactions implementation trade-offs
