Being Transactional in 2019 — What Does That Actually Mean?
summary
Examines real transactional failure modes such as double-booking in reservation systems and explains which isolation levels (read committed, repeatable read, serializable, snapshot) allow those problems like phantom rows and lost updates. Describes concrete systems: MongoDB 4.0 adds multi-document transactions on replica sets but not across shards; Kafka provides atomic producer transactions and atomic commit of consumer offsets for end-to-end exactly-once; Google Spanner uses synchronized clocks to mitigate coordination costs while distributed two-phase commit/consensus imposes latency and availability trade-offs.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 3:10Transaction isolation
- 8:25Isolation level issues
- 13:32Distributed consistency theory
- 21:20Distributed transaction challenges
- 26:18New database approaches
- 33:21Transactions in event logs
- 41:31Summary
related talks
47mBeing transactional in 2019 - what does this actually mean?
41mTransactions: myths, surprises and opportunities
41mConcurrency, Scalability and Transactions -- Myths and Surprises
45mTransactions and Concurrency Control Patterns
17mTransactionalism Explained
16mTransactionality Explained
38mTransactional Work with Topics
28mUnderstanding Transaction Isolation for App Developers