Lost in Transaction?
summary
Shows that cross-resource ACID is impractical for DB + message systems, rejects two‑phase commit, and recommends concrete patterns: use the outbox pattern for atomic DB writes, enforce idempotent remote APIs, and move long retries and compensations into an open‑source workflow engine running BPMN workflows. Demonstrates a small Go service plus a workflow broker that returns 202 immediately, performs stateful retries and compensations, and scales to a real customer case of ~30,000 requests/sec with millisecond extra latency.
AI-written summary · about
chapters (12)
- 0:00Intro
- 1:50Twophase commit
- 5:03Consistency
- 6:29Outbox Pattern
- 8:08Retry
- 9:21Distributed Systems
- 13:51Stateful Cancellations
- 15:11State Machines
- 17:19Example
- 25:02Recommendations
- 26:01Compensation
- 29:36Orchestration
related talks
36mLost In Transaction
46mLost in transaction? Strategies to deal with (in-)consistency in distributed systems
47mBeing transactional in 2019 - what does this actually mean?
28mPractical Approaches To Scaling Distributed Transactions
01h 01mLost in transaction Strategies to manage consistency in distributed systems
55mTackling transactions in Microservices applications
01h 00mDistributed transactions implementation trade-offs
58mDon’t Drop ACID - Transactions in Distributed NoSQL