Data Corruption: Stop the Evil Tribbles
summary
Argues recovery beats heroic prevention, showing concrete fixes: a Return‑to‑Vendor UI disambiguation replaces a destructive migration that hit production, CI failures expose cross‑team mismatches, and simple audit jobs every five minutes catch inconsistencies. Recommends transactions and Rails validations (and acknowledges when service objects or raw SQL force different tradeoffs), cites the 95% case of culture and architecture causing corruption rather than exotic Postgres bugs.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 2:12Talk title change
- 4:20Product changes cause corruption
- 9:28Miscommunication and assumptions
- 14:32Individuals over processes
- 17:42Missing integrity patterns
- 26:02CAP Theorem trade-offs
- 28:14Lightweight audit processes
- 31:02Bizarre recovery cases
- 33:14Truth about integrity work
- 34:42Q&A
related talks
52mData Corruption Bugs: Diagnosis and Lessons
34mToo Big to Fail
28mPerfect is too expensive
31mCode like an accountant: Designing data systems for accuracy, resilience and auditability
37mWhen it all goes Wrong (with Postgres)
37mDatabase in Distress: Testing and Repairing Different Types of Database Corruption
36mTales from the Postgres Front, and What We Can Learn
51mDatabase in Distress: Testing and Repairing Different Types of Database Corruption