Systems playing nicely: Building resilient integrations in an unpredictable world
summary
Parkster runs several hundred thousand parkings a day and uses a transactional outbox plus workers and message queues to guarantee eventual consistency; a database transaction writes both the parking record and the outbox row. Retries use idempotency keys, exponential backoff with jitter, and move failures to an error queue then a dead-letter queue for human fix. Circuit breakers and per-customer bulkheads isolate faults. A monitoring signal that predicts trouble is outbox growth; a polling partner’s 2‑minute queries once exceeded a 1‑minute timeout and triggered rework.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 4:51Resilient integrations planning
- 11:43Robust integration patterns
- 18:57Network unreliability with timeouts
- 25:23Fallbacks and retries
- 33:03Persistent failures with queues
- 36:24Circuit breakers and bulkheads
- 43:08Monitoring integrations
- 47:05Learning from failures
related talks
32mHow does your application behaves when everything goes wrong?
40mPrincipals of System Integration
34mBuilding High Resiliency into Digital Solutions
01h 01mLittle things and big things: lessons for building large systems.
25mAvoiding Overload in Distributed Systems
52mWhy today, more then ever you need Dependency Isolation in Distributed Systems
41mDesign (Failures) Case Study
26mHow to avoid breaking other people's things