Not Invented Here Syndrome and Dark Debt: The PagerDuty Story
summary
PagerDuty builds a Cassandra‑backed distributed queue called WorkQueue in 2011 to get high availability, durability and synchronous replication across vendors. It works initially but accumulates complexity: host‑based partitioning, per‑item queues, time bucketing and many internal adopters. Cassandra API changes and the rise of Kafka and managed pub/sub expose failures and postmortems mention Cassandra repeatedly. Official support ends 2016, usage shrinks by 2017, and the exercise surfaces dark‑debt from proliferated custom tooling.
AI-written summary · about
chapters (12)
- 0:00Start
- 2:49Comparison to technical debt
- 3:31Build vs Buy: The age old question
- 4:27October 24, 2011
- 5:38Background
- 6:21Durability
- 6:51High Scalability
- 7:37Flexible Schema
- 8:07Backups
- 8:42Synchronous Replication
- 13:09Fair partitioning
- 17:27Lessons learnt
related talks
52mWhy today, more then ever you need Dependency Isolation in Distributed Systems
31mFuture Data Platforms and PaaS Services
01h 05mSpeed & Scale: How to Get There
03h 04mHow We Evolve Clever Cloud – Postmortem on a 10‑Year Project
29mComplexity: The Crucial Ingredient in Your Kitchen
47mEverything Loosely Coupled - Advanced Integration Patterns and Their Trade-Offs
19mLearnings & Patterns from Zero-Downtime Migrations
55mCode the future, now