SRE From Inception to Global Scale
summary
LinkedIn pivots from biweekly monster releases to a trunk-based, self-service deployment model with canaries, metric-validation and A/B feature gating, enabling safe rollouts on an active‑active four‑data‑center platform that runs Kafka at trillions of messages per day and a social graph of ~60 billion edges; site reliability teams automate failover (forced data‑center failures three times weekly), retire the NOC with workflow automation (nurse) and Iris escalation, and scale Bangalore to 60+ SREs and 100+ app devs while building tools like DC Manager reduced build time to 30 days and an ML‑driven alert‑correlation system.
AI-written summary · about
chapters (27)
- 0:00Introduction
- 1:10Bruno Conway Introduction
- 1:46Regina Wang Introduction
- 2:27How many people use LinkedIn
- 2:53Getting to know LinkedIn
- 3:17The Economic Graph
- 4:29The Platform
- 5:07The Infrastructure
- 6:23Agent Works
- 6:41Data Centers
- 7:26Engineering Footprint
- 8:58A3 approach
- 12:03Growth curve
- 13:18Technical debt
- 14:34Activeactive model
- 16:56Early decisions
- 17:32Selfservice application
- 19:14Culture of metrics
- 19:26Selfservice communication platform
- 20:53The final early decision
- 21:28LinkedIns selfservice approach
- 22:27NOC is no longer needed
- 25:35Grounding Principles
- 26:26Playbook
- 26:43Hiring
- 34:51Team Characteristics
- 40:17Takeaways
related talks
22mBuilding SRE: Culture from the Outside In
28mEvolution of SRE and Rising Need of SRE Catalyzers
41mBuilding an SRE Capability Inside a Large Organization
21mSRE: An incomplete guide to cultural Narnia
30mThe Path to SRE
16mSRE in Transition: From Startup to Established Business
26mFrom Nothing to SRE: Practical Guidance on Implementing SRE in Smaller Organisations
18mSRE Engagement Model Transition in Building and Expanding SRE Team