The Map Is Not the Territory: How SLOs Lead Us Astray, and What We Can Do about It
Talkat SREcon19 Europe/Middle East/Africa · on Thursday, 2019-10-03· at 09:45· Track 1: The Liffey B·
summary
SLOs rest on shaky assumptions—independent requests, uniform request value and evenly distributed errors—that break on modern microservice stacks and low‑QPS or analytics workloads, causing bad prioritization and invisible customer pain. Practical fixes: measure per customer, add per-customer SLOs and median‑plus‑tail guarantees, document rationale, use canaries and user‑validation, supplement with disaster testing and historical time‑to‑resolution tracking. Don’t sandbag SLOs for catastrophes; use SLAs or other indemnification for rare tail events, and treat queues and async refactors as SLO‑measurement hazards.
AI-written summary · about
chapters (16)
- 0:00Start
- 2:57SLOs by Use Case
- 5:10SLOs as a model of service behavior
- 6:24SLO Use Cases
- 13:02Ongoing Maintenance of SLOS
- 16:16Design - What goes wrong?
- 20:47Supporting Design - Best Practices
- 24:19Mechanics of Response
- 24:38Incident Response - What goes wrong
- 25:32Incident Response - Best Practices
- 26:31Bounding Tail Behavior
- 26:48What goes wrong at the tail?
- 29:56Bounding the Tail - Best practices
- 33:28SLO Model Failures
- 34:01SLO Fidelity
- 34:38SLO Algebra
related talks
28mMaking the Impossible Impossible: Improving Reliability by...
41mOptimizing for Value - Fight Failure Demand With Bottlenecks
39mUnderstanding minimal viable quality promotes craft
45mPanel: Unsolved Problems in SRE
24mIntroducing the Reliability Map – r9y.dev
32mScaling Done Right
52mSOA lessons learnt OR Microservices done better
01h 02mTools for simplified large scale systems