Relia...bility?
summary
Company works in industrial IoT processing sensor data and defines reliability from users’ perspective using metrics → SLIs → SLOs → error budgets → SLAs. Practical specifics: use Prometheus and Sloth for SLO dashboards, common SLI examples like 99% within 200ms or 95% non-500 over 7 days, and availability math (3 nines = ~8 hours/year, 4 nines = ~52 minutes, cost ~10x per extra nine). Policies tie alerts (25/50/75% burn → email/Slack/PagerDuty), multi-window burn detection (fast: 5m/2h burn rate 10; slow: 24h rate 2) and a negative budget policy (halt changes, require postmortem if one incident consumes >20%).
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:59An example from the real world
- 5:45What is Reliability?
- 7:21Service Level Indicator (SLI)
- 8:44Service Level Objective (SLO)
- 10:11How to create a good SLOS
- 13:27SLO Document
- 15:14What is an acceptable target?
- 17:37Back-of-the-envelope costs calculations
- 20:35Service Level Agreement (SLA)
- 21:58Visualization
- 23:20Alerts
- 29:46Error Budget Policy
- 32:12Reliability Stack Extended
- 33:22Why is this important?
- 35:13Shameless plug
related talks
38mImplementing SRE practices: SLI/SLO deep dive
15mShit, our Site is Down! Let's build a culture of reliability
30mMetrics for good, not evil
29mMeasuring Reliability: What Got Us Here Won't Get Us There
30mTechniques for SLOs and Error Budgets at Scale
21mSRE, what is it all about?!
35mSRE 101
37mReal World SLOs and SLIs: A Deep Dive