SLOs that Delight: Master Multi-Service SLO Calculations
summary
Shows how dependencies shrink SLOs with concrete math: serial services multiply probabilities so three 9s -> 2.7 9s; replicas with retries use 1 - (1-p)^n replica math to raise availability. Gives the surprising target tradeoff: jumping from 4 nines to 5 nines cuts monthly downtime from 5 minutes to 26 seconds and forces heavy automation. Compares concrete architectures (multistack vs mesh), recommends measuring SLIs close to the user (client or synthetics), and advises starting with one imperfect SLO and iterate.
AI-written summary · about
chapters (8)
- 0:00Introduction and Session Overview
- 1:08Speaker Background and Experience
- 1:47Principles of SLOs
- 2:57Challenges in Achieving High Reliability
- 6:52Probability and SLO Calculations
- 10:21Architectural Strategies for SLOs
- 22:59Defining SLOs for Standalone Services
- 30:47Q&A Session
related talks
32m9 Things You Should Do When Starting to Use SLOs
31m9 Things You Should Do When Starting to Use SLOs
23mGoing from 30 to 30 Million SLOs
38mImplementing SRE practices: SLI/SLO deep dive
20m9 SLIs; Oh My!
37mReal World SLOs and SLIs: A Deep Dive
35mAvoiding Goodhart's Law - Use SLO's as Tools Not Cudgels
30mTechniques for SLOs and Error Budgets at Scale