Kévin Maschtaler
Developer at marmelab
Errors Budget
An SRE Principle
Errors Budget
Deployment
Reliability
Monitoring
Benjamin Treynor Sloss, VP of Engineering, Google
Get To Know What Really Matters For Your Users
Weather API
Availability
Database
Data Consistency
Stock Exchange App.
Response Time
Define A Service Level Indicator
Availability = \frac{Uptime}{Uptime + Downtime}
Choose A Service Level Objective
Unrealistic & unreachable
Do more harm than good
99% ("two nines"): 3.65 days of downtime
99.9% ("three nines"): 8.77 hours of downtime
99.99% ("four nines"): 52.60 minutes of downtime
99.999% ("five nines"): 5.26 minutes of downtime
Example of realistic SLOs (over a year)
Focus on unplanned downtime
Errors Budget = SLI - SLO
if (budget > 0 && !friday)if (budget <= 0 || friday)
Recap
1. Get To Know What Really Matters For Your Users
2. Measure it (SLI)
3. Choose A Realistic Objective (SLO)
4. Align Team Behavior With The Errors Budget
5. Iterate and goto 1
- Risky
- Not Risky
Focus On Stability
Focus On Velocity
DEMO
About Site Reliability Engineering
- Google (Benjamin Treynor Sloss)
- Microsoft (David N. Blank-Edelman)
- Stack Overflow (Nick Craver)
- leboncoin.fr
- vente-privee.com
- AirBnb
- Amazon
- Apple
- Baidu
- Dropbox
- Etsy
- GitHub
- Netflix
- Uber
- Yahoo!
- Yelp
- ...