SRE (site reliability engineering)

Site reliability engineering (SRE) is a discipline that treats operations as a software problem. SRE teams set measurable reliability targets for a service, automate repetitive operational work, and use data from production to decide when to ship features and when to invest in stability. Google introduced the practice in the early 2000s.

How it works

SRE starts with what users need. The team picks service level indicators (SLIs), such as the share of requests served in under 300 ms, and sets service level objectives (SLOs) on them. The gap between the SLO and 100% is the error budget. A 99.9% monthly SLO allows 43.2 minutes of failure in 30 days. Google's SRE book also suggests capping manual operational work, called toil, at 50% of an engineer's time so the rest goes to automation.

Why it matters in production

Without SRE, reliability depends on whoever has been on call the longest. With it, a team has shared targets, clear rules for when to slow releases, and blameless postmortems that stop the same incident from recurring.