Site Reliability Engineering
Site reliability engineering applies software engineering to operations so teams can run dependable services while continuing to change them.
Download Markdown (.md)Soroush’s Product Wiki · https://wiki.soroush.uk/topic/site-reliability-engineering/
01What it means
Site reliability engineering, or SRE, is an approach to operating software systems with engineering methods. An SRE team treats availability, latency, capacity, and operational work as measurable design concerns. It automates repeatable tasks, defines reliability targets, and uses production evidence to decide where engineering effort belongs.[1][2]
Availability, latency, capacity, graceful failure, monitoring, response, and reduction of operational risk
Releases and experiments that create value while consuming a measured allowance for unreliability
User-centered SLI · SLO · error budget · production evidence
SRE applies software engineering to operations; targets, automation, incident learning, and capacity work must reflect actual user journeys.
02Why it matters
Digital products must improve without making every release a reliability gamble. SRE gives product and engineering teams a shared language for balancing change with dependable service. It also challenges endless manual operations: recurring work should be reduced, automated, or redesigned rather than accepted as permanent toil.
03When to use it
Use SRE practices when a service has meaningful availability expectations, recurring operational work, or many teams contributing changes. A small, low-risk product may begin with lightweight ownership and monitoring, then adopt more formal practices as dependencies and consequences grow.
04How it works
- Define user-centered service-level indicators and objectives.
- Use the resulting error budget to govern release and reliability decisions.
- Instrument the service and maintain actionable alerts, dashboards, and runbooks.
- Automate repetitive operational tasks and design systems for graceful failure.
- Respond to incidents with clear roles, then learn through blameless postmortems.
05A practical example
A payment platform measures the proportion of valid checkout requests completed successfully within two seconds. Its SRE and product teams agree on a monthly objective, monitor budget consumption, and slow risky launches when failures burn the budget unusually fast. Engineers then improve a fragile dependency instead of merely adding more alerts.
06Where it breaks
SRE is not a job-title swap for traditional operations, nor a promise of perfect availability. Metrics can misrepresent user experience, automation can amplify mistakes, and formal processes can become ceremony. The model needs leadership support, product participation, and enough engineering capacity to address systemic risks.
Common mistakes
- Treating SRE as the sole owner of production reliability.
- Copying another company’s targets without examining users and architecture.
- Calling routine manual work acceptable instead of measuring and reducing toil.