What Exactly Does an SRE Do? Unpacking Roles, Skills, and Practices

SRE (Site Reliability Engineering) was introduced by Google to address the instability caused by rapid development cycles, using standardization, automation, and scalability to balance service quality and stability.

Different companies define SRE roles variably, including network SRE, DBA SRE, business SRE, and security SRE, but the common baseline is a maintenance engineer focused on service quality.

Key competencies for SREs include:

Broad skill set covering networking, OS, monitoring, CI/CD, and development.

Product‑mindset communication that breaks traditional ops silos.

Software‑engineered solutions for operational problems.

Strong troubleshooting and abstraction abilities .

In China, SREs are often split into two tiers: PasS‑SRE (platform reliability) and business SRE (application reliability), with the latter resembling traditional business ops.

Observability System

Effective observability consists of three pillars: metric monitoring, log collection, and distributed tracing. It requires clear quality standards and systematic monitoring rather than ad‑hoc checks.

Complete metric collection across devices and tech stacks.

Support for massive numbers of devices.

Storage and analysis of monitoring data for visualization and automated insights.

Enterprise‑grade observability should be platform‑based, allowing configuration or development of new metrics and integration with specialized tools.

Fault Response

When a failure occurs, the response workflow includes alerting, communication, and recovery. Effective alerts must be timely and accurate to avoid alert fatigue.

Alert compression techniques—trend prediction, short‑cycle detection, baseline evaluation—help reduce noise, and health scoring can guide operators to prioritize issues.

Testing and Deployment

Testing aims to limit incidents while enabling rapid releases. Error budgets dictate how much testing resources to allocate: a high budget allows lighter testing, while a low budget requires stricter validation.

Automation pipelines handle compilation, testing, release preparation, alert silencing, service stop/start, and database migrations.

Capacity Planning

Capacity planning predicts future demand and identifies system limits, using massive operational data to assess current capacity, forecast limits, and recommend adjustments.

Effective platforms provide fast data retrieval, multi‑dimensional queries, and robust visualization to support capacity analysis.

Automation Tool Development

SREs spend roughly half their time building tools that automate repetitive tasks, improving efficiency, standardizing operations, and preserving institutional knowledge.

Increased efficiency through code‑driven automation.

Standardized, error‑free operational procedures.

Codified expertise that can be shared across teams.

User Support

SREs prioritize user experience, using logs, metrics, and tracing to reconstruct user journeys and ensure service reliability from the end‑user perspective.

Oncall

Oncall duties involve receiving alerts, diagnosing root causes, and restoring services, often guided by predefined SOPs that dictate immediate remedial actions.

Defining Deliverable SLI/SLO/SLA

SLI (Service Level Indicator) measures specific metrics; SLO (Service Level Objective) sets target thresholds; SLA (Service Level Agreement) formalizes commitments and consequences.

Best practices for SLOs include defining clear time windows, using consistent measurement periods, setting realistic expectations, and maintaining safety buffers.

SLA combines SLOs with penalties or rewards, guiding resource allocation and risk management.

Service

A service is any functional offering to customers, delivered by a provider using software and infrastructure.

SLI

SLIs are carefully chosen metrics that answer what to measure, under which system state, and how to aggregate results.

SLO

SLOs translate SLIs into business‑level targets, such as "99% of requests under 500 ms".

SLA

SLAs bind providers and customers to agreed service levels, often linking unmet SLOs to compensation.

Overall, SRE culture emphasizes blameless post‑mortems, continuous learning, and automation to improve reliability and operational efficiency.