Site Reliability Engineering (SRE): Phase 1 overview#

Reviewed 28 September 2026. This is a broad overview of public practitioner writing, not a survey of every SRE practice or organization.

What SRE means#

Site Reliability Engineering applies software engineering to the work of operating services: engineers build systems that reduce manual operations while taking responsibility for availability, latency, performance, efficiency, change, monitoring, incident response, and capacity. [1] This is Google’s original framing, and its authors describe SRE as a particular implementation of the wider DevOps ideas of collaboration and automation. [1]

The central question is how reliable a service needs to be for its users, rather than how to pursue perfect uptime. [2] Google argues that reliability, new features, and operating cost must be balanced; a higher reliability target can consume engineering time and limit change without necessarily improving the user experience enough to justify it. [2]

Measuring reliability and deciding what to do#

The common SRE vocabulary is a service level indicator (SLI), a measurement of user-relevant behavior; a service level objective (SLO), a target for that measurement; and an error budget, the allowed gap from perfect performance over a defined period. [3] Google recommends agreeing on SLOs with product, development, and operations stakeholders, then documenting what happens when the budget is exhausted. [3] Without an agreed response, it says, an SLO risks becoming only a reporting metric. [3]

Targets should reflect the experience and importance of different work. [4] For example, Spotify’s event pipeline assigns different delivery SLOs to high-, normal-, and low-priority events, so incident response can favor business-critical data and avoid waking engineers for less urgent delays. [4]

Running and improving services#

SRE includes on-call response, but Google’s model reserves at least half of SRE time for engineering projects and treats repetitive operational work, or toil, as something to reduce. [1] Its workbook recommends examining whether recurring tasks can be eliminated, batched, or automated rather than letting them consume the team. [5]

Incidents are also inputs to system improvement. [6] [7] Google advocates blameless postmortems that examine contributing conditions and produce actionable changes, while Slack describes a response process with trained incident commanders, service-owning responders, and a later review. [6] [7]

Recent practitioner reports continue to emphasize safer changes and limiting the scope of failures. [8] [9] After major outages, Cloudflare reported work on safer configuration rollout, failure isolation, emergency access, and incident management; its SRE team has also described building shared automation so product teams can remediate common failures themselves. [8] [9]

Who owns reliability?#

SRE does not imply one fixed team structure. [10] Google describes infrastructure, product-focused, embedded, and consulting arrangements, and suggests that smaller organizations can begin with part-time advocates before staffing a dedicated team. [10] It also warns that an SRE team with an unbounded charter can become overloaded. [10]

The split between specialists and service owners varies in practice. [7] [9] Slack moved routine on-call responsibility to development teams while its Reliability Engineering group maintained the incident response program; Cloudflare describes an SRE team that builds and operates shared platform services for product teams. [7] [9]

There is a debate about labels and ownership. [11] [1] Charity Majors argues that renaming an operations function “SRE” does not improve outcomes by itself and that organizations should be explicit about the operational work they need and who owns it. [11] Google’s own introduction says its model depends on management support, scarce engineering skills, and protecting time for engineering work. [1]

References#

  1. Google SRE Book: Introduction
  2. Google SRE Book: Embracing Risk
  3. Google SRE Workbook: Implementing SLOs
  4. Spotify Engineering: Spotify's Event Delivery
  5. Google SRE Workbook: Eliminating Toil
  6. Google SRE Workbook: Postmortem Culture
  7. Slack Engineering: All Hands on Deck
  8. Cloudflare: Code Orange: Fail Small is Complete
  9. Cloudflare: Improving Platform Resilience through Automation
  10. Google Cloud: How SRE Teams Are Organized
  11. Charity Majors: Bring Back Ops Pride