Skip to content
Nubent

/Glossary · DevOps

What is SRE?

SRE (Site Reliability Engineering) is a discipline that applies software engineering practices to operating systems so they are reliable and scalable. It sets measurable reliability targets and reduces manual operational work through automation.

How SRE works

SRE starts by defining what it means for a service to work well for the user. SLIs are measurable indicators, such as latency or error rate; SLOs are targets on those indicators, for example 99.9% of requests succeeding over a month. The gap to 100% is the error budget.

While budget remains, the team can ship and experiment. Once it runs out, reliability becomes the priority. That turns a subjective argument between product and operations into a data-driven decision. Alerts are set on the budget burn rate, not on every technical metric.

When SRE makes sense

It makes sense when incidents keep recurring, when the team gets so many alerts it stops looking at them, or when there is no agreement on the level of availability the business actually needs. Also when customers demand service level commitments and there is no objective way to measure them.

You don't need a dedicated SRE team to get started. Defining two or three SLOs for critical user flows, instrumenting them and reviewing budget burn every week already changes how you prioritize. A dedicated team makes sense when scale and criticality justify it.

Best practices and common mistakes on AWS

On AWS, CloudWatch Application Signals lets you define SLOs on services instrumented with OpenTelemetry. Add composite alarms, automated runbooks in Systems Manager and resilience testing with AWS Fault Injection Service. After every incident, write a blameless postmortem with concrete action items.

Frequent mistakes: setting 99.99% SLOs with no need for them and no architecture to back them up, measuring availability from the infrastructure instead of from the user, and alerting on CPU usage instead of real impact. Another is writing postmortems whose action items never get done.

/FAQ

Questions about SRE

Another question? Write to hello@nubent.com and an engineer will answer.

What is the difference between an SLA, an SLO and an SLI?

The SLI is the metric, for example the percentage of successful requests. The SLO is the internal target for that metric. The SLA is the contractual commitment to the customer, with consequences if it is missed, and it is usually less strict than the SLO.

Does SRE replace DevOps?

No. SRE can be seen as a concrete implementation of DevOps principles focused on reliability. DevOps defines the culture of collaboration and continuous delivery; SRE adds specific tools such as SLOs, error budgets and the reduction of repetitive manual work.

What is an error budget?

It is the amount of failure that is acceptable within a period according to the SLO. With a 99.9% monthly SLO, the budget comes to about 43 minutes of downtime. While budget remains, you can prioritize speed; once it runs out, stability comes first.

Next step

What should we build next on AWS?

30 minutes with an engineer. Concrete next steps, whether or not you work with us.