Reliability is a system’s ability to remain responsive when failures occur. In a distributed system, this means continuing to provide its intended function even though components may crash or interactions may fail. section 1.0 section 1.4.2
Reliability is not a property of one component alone. It emerges from multiple components and their interactions, because a single component cannot survive every inevitable failure. section 1.4.3 section 1.7.2
Judge reliability through responsiveness:
- A service-level indicator (SLI) quantitatively observes system behavior. section 1.4.3
- A service-level objective (SLO) states whether that observed behavior meets the desired objective. section 1.4.3
- An error rate measures observations that miss their objectives, while an error budget sets the acceptable upper limit. section 1.4.3
A distributed team
A company’s reliability cannot be attributed to one employee: that person may be absent or unable to complete unlimited work. Reliable behavior instead emerges from employees acting together, just as reliable systems are built from unreliable components. section 1.7.2
Reliability guarantees are application-specific. Each system must decide which behavior is desirable, tolerable, or unacceptable, then express those decisions through objectives and error budgets. section 1.4.3
