SLO / SLA / SLI
also: service level objective · error budget
The reliability targets: an SLI measures it, an SLO is your goal for it, an SLA is the contract with penalties.
SLI, SLO, and SLA are the reliability trio. An SLI (indicator) is a measured signal like the fraction of successful requests; an SLO (objective) is your internal target for it, say 99.9%; an SLA (agreement) is the external contract with a customer, with penalties if you miss it.
Worked example: if your SLI is request success rate and your SLO is 99.9% over 30 days, that allows about 43 minutes of downtime a month — the error budget you can spend on risky deploys. Gotcha: set the SLO below the SLA (never promise a customer exactly your internal target), and choose SLIs that reflect user experience — a server that is “up” but returning slow errors can meet a naive uptime SLI while failing users.