Work out availability from error rate instantly with clear inputs, formula shown and shareable results.
Request-based availability is successful requests over total requests, which reflects what users experienced far better than counting minutes the service was down. Converting the error ratio into equivalent full-outage minutes makes it comparable with a time-based SLA, and the nines figure is just the negative base-ten logarithm of the error rate.
Request availability
availability = 1 - failed / total; equivalent downtime = error rate x period minutes; nines = -log10(error rate)
Request-based, for user-facing services. A 30-second outage during peak traffic harms far more users than an hour at 3 am, and time-based measurement treats them as opposite.
Define it as part of the SLI: usually 5xx responses and timeouts, excluding client errors. Whether to count slow-but-successful responses is a separate latency SLI.