metrics
termops culturecloud
Numbers sampled over time - cheap to store, fast to query, and unable to explain a single request.
Metrics answer how much and how often, which makes them right for alerting and wrong for investigation: an aggregate says the error rate rose and cannot say which customer, which version or which query. The recurring mistake is averaging, because an average hides the tail users actually experience - if the mean response is fast and the ninety-ninth percentile is terrible, a meaningful fraction of people are having a bad time invisibly.
Also known as: time series