WHY OBSERVABILITY MATTERS

THE THREE KINDS OF SIGNALS Observability data usually comes in three kinds, and each answers a different question. Logs are time-stamped written records of events, such as "tool search_web was called at 10:42 and returned an error." Logs answer the question of what happened. Metrics are numbers measured over time, such as the number of runs per hour or the average cost per run. Metrics answer the question of how much and how often. Traces are step-by-step records that connect all the events of a single run in order. Traces answer the question of why this particular run went the way it did. Think of a hospital. The patient chart that nurses write is like a log, the monitor that shows heart rate over the day is like a metric, and the full timeline of one patient's treatment, linking tests and decisions, is like a trace. You need all three because each one hides something that the others show. Chapters 28 through 30 teach each in turn.