Histogram
Recording numerical data that's inherent to each event in a system can be expensive, so some sort of pre-aggregation is usually needed to conserve at least partial information about what happened. However, by pre-calculating aggregations on each instance (such as average since process start, rolling window, exponentially weighted, and so on), a lot of granularity is lost and some calculations can be computationally costly. Adding to this, a lot of pre-aggregations can't generally be re-aggregated without losing meaning—the average of a thousand pre-calculated 95th percentiles has no statistical meaning. Similarly, having the 99th percentile of request latency collected from each instance of a given cluster (for example) gives you ...
Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.
Read now
Unlock full access