Interpretation of Latency Percentiles on Service Details
- Pralav Dessai
In APM screens, latency percentile information is available in the following screens
Service List: This displays the latency percentile values for each service, for the overall selected time range
Service Details: This displays multiple time-series charts as well as tables with latency percentiles
Overall latency percentile charts: these charts represent overall service latencies. The time series is bucketed based on overall time range selected. The percentile values in the time series bucket are representative of just the points within the bucket. Hence it is possible that there is fluctuation of the percentile values across the buckets in the time series.
Latency percentile charts broken-down by span-name, deployment and other dimensions: The service detail page also displays charts broken down into smaller components. These time series charts are similar in behavior to the overall latency percentile charts.
Latency percentile tables broken-down by span-name, deployment and other dimensions: These table represent the percentile values over the overall selected time range.
Special behavior in expanded-charts
When latency percentile charts are expanded, the screens display an additional table below the chart. This table should be interpreted as a description of the chart, and does not always represent the actual aggregates of the underlying data. For example, in case of p99 latency, the table below shows values for average and max. These describe the values of the chart data and do not provide information on the underlying metric. For example, the average and max of p99 values over multiple buckets does not have statistical significance.