Glossary
Standard and log-based metrics
Application Insights' two kinds of built-in metric; standard metrics are pre-aggregated as telemetry is collected and unaffected by sampling, log-based metrics are queries over stored rows that sampling does affect.
Microsoft describes three types of Application Insights metric: standard (pre-aggregated), log-based and custom. The first two cover the same everyday numbers, such as request counts, failure rates and durations, in different ways.
Standard metrics are aggregated while telemetry is collected and stored as time series with a few key dimensions, at one-minute granularity. Because they’re counted before sampling or filtering, their accuracy isn’t affected by either, and Microsoft recommends them for near real-time alerting and responsive dashboards. The metric store allows 5,000 time series per metric in 24 hours.
Log-based metrics are a query-time concept: the chart runs a query over the telemetry rows in the logs. Any property of the rows can be a dimension, but sampling and filtering lower their accuracy unless the query weights rows by itemCount.
The difference shows on any sampled app. At 10% sampling, an hour with 60,000 requests reads 60,000 in the standard metric, about 6,000 in a log query that uses count(), and about 60,000 again in a log query that uses sum(itemCount).
Both cost money: ingesting metrics, log-based or pre-aggregated, is billed by the size of the data. For billing, Microsoft treats all OpenTelemetry metrics as Application Insights custom metrics, and the pre-aggregated copy of a custom metric keeps no dimensions unless you turn on the billed “With dimensions” option.