Time series aggregation

Summarizes metric data over a window to observe trends or trigger alerts.

Creating dashboard panels, setting up alert rules, analyzing service latency over the last 5 minutes.

Prometheus monitoring stacks, Grafana dashboards, SRE tooling, cloud-native observability.

avg_over_time(metric[duration])

avg_over_time([...])

metric: selector string (e.g., http_requests_total), duration: time duration string like '5m', '1h'

• sum_over_time
• max_over_time
• rate
• increase

• avg over time can be approximated by avg without brackets for instant vectors
• or using recording rules

Using avg_over_time with an instant vector (missing brackets) causing type errors; confusing with avg aggregation over labels.

avg (instant aggregation) – computes average across label dimensions at a single point; avg_over_time averages over time.

Coming from traditional SQL: may confuse avg_over_time with SQL AVG over time windows; note the bracket syntax is Prometheus-specific.

• time_series_aggregation
• rate_over_time
• increase_over_time

The duration must be a valid Prometheus duration; the function returns a scalar if the range vector resolves to a single series, otherwise a vector of values per series.

Enables concise expression of temporal trends, reduces need for complex recording rules.

Think ‘average over time’ – picture a graph line smoothing.

If the range vector resolves to a single series, the result is a scalar; otherwise a vector of averaged values per series

Precompute with recording rules for efficiency, or combine with sum_over_time for custom aggregations