In PromQL, range vector functions like max_over_time() compute an aggregate value (in this case, the maximum) over all samples within a specified time range. The function ignores NaN (Not-a-Number) values when computing the result.

Given the range vector foo[5m] containing samples [1, 1, NaN], the maximum value among the valid numeric samples is 1. Therefore, max_over_time(foo[5m]) returns 1.

Prometheus functions handle missing or invalid data points gracefully---ignoring NaN ensures stable calculations even when intermittent collection issues or resets occur. The function only errors if the selector is syntactically invalid or if no numeric samples exist at all.

Verified from Prometheus documentation -- PromQL Range Vector Functions, Aggregation Over Time Functions, and Handling NaN Values in PromQL sections.