"Operation xyz completed in 15.445 seconds." This can be expressed as a metric (or a trace) so that the "operation_xyz_completed" is a metric and 15.445 seconds

"Operation xyz completed in 15.445 seconds." This can be expressed as a metric (or a trace) so that the "operation_xyz_completed" is a metric and 15.445 seconds is evaluated as a metric data point. The result is an easily chart to graph the average, p99, whatever of the operation to gauge if this is normal or exceptional. It's dead simple to alert on metrics as well often, so it helps to unlock alerting. Log alerts are valid but often more limiting without a bunch of parsing or being really naive.