rate and irate

The rate() Function

rate() calculates the per-second average increase of a counter over a time range. It is the standard function for converting counters into rates. It handles counter resets gracefully by accounting for drops in value as new counters.

# Requests per second over last 5 minutes
rate(http_requests_total[5m])

# Error rate per second
rate(http_requests_total{status=~"5.."}[5m])

# Bytes per second throughput
rate(network_bytes_total[5m])

The irate() Function

irate() calculates the per-second instantaneous rate using only the last two data points in the range. It responds faster to changes than rate(), but it is more sensitive to noise and scrape jitter.

# Slow and smooth — good for alerts
rate(http_requests_total[5m])

# Fast and responsive — good for dashboards
irate(http_requests_total[5m])

The increase() Function

increase() returns the total increase of a counter over a time range, in the counter's original units. It is equivalent to rate() multiplied by the range duration. Use it when you need absolute counts rather than rates.

# Total requests in last 1 hour
increase(http_requests_total[1h])

# Total errors in last 24 hours
increase(http_requests_total{status=~"5.."}[24h])

# Equivalent to rate() * 3600 for 1 hour
rate(http_requests_total[1h]) * 3600

Counter Reset Handling

When a counter decreases between scrapes, Prometheus assumes a reset (process restart, target re-deployment). rate() and increase() detect resets by adding the current value to the previous value instead of counting the decrease as a negative change.

Sample 1: 100
Sample 2: 150  (+50)
Sample 3: 120  — reset detected, treated as 120, not -30
Sample 4: 200  (+80 from 120)

Choosing the Right Window

The range window controls smoothing. Short windows are responsive but noisy. Long windows are smooth but hide short spikes. The right choice depends on whether you are alerting, graphing, or performing capacity planning.

Common Patterns

# QPS per endpoint
sum by(endpoint) (rate(http_requests_total[5m]))

# Error ratio (5xx / total)
sum(rate(http_requests_total{status=~"5.."}[5m]))
  /
sum(rate(http_requests_total[5m]))

# Request rate per second, only endpoints with traffic
sum by(endpoint) (rate(http_requests_total[5m]))
  > 0.1