rate, irate and increase

Per-second rates that handle resets

rate(counter[5m]) gives the average per-second increase over the window and automatically handles counter resets after restarts. increase(counter[1h]) gives the total increase over the window (rate times the window length). irate uses only the last two samples, which is spiky and better suited to fast-moving graphs than alerts. Use a window of at least four times the scrape interval so each window contains enough samples. Always apply rate before aggregating with sum.

Rates in practice

Common queries.

# requests per second per instance, averaged over 5 minutes
rate(http_requests_total{job="orders-api"}[5m])

# total requests in the last hour
sum(increase(http_requests_total{job="orders-api"}[1h]))

# wrong order: summing counters first breaks reset handling
# rate(sum(http_requests_total)[5m])   <- avoid

rate then sum, never sum then rate

Summing raw counters hides individual resets, so the resulting rate shows false drops and spikes.

Quick check: Why use rate() on a counter instead of graphing it directly?

• Graphing counters is not allowed
• Counters are always zero
• rate() converts gauges to counters
• Counters only grow; rate shows the per-second change and handles resets