Here are the definitions from the official document for rate() and irate(). But if you still don’t quite understand, check the examples below.

In this example, I select all the values I have recorded within the last 1 minute for all time series that have the metric name prometheus_http_requests_total and a handler label set to /metrics:

prometheus_http_requests_total{handler=”/metrics”}[1m]

Output from the Prometheus UI:

# Element
prometheus_http_requests_total{code="200",handler="/metrics",instance="localhost:9090",job="prometheus"}
# ValueHere are the definitions from the official document for rate() and irate(). But if you still don’t quite understand, check the examples below.

In this example, I select all the values I have recorded within the last 1 minute for all time series that have the metric name prometheus_http_requests_total and a handler label set to /metrics:

prometheus_http_requests_total{handler=”/metrics”}[1m]

Output from the Prometheus UI:

# Element
prometheus_http_requests_total{code="200",handler="/metrics",instance="localhost:9090",job="prometheus"}
# Value
878 @1640020217.631
879 @1640020232.605
880 @1640020247.606
881 @1640020262.605

rate()

Now let’s check what will we get when we apply the rate function.

rate(prometheus_http_requests_total{handler=”/metrics”}[1m]) Output from the Prometheus UI:

# Element
{code=”200",handler=”/metrics”,instance=”localhost:9090",job=”prometheus”}
# Value
0.06670520745319518
and here’s the equation taking the first and the last metric point:

irate()

irate(prometheus_http_requests_total{handler="/metrics"}[1m]) Output from the Prometheus UI:

# Element
{code=”200",handler=”/metrics”,instance=”localhost:9090",job=”prometheus”}
# Value
0.06667111140742715
and here’s the equation and here’s the equation taking the last two metric points:

Summary

That why the document points out:

irate should only be used when graphing volatile, fast-moving counters. Use rate for alerts and slow-moving counters, as brief changes in the rate can reset the FOR clause and graphs consisting entirely of rare spikes are hard to read.

• ← Previous Post
• Next Post →