Prometheus irate(): The Fastest Way to Track Metric Spikes

Monitoring systems need to understand not just how much something happened — but how fast it's happening right now. Prometheus gives us two tools to track how quickly counters increase: rate() and irate().

 

In this post, we’ll focus on irate(), the function that gives you the instantaneous speed of change based on the two most recent samples.

 

We’ll explore:

 

·      What irate() does and how it works

·      Why we need to specify a time window like [5m]

·      How irate() compares with rate()

·      When to use irate() vs rate()

 

What does irate() do?

irate() (short for instant rate) calculates the per-second rate of increase for counter metrics, based only on the two most recent samples in a time window.

 

In simple terms: How fast is this counter going up right now? By considering most recent two samples.

 

Counters are metrics that only increase, like:

 

·      http_requests_total

·      bytes_sent_total

·      errors_total

 

How does irate() work?

Suppose you're looking at the metric http_requests_total (a counter), and you have these samples:

 

Time	Metric Value
09:59:00	800
09:59:30	850
10:00:00	900
10:00:30	950
10:01:00	1000
10:01:30	1050

 

Let’s say you query at 10:01:30 with:

irate(http_requests_total[1m])

 

1. Time window

The time window is from 10:00:30 to 10:01:30

 

Samples in that window:

10:00:30 — 950

10:01:00 — 1000

10:01:30 — 1050

 

2. irate() selects the last two samples:

10:01:00 — 1000

10:01:30 — 1050

 

3. Calculate rate:

Delta Value = 1050 - 1000 = 50

Delta Time = 30 seconds

Result = 50 / 30 = 1.6667 req/sec

 

Final Result = irate(http_requests_total[1m]) = 1.67 requests/sec

 

This gives a snapshot of how fast the counter is increasing right now.

 

Why do we need to specify a time window like [5m]?

In PromQL, the time window (like [1m] or [5m]) tells Prometheus how far back to look for data points. Even though irate() only uses the two most recent samples, it still needs a time window to know where to search for those samples. Think of [5m] as “look at the last 5 minutes of data to find the most recent two points.”

 

irate() vs rate()

Feature	irate()	rate()
Purpose	Instant rate (spiky/fast)	Average rate (smooth/stable)
Samples	used  Only 2 most recent	All samples within the window
Output	Spiky, real-time changes	Smoothed over the time window
Use case	Dashboards, alerting on spikes	Trends, long-term graphs

 

When to use irate() vs rate()

 

Situation	Use	Why
Real-time dashboards	irate()	Shows current activity
Alerting on sudden changes/spikes	irate()	Detects fast fluctuations
Historical trends	rate()	Gives stable, averaged results
Smoothing out noisy metrics	rate()	Reduces “jumpy” data

 

In summary,

·      irate() gives you the instantaneous speed of increase based on the last two data points.

·      Always specify a time window like [1m], [5m] to tell Prometheus where to look.

·      irate() is best for real-time monitoring and detecting fast changes.

·      For long-term analysis, prefer rate() to get a stable view of trends.

 

 

 

 

Previous                                                    Next                                                    Home