PromQL Range Vector Increase Calculation Functions

1. PromQL Increase Calculation

PromQL can calculate increase amounts, increase rates, etc. using Range Vectors, and provides the functions increase(), rate(), irate(), changes(), delta(), idelta(), and deriv() for this purpose.

[Table 1] shows example data representing http_requests_total values for each time. We actually calculate the results of each function using the data from [Table 1] to see how each function works.

1. increase() function

# --- Query at 10:20 ---
increase(http_requests_total[10m])

# --- Result ---
# 200 - 140 = 60

The increase() function calculates the increase amount for a given time range. It is a function that calculates the difference between the final value and the initial value of a given Range Vector. It is mainly used for Counter-type metrics where values accumulate. In [Query 1], since 10:20 has the value 200 and 10:15 has the value 160, it returns the value 200 - 160 = 40.

2. rate() function

# --- Query at 10:20 ---
rate(http_requests_total[10m])

# --- Result ---
# (200 - 140) / 10m = 60 / 600s = 0.1

The rate() function calculates the increase rate per second for a given time range. It is a function that calculates the difference between the final value and the initial value of a given Range Vector. It is mainly used for Counter-type metrics where values accumulate. In [Query 2], since 10:20 has the value 200 and 10:15 has the value 160, and it is an increase rate per second, it returns the value (200 - 160) / 10m = 60 / 600s = 0.1.

3. irate() function

# --- Query at 10:20 ---
irate(http_requests_total[10m])

# --- Result ---
# (200 - 160) / 10m = 40 / 300s = 0.1333

The irate() function calculates the increase rate per second for a given time range. It is a function that calculates the difference between the final value and the immediately previous value of a given Range Vector. It is similar to the rate() function, but the rate() function calculates the difference between the final value and the initial value, so results differ depending on the time range of the Range Vector, while the irate() function calculates the difference between the final value and the immediately previous value, so it always returns the same result even if the time range of the Range Vector changes. Because of this characteristic, the rate() function shows overall increase rates, while the irate() function better shows sudden increase rates.

In [Query 3], since 10:20 has the value 200 and 10:15 has the value 160, it returns the value (200 - 160) / 10m = 40 / 300s = 0.1333. Although the time range of the given Range Vector is 10m, it uses the value from 10:15 instead of 10:10 for calculation because it uses the immediately previous value of the final value.

4. changes() function

# --- Query at 10:20 ---
changes(http_requests_total[10m])

# --- Result ---
# 140 -> 160 -> 200 = 3 (changes)

The changes() function calculates the number of changes for a given time range. It is a function that calculates the number of changes for each value of a given Range Vector. In [Query 4], since 10:10 has the value 140, 10:15 has the value 160, and 10:20 has the value 200, it changed 3 times, so it returns the value 3.

5. delta() function

# --- Query at 10:20 ---
delta(http_requests_total[10m])

# --- Result ---
# (200 - 140) / 2 = 60 / 2 = 30

The delta() function calculates the average increase amount for a given time range. In [Query 5], since 10:20 has the value 200 and 10:15 has the value 160, and there are 2 intervals, it returns the value (200 - 140) / 2 = 60 / 2 = 30.

6. idelta() function

# --- Query at 10:20 ---
idelta(http_requests_total[10m])

# --- Result ---
# 200 - 160 = 40

The idelta() function calculates the increase amount of the last interval for a given time range. In [Query 6], since 10:20 has the value 200 and 10:15 has the value 160, it returns the value 200 - 160 = 40. Although the time range of the given Range Vector is 10m, it uses the value from 10:15 instead of 10:10 for calculation because it uses the immediately previous value of the final value.

7. deriv() function

# --- Query at 10:20 ---
deriv(http_requests_total[10m])

# --- Result ---
# simple linear regression = 2.67

The deriv() function calculates the linear regression coefficient for a given time range. In [Query 7], since 10:10 has the value 140, 10:15 has the value 160, and 10:20 has the value 200, it calculates the linear regression coefficient for 3 values and returns the value 2.67.

2. References

• PromQL Functions : https://prometheus.io/docs/prometheus/latest/querying/functions/
• rate() vs irate() : https://techannotation.wordpress.com/2021/07/19/irate-vs-rate-whatre-they-telling-you/