prometheus/promql/promqltest/testdata/limit.test
Neeraj Gartia 38bb6ece25
[BUGFIX] PromQL: Fix behaviour of some aggregations with histograms (#15432)
promql: fix some aggregations for histograms

This PR fixes the behaviour of `topk`,`bottomk`, `limitk` and `limit_ratio` with histograms. The fixed behaviour are as follows:
- For `topk` and `bottomk` histograms are ignored and add info annotations added.
- For `limitk` and `limit_ratio` histograms are included in the results(if applicable).

Signed-off-by: Neeraj Gartia <neerajgartia211002@gmail.com>
2024-11-26 19:12:36 +01:00

159 lines
7 KiB
Plaintext

# Tests for limitk
#
# NB: those many `and http_requests` are to ensure that the series _are_ indeed
# a subset of the original series.
load 5m
http_requests{job="api-server", instance="0", group="production"} 0+10x10
http_requests{job="api-server", instance="1", group="production"} 0+20x10
http_requests{job="api-server", instance="0", group="canary"} 0+30x10
http_requests{job="api-server", instance="1", group="canary"} 0+40x10
http_requests{job="api-server", instance="2", group="canary"} 0+50x10
http_requests{job="api-server", instance="3", group="canary"} 0+60x10
http_requests{job="api-server", instance="histogram_1", group="canary"} {{schema:0 sum:10 count:10}}x11
http_requests{job="api-server", instance="histogram_2", group="canary"} {{schema:0 sum:20 count:20}}x11
eval instant at 50m count(limitk by (group) (0, http_requests))
# empty
eval instant at 50m count(limitk by (group) (-1, http_requests))
# empty
# Exercise k==1 special case (as sample is added before the main series loop).
eval instant at 50m count(limitk by (group) (1, http_requests) and http_requests)
{} 2
eval instant at 50m count(limitk by (group) (2, http_requests) and http_requests)
{} 4
eval instant at 50m count(limitk(100, http_requests) and http_requests)
{} 8
# Exercise k==1 special case (as sample is added before the main series loop).
eval instant at 50m count(limitk by (group) (1, http_requests) and http_requests)
{} 2
eval instant at 50m count(limitk by (group) (2, http_requests) and http_requests)
{} 4
eval instant at 50m count(limitk(100, http_requests) and http_requests)
{} 8
# Test for histograms.
# k==1: verify that histogram is included in the result.
eval instant at 50m limitk(1, http_requests{instance="histogram_1"})
{__name__="http_requests", group="canary", instance="histogram_1", job="api-server"} {{count:10 sum:10}}
eval range from 0 to 50m step 5m limitk(1, http_requests{instance="histogram_1"})
{__name__="http_requests", group="canary", instance="histogram_1", job="api-server"} {{count:10 sum:10}}x10
# Histogram is included with mix of floats as well.
eval instant at 50m limitk(8, http_requests{instance=~"(histogram_2|0)"})
{__name__="http_requests", group="canary", instance="histogram_2", job="api-server"} {{count:20 sum:20}}
{__name__="http_requests", group="production", instance="0", job="api-server"} 100
{__name__="http_requests", group="canary", instance="0", job="api-server"} 300
eval range from 0 to 50m step 5m limitk(8, http_requests{instance=~"(histogram_2|0)"})
{__name__="http_requests", group="canary", instance="histogram_2", job="api-server"} {{count:20 sum:20}}x10
{__name__="http_requests", group="production", instance="0", job="api-server"} 0+10x10
{__name__="http_requests", group="canary", instance="0", job="api-server"} 0+30x10
eval instant at 50m count(limitk(2, http_requests{instance=~"histogram_[0-9]"}))
{} 2
eval range from 0 to 50m step 5m count(limitk(2, http_requests{instance=~"histogram_[0-9]"}))
{} 2+0x10
eval instant at 50m count(limitk(1000, http_requests{instance=~"histogram_[0-9]"}))
{} 2
eval range from 0 to 50m step 5m count(limitk(1000, http_requests{instance=~"histogram_[0-9]"}))
{} 2+0x10
# limit_ratio
eval range from 0 to 50m step 5m count(limit_ratio(0.0, http_requests))
# empty
# limitk(2, ...) should always return a 2-count subset of the timeseries (hence the AND'ing)
eval range from 0 to 50m step 5m count(limitk(2, http_requests) and http_requests)
{} 2+0x10
# Tests for limit_ratio.
#
# NB: below 0.5 ratio will depend on some hashing "luck" (also there's no guarantee that
# an integer comes from: total number of series * ratio), as it depends on:
#
# * ratioLimit = [0.0, 1.0]:
# float64(sample.Metric.Hash()) / float64MaxUint64 < Ratio ?
# * ratioLimit = [-1.0, 1.0):
# float64(sample.Metric.Hash()) / float64MaxUint64 >= (1.0 + Ratio) ?
#
# See `AddRatioSample()` in promql/engine.go for more details.
# Half~ish samples: verify we get "near" 3 (of 0.5 * 6).
eval range from 0 to 50m step 5m count(limit_ratio(0.5, http_requests) and http_requests) <= bool (4+1)
{} 1+0x10
eval range from 0 to 50m step 5m count(limit_ratio(0.5, http_requests) and http_requests) >= bool (4-1)
{} 1+0x10
# All samples.
eval range from 0 to 50m step 5m count(limit_ratio(1.0, http_requests) and http_requests)
{} 8+0x10
# All samples.
eval range from 0 to 50m step 5m count(limit_ratio(-1.0, http_requests) and http_requests)
{} 8+0x10
# Capped to 1.0 -> all samples.
eval_warn range from 0 to 50m step 5m count(limit_ratio(1.1, http_requests) and http_requests)
{} 8+0x10
# Capped to -1.0 -> all samples.
eval_warn range from 0 to 50m step 5m count(limit_ratio(-1.1, http_requests) and http_requests)
{} 8+0x10
# Verify that limit_ratio(value) and limit_ratio(1.0-value) return the "complement" of each other.
# Complement below for [0.2, -0.8].
#
# Complement 1of2: `or` should return all samples.
eval range from 0 to 50m step 5m count(limit_ratio(0.2, http_requests) or limit_ratio(-0.8, http_requests))
{} 8+0x10
# Complement 2of2: `and` should return no samples.
eval range from 0 to 50m step 5m count(limit_ratio(0.2, http_requests) and limit_ratio(-0.8, http_requests))
# empty
# Complement below for [0.5, -0.5].
eval range from 0 to 50m step 5m count(limit_ratio(0.5, http_requests) or limit_ratio(-0.5, http_requests))
{} 8+0x10
eval range from 0 to 50m step 5m count(limit_ratio(0.5, http_requests) and limit_ratio(-0.5, http_requests))
# empty
# Complement below for [0.8, -0.2].
eval range from 0 to 50m step 5m count(limit_ratio(0.8, http_requests) or limit_ratio(-0.2, http_requests))
{} 8+0x10
eval range from 0 to 50m step 5m count(limit_ratio(0.8, http_requests) and limit_ratio(-0.2, http_requests))
# empty
# Complement below for [some_ratio, 1.0 - some_ratio], some_ratio derived from time(),
# using a small prime number to avoid rounded ratio values, and a small set of them.
eval range from 0 to 50m step 5m count(limit_ratio(time() % 17/17, http_requests) or limit_ratio(1.0 - (time() % 17/17), http_requests))
{} 8+0x10
eval range from 0 to 50m step 5m count(limit_ratio(time() % 17/17, http_requests) and limit_ratio(1.0 - (time() % 17/17), http_requests))
# empty
# Poor man's normality check: ok (loaded samples follow a nice linearity over labels and time).
# The check giving: 1 (i.e. true).
eval range from 0 to 50m step 5m abs(avg(limit_ratio(0.5, http_requests{instance!~"histogram_[0-9]"})) - avg(limit_ratio(-0.5, http_requests{instance!~"histogram_[0-9]"}))) <= bool stddev(http_requests{instance!~"histogram_[0-9]"})
{} 1+0x10
# All specified histograms are included for r=1.
eval instant at 50m limit_ratio(1, http_requests{instance="histogram_1"})
{__name__="http_requests", group="canary", instance="histogram_1", job="api-server"} {{count:10 sum:10}}
eval range from 0 to 50m step 5m limit_ratio(1, http_requests{instance="histogram_1"})
{__name__="http_requests", group="canary", instance="histogram_1", job="api-server"} {{count:10 sum:10}}x10