mudhorn/prometheus
1
0
Fork 0
mirror of https://github.com/prometheus/prometheus.git synced 2024-11-15 10:04:07 -08:00
Code Issues Actions 8 Packages Projects Releases Wiki Activity
prometheus/promql/promqltest/testdata/histograms.test
György Krajcsovits 06a8886b94 Native histograms: define behavior when rate is null. 
Histogram quantile returns NaN in this case, which might be
surprising, so add a unit test that clarifies that this is
intentional.

Signed-off-by: György Krajcsovits <gyorgy.krajcsovits@grafana.com>
2024-08-12 10:40:21 +02:00

514 lines
18 KiB
Plaintext
Raw Blame History

# Two histograms with 4 buckets each (x_sum and x_count not included,
# only buckets). Lowest bucket for one histogram < 0, for the other >
# 0. They have the same name, just separated by label. Not useful in
# practice, but can happen (if clients change bucketing), and the
# server has to cope with it.
# Test histogram.
load_with_nhcb 5m
testhistogram_bucket{le="0.1", start="positive"} 0+5x10
testhistogram_bucket{le=".2", start="positive"} 0+7x10
testhistogram_bucket{le="1e0", start="positive"} 0+11x10
testhistogram_bucket{le="+Inf", start="positive"} 0+12x10
testhistogram_bucket{le="-.2", start="negative"} 0+1x10
testhistogram_bucket{le="-0.1", start="negative"} 0+2x10
testhistogram_bucket{le="0.3", start="negative"} 0+2x10
testhistogram_bucket{le="+Inf", start="negative"} 0+3x10
# Another test histogram, where q(1/6), q(1/2), and q(5/6) are each in
# the middle of a bucket and should therefore be 1, 3, and 5,
# respectively.
load_with_nhcb 5m
testhistogram2_bucket{le="0"} 0+0x10
testhistogram2_bucket{le="2"} 0+1x10
testhistogram2_bucket{le="4"} 0+2x10
testhistogram2_bucket{le="6"} 0+3x10
testhistogram2_bucket{le="+Inf"} 0+3x10
# Another test histogram, this time without any observations in the +Inf bucket.
# This enables a meaningful calculation of standard deviation and variance.
load_with_nhcb 5m
testhistogram3_bucket{le="0", start="positive"} 0+0x10
testhistogram3_bucket{le="0.1", start="positive"} 0+5x10
testhistogram3_bucket{le=".2", start="positive"} 0+7x10
testhistogram3_bucket{le="1e0", start="positive"} 0+11x10
testhistogram3_bucket{le="+Inf", start="positive"} 0+11x10
testhistogram3_sum{start="positive"} 0+33x10
testhistogram3_count{start="positive"} 0+11x10
testhistogram3_bucket{le="-.25", start="negative"} 0+0x10
testhistogram3_bucket{le="-.2", start="negative"} 0+1x10
testhistogram3_bucket{le="-0.1", start="negative"} 0+2x10
testhistogram3_bucket{le="0.3", start="negative"} 0+2x10
testhistogram3_bucket{le="+Inf", start="negative"} 0+2x10
testhistogram3_sum{start="negative"} 0+8x10
testhistogram3_count{start="negative"} 0+2x10
# Now a more realistic histogram per job and instance to test aggregation.
load_with_nhcb 5m
request_duration_seconds_bucket{job="job1", instance="ins1", le="0.1"} 0+1x10
request_duration_seconds_bucket{job="job1", instance="ins1", le="0.2"} 0+3x10
request_duration_seconds_bucket{job="job1", instance="ins1", le="+Inf"} 0+4x10
request_duration_seconds_bucket{job="job1", instance="ins2", le="0.1"} 0+2x10
request_duration_seconds_bucket{job="job1", instance="ins2", le="0.2"} 0+5x10
request_duration_seconds_bucket{job="job1", instance="ins2", le="+Inf"} 0+6x10
request_duration_seconds_bucket{job="job2", instance="ins1", le="0.1"} 0+3x10
request_duration_seconds_bucket{job="job2", instance="ins1", le="0.2"} 0+4x10
request_duration_seconds_bucket{job="job2", instance="ins1", le="+Inf"} 0+6x10
request_duration_seconds_bucket{job="job2", instance="ins2", le="0.1"} 0+4x10
request_duration_seconds_bucket{job="job2", instance="ins2", le="0.2"} 0+7x10
request_duration_seconds_bucket{job="job2", instance="ins2", le="+Inf"} 0+9x10
# Different le representations in one histogram.
load_with_nhcb 5m
mixed_bucket{job="job1", instance="ins1", le="0.1"} 0+1x10
mixed_bucket{job="job1", instance="ins1", le="0.2"} 0+1x10
mixed_bucket{job="job1", instance="ins1", le="2e-1"} 0+1x10
mixed_bucket{job="job1", instance="ins1", le="2.0e-1"} 0+1x10
mixed_bucket{job="job1", instance="ins1", le="+Inf"} 0+4x10
mixed_bucket{job="job1", instance="ins2", le="+inf"} 0+0x10
mixed_bucket{job="job1", instance="ins2", le="+Inf"} 0+0x10
# Test histogram_count.
eval instant at 50m histogram_count(testhistogram3)
{start="positive"} 110
{start="negative"} 20
# Classic way of accessing the count still works.
eval instant at 50m testhistogram3_count
testhistogram3_count{start="positive"} 110
testhistogram3_count{start="negative"} 20
# Test histogram_sum.
eval instant at 50m histogram_sum(testhistogram3)
{start="positive"} 330
{start="negative"} 80
# Classic way of accessing the sum still works.
eval instant at 50m testhistogram3_sum
testhistogram3_sum{start="positive"} 330
testhistogram3_sum{start="negative"} 80
# Test histogram_avg. This has no classic equivalent.
eval instant at 50m histogram_avg(testhistogram3)
{start="positive"} 3
{start="negative"} 4
# Test histogram_stddev. This has no classic equivalent.
eval instant at 50m histogram_stddev(testhistogram3)
{start="positive"} 2.8189265757336734
{start="negative"} 4.182715937754936
# Test histogram_stdvar. This has no classic equivalent.
eval instant at 50m histogram_stdvar(testhistogram3)
{start="positive"} 7.946347039377573
{start="negative"} 17.495112615949154
# Test histogram_fraction.
eval instant at 50m histogram_fraction(0, 0.2, testhistogram3)
{start="positive"} 0.6363636363636364
{start="negative"} 0
eval instant at 50m histogram_fraction(0, 0.2, rate(testhistogram3[5m]))
{start="positive"} 0.6363636363636364
{start="negative"} 0
# In the classic histogram, we can access the corresponding bucket (if
# it exists) and divide by the count to get the same result.
eval instant at 50m testhistogram3_bucket{le=".2"} / ignoring(le) testhistogram3_count
{start="positive"} 0.6363636363636364
eval instant at 50m rate(testhistogram3_bucket{le=".2"}[5m]) / ignoring(le) rate(testhistogram3_count[5m])
{start="positive"} 0.6363636363636364
# Test histogram_quantile, native and classic.
eval instant at 50m histogram_quantile(0, testhistogram3)
{start="positive"} 0
{start="negative"} -0.25
eval instant at 50m histogram_quantile(0, testhistogram3_bucket)
{start="positive"} 0
{start="negative"} -0.25
eval instant at 50m histogram_quantile(0.25, testhistogram3)
{start="positive"} 0.055
{start="negative"} -0.225
eval instant at 50m histogram_quantile(0.25, testhistogram3_bucket)
{start="positive"} 0.055
{start="negative"} -0.225
eval instant at 50m histogram_quantile(0.5, testhistogram3)
{start="positive"} 0.125
{start="negative"} -0.2
eval instant at 50m histogram_quantile(0.5, testhistogram3_bucket)
{start="positive"} 0.125
{start="negative"} -0.2
eval instant at 50m histogram_quantile(0.75, testhistogram3)
{start="positive"} 0.45
{start="negative"} -0.15
eval instant at 50m histogram_quantile(0.75, testhistogram3_bucket)
{start="positive"} 0.45
{start="negative"} -0.15
eval instant at 50m histogram_quantile(1, testhistogram3)
{start="positive"} 1
{start="negative"} -0.1
eval instant at 50m histogram_quantile(1, testhistogram3_bucket)
{start="positive"} 1
{start="negative"} -0.1
# Quantile too low.
eval_warn instant at 50m histogram_quantile(-0.1, testhistogram)
{start="positive"} -Inf
{start="negative"} -Inf
eval_warn instant at 50m histogram_quantile(-0.1, testhistogram_bucket)
{start="positive"} -Inf
{start="negative"} -Inf
# Quantile too high.
eval_warn instant at 50m histogram_quantile(1.01, testhistogram)
{start="positive"} +Inf
{start="negative"} +Inf
eval_warn instant at 50m histogram_quantile(1.01, testhistogram_bucket)
{start="positive"} +Inf
{start="negative"} +Inf
# Quantile invalid.
eval_warn instant at 50m histogram_quantile(NaN, testhistogram)
{start="positive"} NaN
{start="negative"} NaN
eval_warn instant at 50m histogram_quantile(NaN, testhistogram_bucket)
{start="positive"} NaN
{start="negative"} NaN
# Quantile value in lowest bucket.
eval instant at 50m histogram_quantile(0, testhistogram)
{start="positive"} 0
{start="negative"} -0.2
eval instant at 50m histogram_quantile(0, testhistogram_bucket)
{start="positive"} 0
{start="negative"} -0.2
# Quantile value in highest bucket.
eval instant at 50m histogram_quantile(1, testhistogram)
{start="positive"} 1
{start="negative"} 0.3
eval instant at 50m histogram_quantile(1, testhistogram_bucket)
{start="positive"} 1
{start="negative"} 0.3
# Finally some useful quantiles.
eval instant at 50m histogram_quantile(0.2, testhistogram)
{start="positive"} 0.048
{start="negative"} -0.2
eval instant at 50m histogram_quantile(0.2, testhistogram_bucket)
{start="positive"} 0.048
{start="negative"} -0.2
eval instant at 50m histogram_quantile(0.5, testhistogram)
{start="positive"} 0.15
{start="negative"} -0.15
eval instant at 50m histogram_quantile(0.5, testhistogram_bucket)
{start="positive"} 0.15
{start="negative"} -0.15
eval instant at 50m histogram_quantile(0.8, testhistogram)
{start="positive"} 0.72
{start="negative"} 0.3
eval instant at 50m histogram_quantile(0.8, testhistogram_bucket)
{start="positive"} 0.72
{start="negative"} 0.3
# More realistic with rates.
eval instant at 50m histogram_quantile(0.2, rate(testhistogram[5m]))
{start="positive"} 0.048
{start="negative"} -0.2
eval instant at 50m histogram_quantile(0.2, rate(testhistogram_bucket[5m]))
{start="positive"} 0.048
{start="negative"} -0.2
eval instant at 50m histogram_quantile(0.5, rate(testhistogram[5m]))
{start="positive"} 0.15
{start="negative"} -0.15
eval instant at 50m histogram_quantile(0.5, rate(testhistogram_bucket[5m]))
{start="positive"} 0.15
{start="negative"} -0.15
eval instant at 50m histogram_quantile(0.8, rate(testhistogram[5m]))
{start="positive"} 0.72
{start="negative"} 0.3
eval instant at 50m histogram_quantile(0.8, rate(testhistogram_bucket[5m]))
{start="positive"} 0.72
{start="negative"} 0.3
# Want results exactly in the middle of the bucket.
eval instant at 7m histogram_quantile(1./6., testhistogram2)
{} 1
eval instant at 7m histogram_quantile(1./6., testhistogram2_bucket)
{} 1
eval instant at 7m histogram_quantile(0.5, testhistogram2)
{} 3
eval instant at 7m histogram_quantile(0.5, testhistogram2_bucket)
{} 3
eval instant at 7m histogram_quantile(5./6., testhistogram2)
{} 5
eval instant at 7m histogram_quantile(5./6., testhistogram2_bucket)
{} 5
eval instant at 47m histogram_quantile(1./6., rate(testhistogram2[15m]))
{} 1
eval instant at 47m histogram_quantile(1./6., rate(testhistogram2_bucket[15m]))
{} 1
eval instant at 47m histogram_quantile(0.5, rate(testhistogram2[15m]))
{} 3
eval instant at 47m histogram_quantile(0.5, rate(testhistogram2_bucket[15m]))
{} 3
eval instant at 47m histogram_quantile(5./6., rate(testhistogram2[15m]))
{} 5
eval instant at 47m histogram_quantile(5./6., rate(testhistogram2_bucket[15m]))
{} 5
# Aggregated histogram: Everything in one. Note how native histograms
# don't require aggregation by le.
eval instant at 50m histogram_quantile(0.3, sum(rate(request_duration_seconds[5m])))
{} 0.075
eval instant at 50m histogram_quantile(0.3, sum(rate(request_duration_seconds_bucket[5m])) by (le))
{} 0.075
eval instant at 50m histogram_quantile(0.5, sum(rate(request_duration_seconds[5m])))
{} 0.1277777777777778
eval instant at 50m histogram_quantile(0.5, sum(rate(request_duration_seconds_bucket[5m])) by (le))
{} 0.1277777777777778
# Aggregated histogram: Everything in one. Now with avg, which does not change anything.
eval instant at 50m histogram_quantile(0.3, avg(rate(request_duration_seconds[5m])))
{} 0.075
eval instant at 50m histogram_quantile(0.3, avg(rate(request_duration_seconds_bucket[5m])) by (le))
{} 0.075
eval instant at 50m histogram_quantile(0.5, avg(rate(request_duration_seconds[5m])))
{} 0.12777777777777778
eval instant at 50m histogram_quantile(0.5, avg(rate(request_duration_seconds_bucket[5m])) by (le))
{} 0.12777777777777778
# Aggregated histogram: By instance.
eval instant at 50m histogram_quantile(0.3, sum(rate(request_duration_seconds[5m])) by (instance))
{instance="ins1"} 0.075
{instance="ins2"} 0.075
eval instant at 50m histogram_quantile(0.3, sum(rate(request_duration_seconds_bucket[5m])) by (le, instance))
{instance="ins1"} 0.075
{instance="ins2"} 0.075
eval instant at 50m histogram_quantile(0.5, sum(rate(request_duration_seconds[5m])) by (instance))
{instance="ins1"} 0.1333333333
{instance="ins2"} 0.125
eval instant at 50m histogram_quantile(0.5, sum(rate(request_duration_seconds_bucket[5m])) by (le, instance))
{instance="ins1"} 0.1333333333
{instance="ins2"} 0.125
# Aggregated histogram: By job.
eval instant at 50m histogram_quantile(0.3, sum(rate(request_duration_seconds[5m])) by (job))
{job="job1"} 0.1
{job="job2"} 0.0642857142857143
eval instant at 50m histogram_quantile(0.3, sum(rate(request_duration_seconds_bucket[5m])) by (le, job))
{job="job1"} 0.1
{job="job2"} 0.0642857142857143
eval instant at 50m histogram_quantile(0.5, sum(rate(request_duration_seconds[5m])) by (job))
{job="job1"} 0.14
{job="job2"} 0.1125
eval instant at 50m histogram_quantile(0.5, sum(rate(request_duration_seconds_bucket[5m])) by (le, job))
{job="job1"} 0.14
{job="job2"} 0.1125
# Aggregated histogram: By job and instance.
eval instant at 50m histogram_quantile(0.3, sum(rate(request_duration_seconds[5m])) by (job, instance))
{instance="ins1", job="job1"} 0.11
{instance="ins2", job="job1"} 0.09
{instance="ins1", job="job2"} 0.06
{instance="ins2", job="job2"} 0.0675
eval instant at 50m histogram_quantile(0.3, sum(rate(request_duration_seconds_bucket[5m])) by (le, job, instance))
{instance="ins1", job="job1"} 0.11
{instance="ins2", job="job1"} 0.09
{instance="ins1", job="job2"} 0.06
{instance="ins2", job="job2"} 0.0675
eval instant at 50m histogram_quantile(0.5, sum(rate(request_duration_seconds[5m])) by (job, instance))
{instance="ins1", job="job1"} 0.15
{instance="ins2", job="job1"} 0.1333333333333333
{instance="ins1", job="job2"} 0.1
{instance="ins2", job="job2"} 0.1166666666666667
eval instant at 50m histogram_quantile(0.5, sum(rate(request_duration_seconds_bucket[5m])) by (le, job, instance))
{instance="ins1", job="job1"} 0.15
{instance="ins2", job="job1"} 0.1333333333333333
{instance="ins1", job="job2"} 0.1
{instance="ins2", job="job2"} 0.1166666666666667
# The unaggregated histogram for comparison. Same result as the previous one.
eval instant at 50m histogram_quantile(0.3, rate(request_duration_seconds[5m]))
{instance="ins1", job="job1"} 0.11
{instance="ins2", job="job1"} 0.09
{instance="ins1", job="job2"} 0.06
{instance="ins2", job="job2"} 0.0675
eval instant at 50m histogram_quantile(0.3, rate(request_duration_seconds_bucket[5m]))
{instance="ins1", job="job1"} 0.11
{instance="ins2", job="job1"} 0.09
{instance="ins1", job="job2"} 0.06
{instance="ins2", job="job2"} 0.0675
eval instant at 50m histogram_quantile(0.5, rate(request_duration_seconds[5m]))
{instance="ins1", job="job1"} 0.15
{instance="ins2", job="job1"} 0.13333333333333333
{instance="ins1", job="job2"} 0.1
{instance="ins2", job="job2"} 0.11666666666666667
eval instant at 50m histogram_quantile(0.5, rate(request_duration_seconds_bucket[5m]))
{instance="ins1", job="job1"} 0.15
{instance="ins2", job="job1"} 0.13333333333333333
{instance="ins1", job="job2"} 0.1
{instance="ins2", job="job2"} 0.11666666666666667
# All NHCBs summed into one.
eval instant at 50m sum(request_duration_seconds)
{} {{schema:-53 count:250 custom_values:[0.1 0.2] buckets:[100 90 60]}}
# A histogram with nonmonotonic bucket counts. This may happen when recording
# rule evaluation or federation races scrape ingestion, causing some buckets
# counts to be derived from fewer samples.
load 5m
nonmonotonic_bucket{le="0.1"} 0+2x10
nonmonotonic_bucket{le="1"} 0+1x10
nonmonotonic_bucket{le="10"} 0+5x10
nonmonotonic_bucket{le="100"} 0+4x10
nonmonotonic_bucket{le="1000"} 0+9x10
nonmonotonic_bucket{le="+Inf"} 0+8x10
# Nonmonotonic buckets
eval instant at 50m histogram_quantile(0.01, nonmonotonic_bucket)
{} 0.0045
eval instant at 50m histogram_quantile(0.5, nonmonotonic_bucket)
{} 8.5
eval instant at 50m histogram_quantile(0.99, nonmonotonic_bucket)
{} 979.75
# Buckets with different representations of the same upper bound.
eval instant at 50m histogram_quantile(0.5, rate(mixed_bucket[5m]))
{instance="ins1", job="job1"} 0.15
{instance="ins2", job="job1"} NaN
eval instant at 50m histogram_quantile(0.5, rate(mixed[5m]))
{instance="ins1", job="job1"} 0.2
{instance="ins2", job="job1"} NaN
eval instant at 50m histogram_quantile(0.75, rate(mixed_bucket[5m]))
{instance="ins1", job="job1"} 0.2
{instance="ins2", job="job1"} NaN
eval instant at 50m histogram_quantile(1, rate(mixed_bucket[5m]))
{instance="ins1", job="job1"} 0.2
{instance="ins2", job="job1"} NaN
load_with_nhcb 5m
empty_bucket{le="0.1", job="job1", instance="ins1"} 0x10
empty_bucket{le="0.2", job="job1", instance="ins1"} 0x10
empty_bucket{le="+Inf", job="job1", instance="ins1"} 0x10
eval instant at 50m histogram_quantile(0.2, rate(empty_bucket[5m]))
{instance="ins1", job="job1"} NaN
# Load a duplicate histogram with a different name to test failure scenario on multiple histograms with the same label set.
# https://github.com/prometheus/prometheus/issues/9910
load_with_nhcb 5m
request_duration_seconds2_bucket{job="job1", instance="ins1", le="0.1"} 0+1x10
request_duration_seconds2_bucket{job="job1", instance="ins1", le="0.2"} 0+3x10
request_duration_seconds2_bucket{job="job1", instance="ins1", le="+Inf"} 0+4x10
eval_fail instant at 50m histogram_quantile(0.99, {__name__=~"request_duration_seconds\\d*_bucket"})
eval_fail instant at 50m histogram_quantile(0.99, {__name__=~"request_duration_seconds\\d*"})
# Histogram with constant buckets.
load_with_nhcb 1m
const_histogram_bucket{le="0.0"} 1 1 1 1 1
const_histogram_bucket{le="1.0"} 1 1 1 1 1
const_histogram_bucket{le="2.0"} 1 1 1 1 1
const_histogram_bucket{le="+Inf"} 1 1 1 1 1
# There is no change to the bucket count over time, thus rate is 0 in each bucket.
eval instant at 5m rate(const_histogram_bucket[5m])
{le="0.0"} 0
{le="1.0"} 0
{le="2.0"} 0
{le="+Inf"} 0
# There is no change to the bucket count over time, thus rate is 0 in each bucket.
# However native histograms do not represent empty buckets, so here the zeros are implicit.
eval instant at 5m rate(const_histogram[5m])
{} {{schema:-53 sum:0 count:0 custom_values:[0.0 1.0 2.0]}}
# Zero buckets mean no observations, so there is no value that observations fall bellow,
# which means that any quantile is a NaN.
eval instant at 5m histogram_quantile(1.0, sum by (le) (rate(const_histogram_bucket[5m])))
{} NaN
# Zero buckets mean no observations, so there is no value that observations fall bellow,
# which means that any quantile is a NaN.
eval instant at 5m histogram_quantile(1.0, sum(rate(const_histogram[5m])))
{} NaN
Reference in a new issue View git blame Copy permalink