prometheus/promql/promqltest
beorn7 c46074f4dd promql: make avg aggregation more precise and less expensive
The basic idea here is that the previous code was always doing
incremental calculation of the mean value, which is more costly and
can be less precise. It protects against overflows, but in most cases,
an overflow doesn't happen anyway.

The other idea applied here is to expand on #14074, where Kahan
summation was applied to sum().

With this commit, the average is calculated in a conventional way
(adding everything up and divide in the end) as long as the sum isn't
overflowing float64. This is combined with Kahan summation so that the
avg aggregation, in most cases, is really equivalent to the sum
aggregation with a following division (which is the user's expectation
as avg is supposed to be syntactic sugar for sum with a following
divison).

If the sum hits ±Inf, the calculation reverts to incremental
calculation of the mean value. Kahan summation is also applied here,
although it cannot fully compensate for the numerical errors
introduced by the incremental mean calculation. (The tests added in
this commit would fail if incremental mean calculation was always
used.)

Signed-off-by: beorn7 <beorn@grafana.com>
2024-07-10 19:20:24 +02:00
..
testdata promql: make avg aggregation more precise and less expensive 2024-07-10 19:20:24 +02:00
README.md
test.go
test_test.go

README.md

The PromQL test scripting language

This package contains two things:

  • an implementation of a test scripting language for PromQL engines
  • a predefined set of tests written in that scripting language

The predefined set of tests can be run against any PromQL engine implementation by calling promqltest.RunBuiltinTests(). Any other test script can be run with promqltest.RunTest().

The rest of this document explains the test scripting language.

Each test script is written in plain text.

Comments can be given by prefixing the comment with a #, for example:

# This is a comment.

Each test file contains a series of commands. There are three kinds of commands:

  • load
  • clear
  • eval

Each command is executed in the order given in the file.

load command

load adds some data to the test environment.

The syntax is as follows:

load <interval>
    <series> <points>
    ...
    <series> <points>
  • <interval> is the step between points (eg. 1m or 30s)
  • <series> is a Prometheus series name in the usual metric{label="value"} syntax
  • <points> is a specification of the points to add for that series, following the same expanding syntax as for promtool unittest documented here

For example:

load 1m
    my_metric{env="prod"} 5 2+3x2 _ stale {{schema:1 sum:3 count:22 buckets:[5 10 7]}}

...will create a single series with labels my_metric{env="prod"}, with the following points:

  • t=0: value is 5
  • t=1m: value is 2
  • t=2m: value is 5
  • t=3m: value is 7
  • t=4m: no point
  • t=5m: stale marker
  • t=6m: native histogram with schema 1, sum -3, count 22 and bucket counts 5, 10 and 7

Each load command is additive - it does not replace any data loaded in a previous load command. Use clear to remove all loaded data.

Native histograms with custom buckets (NHCB)

When loading a batch of classic histogram float series, you can optionally append the suffix _with_nhcb to convert them to native histograms with custom buckets and load both the original float series and the new histogram series.

clear command

clear removes all data previously loaded with load commands.

eval command

eval runs a query against the test environment and asserts that the result is as expected.

Both instant and range queries are supported.

The syntax is as follows:

# Instant query
eval instant at <time> <query>
    <series> <points>
    ...
    <series> <points>
    
# Range query
eval range from <start> to <end> step <step> <query>
    <series> <points>
    ...
    <series> <points>
  • <time> is the timestamp to evaluate the instant query at (eg. 1m)
  • <start> and <end> specify the time range of the range query, and use the same syntax as <time>
  • <step> is the step of the range query, and uses the same syntax as <time> (eg. 30s)
  • <series> and <points> specify the expected values, and follow the same syntax as for load above

For example:

eval instant at 1m sum by (env) (my_metric)
    {env="prod"} 5
    {env="test"} 20
    
eval range from 0 to 3m step 1m sum by (env) (my_metric)
    {env="prod"} 2 5 10 20
    {env="test"} 10 20 30 45

Instant queries also support asserting that the series are returned in exactly the order specified: use eval_ordered instant ... instead of eval instant .... This is not supported for range queries.

It is also possible to test that queries fail: use eval_fail instant ... or eval_fail range .... eval_fail optionally takes an expected error message string or regexp to assert that the error message is as expected.

For example:

# Assert that the query fails for any reason without asserting on the error message.
eval_fail instant at 1m ceil({__name__=~'testmetric1|testmetric2'})

# Assert that the query fails with exactly the provided error message string.
eval_fail instant at 1m ceil({__name__=~'testmetric1|testmetric2'})
    expected_fail_message vector cannot contain metrics with the same labelset

# Assert that the query fails with an error message matching the regexp provided.
eval_fail instant at 1m ceil({__name__=~'testmetric1|testmetric2'})
    expected_fail_regexp (vector cannot contain metrics .*|something else went wrong)