VictoriaMetrics/app/vmselect/promql/transform_test.go
Zakhar Bessarab 401ae72587
app/vmselect/promql: propagate lower bucket values when fixing a histogram (#6547)
### Describe Your Changes

In most cases histograms are exposed in sorted manner with lower buckets
being first. This means that during scraping buckets with lower bounds
have higher chance of being updated earlier than upper ones.

Previously, values were propagated from upper to lower bounds, which
means that in most cases that would produce results higher than expected
once all buckets will become updated.
Propagating from upper bound effectively limits highest value of
histogram to the value of previous scrape. Once the data will become
consistent in the subsequent evaluation this causes spikes in the
result.

Changing propagation to be from lower to higher buckets reduces value
spikes in most cases due to nature of the original inconsistency.

 See: https://github.com/VictoriaMetrics/VictoriaMetrics/issues/4580

An example histogram with previous(red) and updated(blue) versions:

![1719565540](https://github.com/VictoriaMetrics/VictoriaMetrics/assets/1367798/605c5e60-6abe-45b5-89b2-d470b60127b8)

This also makes logic of filling nan values with lower buckets values: [1 2 3 nan nan nan] => [1 2 3 3 3 3] obsolete.
Since buckets are now fixed from lower ones to upper this happens in the main loop, so there is no need in a second one.

---------

Signed-off-by: Zakhar Bessarab <z.bessarab@victoriametrics.com>
Signed-off-by: hagen1778 <roman@victoriametrics.com>
Co-authored-by: Andrii Chubatiuk <andrew.chubatiuk@gmail.com>
Co-authored-by: hagen1778 <roman@victoriametrics.com>
(cherry picked from commit 6a4bd5049b)
2024-07-10 15:17:08 +02:00

362 lines
7.9 KiB
Go

package promql
import (
"fmt"
"reflect"
"strconv"
"strings"
"testing"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/protoparser/prometheus"
"github.com/VictoriaMetrics/VictoriaMetrics/lib/storage"
)
func TestFixBrokenBuckets(t *testing.T) {
f := func(values, expectedResult []float64) {
t.Helper()
xss := make([]leTimeseries, len(values))
for i, v := range values {
xss[i].ts = &timeseries{
Values: []float64{v},
}
}
fixBrokenBuckets(0, xss)
result := make([]float64, len(values))
for i, xs := range xss {
result[i] = xs.ts.Values[0]
}
if !reflect.DeepEqual(result, expectedResult) {
t.Fatalf("unexpected result for values=%v\ngot\n%v\nwant\n%v", values, result, expectedResult)
}
}
f(nil, []float64{})
f([]float64{1}, []float64{1})
f([]float64{1, 2}, []float64{1, 2})
f([]float64{2, 1}, []float64{2, 2})
f([]float64{1, 2, 3, nan, nan}, []float64{1, 2, 3, 3, 3})
f([]float64{5, 1, 2, 3, nan}, []float64{5, 5, 5, 5, 5})
f([]float64{1, 5, 2, nan, 6, 3}, []float64{1, 5, 5, 5, 6, 6})
f([]float64{5, 10, 4, 3}, []float64{5, 10, 10, 10})
}
func TestVmrangeBucketsToLE(t *testing.T) {
f := func(buckets, bucketsExpected string) {
t.Helper()
tss := promMetricsToTimeseries(buckets)
result := vmrangeBucketsToLE(tss)
resultBuckets := timeseriesToPromMetrics(result)
if !reflect.DeepEqual(resultBuckets, bucketsExpected) {
t.Errorf("unexpected vmrangeBucketsToLE(); got\n%v\nwant\n%v", resultBuckets, bucketsExpected)
}
}
// A single non-empty vmrange bucket
f(
`foo{vmrange="4.084e+02...4.642e+02"} 2 123`,
`foo{le="4.084e+02"} 0 123
foo{le="4.642e+02"} 2 123
foo{le="+Inf"} 2 123`,
)
f(
`foo{vmrange="0...+Inf"} 5 123`,
`foo{le="+Inf"} 5 123`,
)
f(
`foo{vmrange="-Inf...0"} 4 123`,
`foo{le="-Inf"} 0 123
foo{le="0"} 4 123
foo{le="+Inf"} 4 123`,
)
f(
`foo{vmrange="-Inf...+Inf"} 1.23 456`,
`foo{le="-Inf"} 0 456
foo{le="+Inf"} 1.23 456`,
)
f(
`foo{vmrange="0...0"} 5.3 0`,
`foo{le="0"} 5.3 0
foo{le="+Inf"} 5.3 0`,
)
// Adjacent empty vmrange bucket
f(
`foo{vmrange="7.743e+05...8.799e+05"} 5 123
foo{vmrange="6.813e+05...7.743e+05"} 0 123`,
`foo{le="7.743e+05"} 0 123
foo{le="8.799e+05"} 5 123
foo{le="+Inf"} 5 123`,
)
// Multiple adjacent empty vmrange bucket
f(
`foo{vmrange="7.743e+05...8.799e+05"} 5 123
foo{vmrange="6.813e+05...7.743e+05"} 0 123
foo{vmrange="5.813e+05...6.813e+05"} 0 123
`,
`foo{le="7.743e+05"} 0 123
foo{le="8.799e+05"} 5 123
foo{le="+Inf"} 5 123`,
)
f(
`foo{vmrange="8.799e+05...9.813e+05"} 0 123
foo{vmrange="7.743e+05...8.799e+05"} 5 123
foo{vmrange="6.813e+05...7.743e+05"} 0 123
foo{vmrange="5.813e+05...6.813e+05"} 0 123
`,
`foo{le="7.743e+05"} 0 123
foo{le="8.799e+05"} 5 123
foo{le="+Inf"} 5 123`,
)
// Multiple non-empty vmrange buckets
f(
`foo{vmrange="4.084e+02...4.642e+02"} 2 123
foo{vmrange="1.234e+02...4.084e+02"} 3 123
`,
`foo{le="1.234e+02"} 0 123
foo{le="4.084e+02"} 3 123
foo{le="4.642e+02"} 5 123
foo{le="+Inf"} 5 123`,
)
// Multiple disjoint vmrange buckets
f(
`foo{vmrange="1...2"} 2 123
foo{vmrange="4...6"} 3 123
`,
`foo{le="1"} 0 123
foo{le="2"} 2 123
foo{le="4"} 2 123
foo{le="6"} 5 123
foo{le="+Inf"} 5 123`,
)
// Multiple intersected vmrange buckets
f(
`foo{vmrange="1...5"} 2 123
foo{vmrange="4...6"} 3 123
`,
`foo{le="1"} 0 123
foo{le="5"} 2 123
foo{le="4"} 2 123
foo{le="6"} 5 123
foo{le="+Inf"} 5 123`,
)
// Multiple vmrange buckets with the same end range
f(
`foo{vmrange="1...5"} 2 123
foo{vmrange="0...5"} 3 123
`,
`foo{le="1"} 0 123
foo{le="5"} 2 123
foo{le="0"} 2 123
foo{le="+Inf"} 2 123`,
)
// A single empty vmrange bucket
f(
`foo{vmrange="0...1"} 0 123`,
``,
)
f(
`foo{vmrange="0...+Inf"} 0 123`,
``,
)
f(
`foo{vmrange="-Inf...0"} 0 123`,
``,
)
f(
`foo{vmrange="0...0"} 0 0`,
``,
)
f(
`foo{vmrange="-Inf...+Inf"} 0 456`,
``,
)
// Multiple empty vmrange buckets
f(
`foo{vmrange="2...3"} 0 123
foo{vmrange="1...2"} 0 123`,
``,
)
// The bucket with negative value
f(
`foo{vmrange="4.084e+02...4.642e+02"} -5 1`,
``,
)
// Missing vmrange in the original metric
f(
`foo 3 6`,
``,
)
// Missing le label in the original metric
f(
`foo{le="456"} 3 6`,
`foo{le="456"} 3 6`,
)
// Invalid vmrange label value
f(
`foo{vmrange="foo...bar"} 1 1`,
``,
)
f(
`foo{vmrange="4.084e+02"} 1 1`,
``,
)
f(
`foo{vmrange="4.084e+02...foo"} 1 1`,
``,
)
}
func promMetricsToTimeseries(s string) []*timeseries {
var rows prometheus.Rows
rows.UnmarshalWithErrLogger(s, func(errStr string) {
panic(fmt.Errorf("cannot parse %q: %s", s, errStr))
})
var tss []*timeseries
for _, row := range rows.Rows {
var tags []storage.Tag
for _, tag := range row.Tags {
tags = append(tags, storage.Tag{
Key: []byte(tag.Key),
Value: []byte(tag.Value),
})
}
var ts timeseries
ts.MetricName.MetricGroup = []byte(row.Metric)
ts.MetricName.Tags = tags
ts.Timestamps = append(ts.Timestamps, row.Timestamp/1000)
ts.Values = append(ts.Values, row.Value)
tss = append(tss, &ts)
}
return tss
}
func timeseriesToPromMetrics(tss []*timeseries) string {
var a []string
for _, ts := range tss {
metricName := ts.MetricName.String()
for i := range ts.Timestamps {
metricName = strings.TrimPrefix(metricName, "AccountID=0, ProjectID=0, ")
line := fmt.Sprintf("%s %v %d", metricName, ts.Values[i], ts.Timestamps[i])
a = append(a, line)
}
}
return strings.Join(a, "\n")
}
func TestGetNumPrefix(t *testing.T) {
f := func(s, prefixExpected string) {
t.Helper()
prefix := getNumPrefix(s)
if prefix != prefixExpected {
t.Fatalf("unexpected getNumPrefix(%q): got %q; want %q", s, prefix, prefixExpected)
}
if len(prefix) > 0 {
if _, err := strconv.ParseFloat(prefix, 64); err != nil {
t.Fatalf("cannot parse num %q: %s", prefix, err)
}
}
}
f("", "")
f("foo", "")
f("-", "")
f(".", "")
f("-.", "")
f("+..", "")
f("1", "1")
f("12", "12")
f("1foo", "1")
f("-123", "-123")
f("-123bar", "-123")
f("+123", "+123")
f("+123.", "+123.")
f("+123..", "+123.")
f("+123.-", "+123.")
f("12.34..", "12.34")
f("-12.34..", "-12.34")
f("-12.-34..", "-12.")
}
func TestNumericLess(t *testing.T) {
f := func(a, b string, want bool) {
t.Helper()
if got := numericLess(a, b); got != want {
t.Fatalf("unexpected numericLess(%q, %q): got %v; want %v", a, b, got, want)
}
}
// empty strings
f("", "", false)
f("", "321", true)
f("321", "", false)
f("", "abc", true)
f("abc", "", false)
f("foo", "123", false)
f("123", "foo", true)
// same length numbers
f("123", "321", true)
f("321", "123", false)
f("123", "123", false)
// same length strings
f("a", "b", true)
f("b", "a", false)
f("a", "a", false)
// identical string prefix
f("foo123", "foo", false)
f("foo", "foo123", true)
f("foo", "foo", false)
// identical num prefix
f("123foo", "123bar", false)
f("123bar", "123foo", true)
f("123bar", "123bar", false)
// numbers with special chars
f("1:0:0", "1:0:2", true)
// numbers with special chars and different number rank
f("1:0:15", "1:0:2", false)
// multiple zeroes"
f("0", "00", false)
// only chars
f("aa", "ab", true)
// strings with different lengths
f("ab", "abc", true)
// multiple zeroes after equal char
f("a0001", "a0000001", false)
// short first string with numbers and highest rank
f("a10", "abcdefgh2", true)
// less as second string
f("a1b", "a01b", false)
// equal strings by length with different number rank
f("a001b01", "a01b001", false)
// different numbers rank
f("a01b001", "a001b01", false)
// different numbers rank
f("a01b001", "a001b01", false)
// highest char and number
f("a1", "a1x", true)
// highest number reverse chars
f("1b", "1ax", false)
// numbers with leading zero
f("082", "83", true)
// numbers with leading zero and chars
f("083a", "9a", false)
f("083a", "94a", true)
// negative number
f("-123", "123", true)
f("-123", "+123", true)
f("-123", "-123", false)
f("123", "-123", false)
// fractional number
f("12.9", "12.56", false)
f("12.56", "12.9", true)
f("12.9", "12.9", false)
}