Files
go-otel/internal/features/spike_test.go
T
mats 94f30165f9
Build and Publish / Metadata (push) Successful in 3s
Build and Publish / BuildAndDeployAmd64 (push) Successful in 1m21s
more ml
2026-09-19 22:00:19 +02:00

80 lines
2.9 KiB
Go

package features
import (
"testing"
"time"
)
func TestDetectSpikesIgnoresChronicPatterns(t *testing.T) {
at := time.Date(2026, 9, 19, 12, 0, 0, 0, time.UTC)
cfg := SpikeConfig{Window: 5 * time.Minute, BaselineWindows: 12}
var events []Event
add := func(sig string, n int, when time.Time) {
for i := 0; i < n; i++ {
events = append(events, Event{Timestamp: when.Add(time.Duration(i) * time.Second), Service: "api", IsError: true, PatternSignature: sig})
}
}
// chronic poller: 50 errors in every window, including the current one
for w := 0; w <= 12; w++ {
add("poll-401", 50, at.Add(-time.Duration(w+1)*cfg.Window))
}
if got := DetectSpikes(events, at, cfg); len(got) != 0 {
t.Fatalf("chronic pattern alone alerted: %+v", got)
}
// a novel failure appears in the current window
add("db-timeout", 20, at.Add(-cfg.Window))
got := DetectSpikes(events, at, cfg)
if len(got) != 1 || got[0].Patterns[0].Signature != "db-timeout" || !got[0].Patterns[0].Novel {
t.Fatalf("expected db-timeout spike, got %+v", got)
}
}
func TestDetectSpikesRequiresMinimumExcess(t *testing.T) {
at := time.Date(2026, 9, 19, 12, 0, 0, 0, time.UTC)
events := []Event{}
for i := 0; i < 3; i++ {
events = append(events, Event{Timestamp: at.Add(-time.Minute), Service: "api", IsError: true, PatternSignature: "x"})
}
if got := DetectSpikes(events, at, SpikeConfig{}); len(got) != 0 {
t.Fatalf("3 errors should stay under the floor: %+v", got)
}
}
func TestBaselineTrimIgnoresEarlierStorm(t *testing.T) {
at := time.Date(2026, 9, 19, 12, 0, 0, 0, time.UTC)
cfg := SpikeConfig{Window: 5 * time.Minute, BaselineWindows: 10}
var events []Event
add := func(sig string, n int, when time.Time) {
for i := 0; i < n; i++ {
events = append(events, Event{Timestamp: when.Add(time.Duration(i) * time.Millisecond), Service: "api", IsError: true, PatternSignature: sig})
}
}
// one earlier storm in 1 of 10 baseline windows, then a smaller repeat now
add("storm", 1000, at.Add(-6*cfg.Window))
add("storm", 60, at.Add(-cfg.Window))
got := DetectSpikes(events, at, cfg)
if len(got) != 1 || got[0].Patterns[0].BaselineMean != 0 {
t.Fatalf("storm baseline should be trimmed to 0, got %+v", got)
}
}
func TestSuppressorCooldownAndEscalation(t *testing.T) {
at := time.Date(2026, 9, 19, 12, 0, 0, 0, time.UTC)
spike := func(at time.Time, count int) []Spike {
return []Spike{{Service: "api", WindowEnd: at, Patterns: []PatternDelta{{Signature: "x", Count: count, Excess: float64(count)}}}}
}
s := NewSuppressor(30 * time.Minute)
if len(s.Filter(spike(at, 20))) != 1 {
t.Fatal("first alert should pass")
}
if len(s.Filter(spike(at.Add(5*time.Minute), 22))) != 0 {
t.Fatal("repeat inside cooldown should be suppressed")
}
if len(s.Filter(spike(at.Add(10*time.Minute), 45))) != 1 {
t.Fatal("2x escalation should re-alert")
}
if len(s.Filter(spike(at.Add(50*time.Minute), 46))) != 1 {
t.Fatal("alert after cooldown should pass")
}
}