fix(alert): keep sample history by rule-defined retention window

Trim alert samples using AlertRule.RetentionWindow() (max rule duration)
instead of Check()'s verdict max, which is 0 while a rule's window is still
filling. The old max<=0 trim wiped history every tick, so Duration>=2 general
rules never accumulated enough samples and never fired a notification.

Add model + end-to-end singleton regression tests covering sample
accumulation, the retention-window contract, and the actual notify path.
This commit is contained in:
naiba
2026-06-07 16:11:43 +00:00
parent 66de244c2e
commit e756b4540c
4 changed files with 178 additions and 8 deletions
+19
View File
@@ -196,6 +196,25 @@ func (r *AlertRule) Check(points [][]bool) (int, bool) {
return slices.Max(durations), hasPassedRule
}
// RetentionWindow 返回保留历史采样所需的长度(各规则窗口的最大值),只依赖
// 规则定义而非 Check 的判定结果——否则窗口未填满时 Check 返回的 max=0 会被
// 误判为"无需历史"而清空采样,使规则永远攒不够样本。
func (r *AlertRule) RetentionWindow() int {
window := 0
for _, rule := range r.Rules {
var need int
if rule.IsTransferDurationRule() || rule.IsOfflineRule() {
need = 1
} else if d := int(rule.Duration); d > 0 {
need = d
}
if need > window {
window = need
}
}
return window
}
func boundCheck(length, duration int, passed bool) bool {
if passed {
return true
+64
View File
@@ -360,3 +360,67 @@ func TestAlertRule_ZeroDurationMixedWithValidRule(t *testing.T) {
t.Fatalf("the valid Duration:3 rule must still set max=3, got %d", maxD)
}
}
// trimSamples mirrors singleton.checkStatus retention: keep the most recent
// `window` samples, clear when window<=0. window comes from RetentionWindow(),
// the production code under test.
func trimSamples(samples [][]bool, window int) [][]bool {
if window <= 0 {
return samples[:0]
} else if window < len(samples) {
return samples[len(samples)-window:]
}
return samples
}
// TestAlertRule_GeneralRuleAccumulatesSamples is a regression guard: a normal
// Duration>1 general rule must be able to fire. checkStatus appends one sample
// per tick then trims to the retention window; if the window is derived from
// Check's verdict (which is 0 while the rule is still filling) the history is
// wiped every tick, the window never reaches Duration, and the alert never
// raises. RetentionWindow() must keep enough samples for the rule to converge.
func TestAlertRule_GeneralRuleAccumulatesSamples(t *testing.T) {
const duration = 10
rule := &AlertRule{
Rules: []*Rule{{Type: "cpu", Duration: duration}},
}
var samples [][]bool
var lastPassed bool
maxLen := 0
for tick := 0; tick < duration*3; tick++ {
samples = append(samples, []bool{false}) // failing sample
_, lastPassed = rule.Check(samples)
samples = trimSamples(samples, rule.RetentionWindow())
if len(samples) > maxLen {
maxLen = len(samples)
}
}
if maxLen < duration {
t.Fatalf("samples never accumulated to Duration: max window reached %d, want >= %d", maxLen, duration)
}
if lastPassed {
t.Fatalf("a server failing every tick must eventually fail the check (passed=false), got passed=true")
}
}
// TestAlertRule_RetentionWindow pins the retention contract directly.
func TestAlertRule_RetentionWindow(t *testing.T) {
cases := []struct {
msg string
rule *AlertRule
want int
}{
{"single general", &AlertRule{Rules: []*Rule{{Type: "cpu", Duration: 10}}}, 10},
{"zero duration only", &AlertRule{Rules: []*Rule{{Type: "cpu", Duration: 0}}}, 0},
{"mixed picks max", &AlertRule{Rules: []*Rule{{Type: "cpu", Duration: 0}, {Type: "cpu", Duration: 7}}}, 7},
{"offline looks back one", &AlertRule{Rules: []*Rule{{Type: "offline", Duration: 30}}}, 1},
{"cycle looks back one", &AlertRule{Rules: []*Rule{{Type: "net_in_speed_cycle"}}}, 1},
}
for _, c := range cases {
if got := c.rule.RetentionWindow(); got != c.want {
t.Fatalf("%s: RetentionWindow()=%d want %d", c.msg, got, c.want)
}
}
}