cloudflare/pint
Publicmirrored from https://github.com/cloudflare/pintAvailable
internal/checks/alerts_count.go
132lines · modecode
| 1 | package checks |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "fmt" |
| 6 | "log/slog" |
| 7 | "net/url" |
| 8 | "sort" |
| 9 | "time" |
| 10 | |
| 11 | "github.com/prometheus/common/model" |
| 12 | |
| 13 | "github.com/cloudflare/pint/internal/discovery" |
| 14 | "github.com/cloudflare/pint/internal/output" |
| 15 | "github.com/cloudflare/pint/internal/parser" |
| 16 | "github.com/cloudflare/pint/internal/promapi" |
| 17 | ) |
| 18 | |
| 19 | const ( |
| 20 | AlertsCheckName = "alerts/count" |
| 21 | ) |
| 22 | |
| 23 | func NewAlertsCheck(prom *promapi.FailoverGroup, lookBack, step, resolve time.Duration, minCount int, severity Severity) AlertsCheck { |
| 24 | return AlertsCheck{ |
| 25 | prom: prom, |
| 26 | lookBack: lookBack, |
| 27 | step: step, |
| 28 | resolve: resolve, |
| 29 | minCount: minCount, |
| 30 | severity: severity, |
| 31 | } |
| 32 | } |
| 33 | |
| 34 | type AlertsCheck struct { |
| 35 | prom *promapi.FailoverGroup |
| 36 | lookBack time.Duration |
| 37 | step time.Duration |
| 38 | resolve time.Duration |
| 39 | minCount int |
| 40 | severity Severity |
| 41 | } |
| 42 | |
| 43 | func (c AlertsCheck) Meta() CheckMeta { |
| 44 | return CheckMeta{IsOnline: true} |
| 45 | } |
| 46 | |
| 47 | func (c AlertsCheck) String() string { |
| 48 | return fmt.Sprintf("%s(%s)", AlertsCheckName, c.prom.Name()) |
| 49 | } |
| 50 | |
| 51 | func (c AlertsCheck) Reporter() string { |
| 52 | return AlertsCheckName |
| 53 | } |
| 54 | |
| 55 | func (c AlertsCheck) Check(ctx context.Context, _ string, rule parser.Rule, _ []discovery.Entry) (problems []Problem) { |
| 56 | if rule.AlertingRule == nil { |
| 57 | return problems |
| 58 | } |
| 59 | |
| 60 | if rule.AlertingRule.Expr.SyntaxError != nil { |
| 61 | return problems |
| 62 | } |
| 63 | |
| 64 | params := promapi.NewRelativeRange(c.lookBack, c.step) |
| 65 | |
| 66 | qr, err := c.prom.RangeQuery(ctx, rule.AlertingRule.Expr.Value.Value, params) |
| 67 | if err != nil { |
| 68 | text, severity := textAndSeverityFromError(err, c.Reporter(), c.prom.Name(), Bug) |
| 69 | problems = append(problems, Problem{ |
| 70 | Fragment: rule.AlertingRule.Expr.Value.Value, |
| 71 | Lines: rule.AlertingRule.Expr.Lines(), |
| 72 | Reporter: c.Reporter(), |
| 73 | Text: text, |
| 74 | Severity: severity, |
| 75 | }) |
| 76 | return problems |
| 77 | } |
| 78 | |
| 79 | if len(qr.Series.Ranges) > 0 { |
| 80 | promUptime, err := c.prom.RangeQuery(ctx, "count(up)", params) |
| 81 | if err != nil { |
| 82 | slog.Warn("Cannot detect Prometheus uptime gaps", slog.Any("err", err), slog.String("name", c.prom.Name())) |
| 83 | } else { |
| 84 | // FIXME: gaps are not used |
| 85 | qr.Series.FindGaps(promUptime.Series, qr.Series.From, qr.Series.Until) |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | var forDur model.Duration |
| 90 | if rule.AlertingRule.For != nil { |
| 91 | forDur, _ = model.ParseDuration(rule.AlertingRule.For.Value.Value) |
| 92 | } |
| 93 | var keepFiringForDur model.Duration |
| 94 | if rule.AlertingRule.KeepFiringFor != nil { |
| 95 | keepFiringForDur, _ = model.ParseDuration(rule.AlertingRule.KeepFiringFor.Value.Value) |
| 96 | } |
| 97 | |
| 98 | var alerts int |
| 99 | for _, r := range qr.Series.Ranges { |
| 100 | // If `keepFiringFor` is not defined its Duration will be 0 |
| 101 | if r.End.Sub(r.Start) > (time.Duration(forDur) + time.Duration(keepFiringForDur)) { |
| 102 | alerts++ |
| 103 | } |
| 104 | } |
| 105 | |
| 106 | if alerts < c.minCount { |
| 107 | return problems |
| 108 | } |
| 109 | |
| 110 | lines := []int{} |
| 111 | lines = append(lines, rule.AlertingRule.Expr.Lines()...) |
| 112 | if rule.AlertingRule.For != nil { |
| 113 | lines = append(lines, rule.AlertingRule.For.Lines()...) |
| 114 | } |
| 115 | if rule.AlertingRule.KeepFiringFor != nil { |
| 116 | lines = append(lines, rule.AlertingRule.KeepFiringFor.Lines()...) |
| 117 | } |
| 118 | sort.Ints(lines) |
| 119 | |
| 120 | delta := qr.Series.Until.Sub(qr.Series.From).Round(time.Minute) |
| 121 | problems = append(problems, Problem{ |
| 122 | Fragment: rule.AlertingRule.Expr.Value.Value, |
| 123 | Lines: lines, |
| 124 | Reporter: c.Reporter(), |
| 125 | Text: fmt.Sprintf("%s would trigger %d alert(s) in the last %s.", promText(c.prom.Name(), qr.URI), alerts, output.HumanizeDuration(delta)), |
| 126 | Details: fmt.Sprintf(`To get a preview of the alerts that would fire please [click here](%s/graph?g0.expr=%s&g0.range_input=%s).`, |
| 127 | qr.PublicURI, url.QueryEscape(rule.AlertingRule.Expr.Value.Value), output.HumanizeDuration(delta), |
| 128 | ), |
| 129 | Severity: c.severity, |
| 130 | }) |
| 131 | return problems |
| 132 | } |