Initial commit

checker/metrics.go

@@ -0,0 +1,137 @@
+package checker
+
+import (
+	"encoding/json"
+	"fmt"
+	"time"
+
+	sdk "git.happydns.org/checker-sdk-go/checker"
+)
+
+// ExtractMetrics implements sdk.CheckerMetricsReporter. It consumes the raw
+// observation, derives consensus on the fly, and emits time-series for
+// dashboards. Severity counters are computed from the rule states carried
+// in the ReportContext rather than re-derived from raw data.
+func (p *resolverPropagationProvider) ExtractMetrics(ctx sdk.ReportContext, collectedAt time.Time) ([]sdk.CheckMetric, error) {
+	var data ResolverPropagationData
+	if err := json.Unmarshal(ctx.Data(), &data); err != nil {
+		return nil, fmt.Errorf("resolver-propagation: decoding observation: %w", err)
+	}
+	deriveView(&data)
+
+	var out []sdk.CheckMetric
+	zone := data.Zone
+
+	rollups := []struct {
+		name string
+		val  float64
+	}{
+		{"resolver_propagation_resolvers_total", float64(data.Stats.TotalResolvers)},
+		{"resolver_propagation_resolvers_reachable", float64(data.Stats.ReachableResolvers)},
+		{"resolver_propagation_unfiltered_agreeing", float64(data.Stats.UnfilteredAgreeing)},
+		{"resolver_propagation_regions_covered", float64(data.Stats.CountriesCovered)},
+		{"resolver_propagation_run_duration_ms", float64(data.RunDurationMs)},
+	}
+	for _, r := range rollups {
+		out = append(out, sdk.CheckMetric{
+			Name:      r.name,
+			Value:     r.val,
+			Labels:    map[string]string{"zone": zone},
+			Timestamp: collectedAt,
+		})
+	}
+
+	if data.DeclaredSerial != 0 {
+		out = append(out, sdk.CheckMetric{
+			Name:      "resolver_propagation_declared_serial",
+			Value:     float64(data.DeclaredSerial),
+			Labels:    map[string]string{"zone": zone},
+			Timestamp: collectedAt,
+		})
+	}
+
+	soaKey := rrsetKey(zone, "SOA")
+	var staleResolvers int
+	for id, rv := range data.Resolvers {
+		if rv.Filtered {
+			continue
+		}
+		p := rv.Probes[soaKey]
+		if p == nil || p.Error != "" || p.Rcode != "NOERROR" {
+			continue
+		}
+		s := extractSerial(p.Records)
+		if s == 0 {
+			continue
+		}
+		out = append(out, sdk.CheckMetric{
+			Name:  "resolver_propagation_observed_serial",
+			Value: float64(s),
+			Labels: map[string]string{
+				"zone":     zone,
+				"resolver": id,
+			},
+			Timestamp: collectedAt,
+		})
+		if data.DeclaredSerial != 0 && s < data.DeclaredSerial {
+			staleResolvers++
+		}
+	}
+	if data.DeclaredSerial != 0 {
+		out = append(out, sdk.CheckMetric{
+			Name:      "resolver_propagation_serial_drift_resolvers",
+			Value:     float64(staleResolvers),
+			Labels:    map[string]string{"zone": zone},
+			Timestamp: collectedAt,
+		})
+	}
+
+	for id, rv := range data.Resolvers {
+		labels := map[string]string{
+			"zone":      zone,
+			"resolver":  id,
+			"ip":        rv.IP,
+			"region":    rv.Region,
+			"transport": string(rv.Transport),
+		}
+
+		up := float64(0)
+		if rv.Reachable {
+			up = 1
+		}
+		out = append(out, sdk.CheckMetric{
+			Name:  "resolver_propagation_resolver_up",
+			Value: up, Labels: labels, Timestamp: collectedAt,
+		})
+
+		var total, n int64
+		for _, p := range rv.Probes {
+			if p.Error != "" {
+				continue
+			}
+			total += p.LatencyMs
+			n++
+		}
+		if n > 0 {
+			out = append(out, sdk.CheckMetric{
+				Name:  "resolver_propagation_resolver_latency_ms",
+				Value: float64(total) / float64(n), Unit: "ms",
+				Labels: labels, Timestamp: collectedAt,
+			})
+		}
+	}
+
+	for key, v := range data.RRsets {
+		out = append(out, sdk.CheckMetric{
+			Name:  "resolver_propagation_rrset_signatures",
+			Value: float64(len(v.Groups)),
+			Labels: map[string]string{
+				"zone":  zone,
+				"rrset": key,
+			},
+			Timestamp: collectedAt,
+		})
+	}
+
+	return out, nil
+}