Initial commit

checker/evaluate.go

@@ -0,0 +1,120 @@
+package checker
+
+import (
+	"encoding/json"
+	"fmt"
+	"time"
+
+	sdk "git.happydns.org/checker-sdk-go/checker"
+)
+
+// ExtractMetrics turns the stored data into time-series metrics so the host
+// can feed Prometheus / Grafana dashboards. Implements
+// sdk.CheckerMetricsReporter.
+//
+// Findings-based counters are re-derived from the raw observation using
+// default option values, since the rule set (and hence severity counts) is
+// no longer serialized with the observation.
+func (p *authoritativeConsistencyProvider) ExtractMetrics(ctx sdk.ReportContext, collectedAt time.Time) ([]sdk.CheckMetric, error) {
+	var data ObservationData
+	if err := json.Unmarshal(ctx.Data(), &data); err != nil {
+		return nil, fmt.Errorf("checker: decoding observation: %w", err)
+	}
+
+	var out []sdk.CheckMetric
+
+	// One "reachability" pair per NS, easy to alert on individual servers.
+	for name, r := range data.Results {
+		labels := map[string]string{"zone": data.Zone, "ns": name}
+
+		up := float64(0)
+		if r.UDPReachable && r.Authoritative {
+			up = 1
+		}
+		out = append(out, sdk.CheckMetric{
+			Name:      "authoritative_consistency_ns_up",
+			Value:     up,
+			Labels:    labels,
+			Timestamp: collectedAt,
+		})
+
+		tcp := float64(0)
+		if r.TCPReachable {
+			tcp = 1
+		}
+		out = append(out, sdk.CheckMetric{
+			Name:      "authoritative_consistency_ns_tcp",
+			Value:     tcp,
+			Labels:    labels,
+			Timestamp: collectedAt,
+		})
+
+		if r.LatencyMs > 0 {
+			out = append(out, sdk.CheckMetric{
+				Name:      "authoritative_consistency_ns_latency_ms",
+				Value:     float64(r.LatencyMs),
+				Unit:      "ms",
+				Labels:    labels,
+				Timestamp: collectedAt,
+			})
+		}
+
+		if r.Serial > 0 {
+			out = append(out, sdk.CheckMetric{
+				Name:      "authoritative_consistency_ns_serial",
+				Value:     float64(r.Serial),
+				Labels:    labels,
+				Timestamp: collectedAt,
+			})
+		}
+	}
+
+	// Zone-wide metrics.
+	uniqueSerials := map[uint32]struct{}{}
+	for _, r := range data.Results {
+		if r.Serial != 0 {
+			uniqueSerials[r.Serial] = struct{}{}
+		}
+	}
+	out = append(out, sdk.CheckMetric{
+		Name:      "authoritative_consistency_unique_serials",
+		Value:     float64(len(uniqueSerials)),
+		Labels:    map[string]string{"zone": data.Zone},
+		Timestamp: collectedAt,
+	})
+
+	// Severity counters sourced from ctx.States(). When the host does not
+	// provide states (e.g. a metrics-only render path), the counters are
+	// simply emitted as zero; we do NOT re-derive from the raw data.
+	states := ctx.States()
+	if states != nil {
+		var crit, warn, info int
+		for _, st := range states {
+			switch st.Status {
+			case sdk.StatusCrit:
+				crit++
+			case sdk.StatusWarn:
+				warn++
+			case sdk.StatusInfo:
+				info++
+			}
+		}
+		for _, x := range []struct {
+			name string
+			val  int
+		}{
+			{"authoritative_consistency_findings_critical", crit},
+			{"authoritative_consistency_findings_warning", warn},
+			{"authoritative_consistency_findings_info", info},
+		} {
+			out = append(out, sdk.CheckMetric{
+				Name:      x.name,
+				Value:     float64(x.val),
+				Labels:    map[string]string{"zone": data.Zone},
+				Timestamp: collectedAt,
+			})
+		}
+	}
+
+	return out, nil
+}