add prometheus instrumentation

internal/metrics/metrics.go

@@ -0,0 +1,125 @@
+package metrics
+
+import (
+	"net/http"
+	"time"
+
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promhttp"
+)
+
+var (
+	registry = prometheus.NewRegistry()
+
+	HourlySnapshotTotal    = prometheus.NewCounter(prometheus.CounterOpts{Name: "vctp_hourly_snapshots_total", Help: "Total number of hourly snapshot jobs completed."})
+	HourlySnapshotFailures = prometheus.NewCounter(prometheus.CounterOpts{Name: "vctp_hourly_snapshots_failed_total", Help: "Hourly snapshot jobs that failed."})
+	HourlySnapshotLast     = prometheus.NewGauge(prometheus.GaugeOpts{Name: "vctp_hourly_snapshot_last_unix", Help: "Unix timestamp of the last hourly snapshot start time."})
+	HourlySnapshotRows     = prometheus.NewGauge(prometheus.GaugeOpts{Name: "vctp_hourly_snapshot_last_rows", Help: "Row count of the last hourly snapshot table."})
+
+	DailyAggregationsTotal   = prometheus.NewCounter(prometheus.CounterOpts{Name: "vctp_daily_aggregations_total", Help: "Total number of daily aggregation jobs completed."})
+	DailyAggregationFailures = prometheus.NewCounter(prometheus.CounterOpts{Name: "vctp_daily_aggregations_failed_total", Help: "Daily aggregation jobs that failed."})
+	DailyAggregationDuration = prometheus.NewHistogram(prometheus.HistogramOpts{
+		Name:    "vctp_daily_aggregation_duration_seconds",
+		Help:    "Duration of daily aggregation jobs.",
+		Buckets: prometheus.ExponentialBuckets(1, 2, 10),
+	})
+
+	MonthlyAggregationsTotal   = prometheus.NewCounter(prometheus.CounterOpts{Name: "vctp_monthly_aggregations_total", Help: "Total number of monthly aggregation jobs completed."})
+	MonthlyAggregationFailures = prometheus.NewCounter(prometheus.CounterOpts{Name: "vctp_monthly_aggregations_failed_total", Help: "Monthly aggregation jobs that failed."})
+	MonthlyAggregationDuration = prometheus.NewHistogram(prometheus.HistogramOpts{
+		Name:    "vctp_monthly_aggregation_duration_seconds",
+		Help:    "Duration of monthly aggregation jobs.",
+		Buckets: prometheus.ExponentialBuckets(1, 2, 10),
+	})
+
+	ReportsAvailable = prometheus.NewGauge(prometheus.GaugeOpts{
+		Name: "vctp_reports_available",
+		Help: "Number of downloadable reports present on disk.",
+	})
+
+	VcenterConnectFailures = prometheus.NewCounterVec(prometheus.CounterOpts{
+		Name: "vctp_vcenter_connect_failures_total",
+		Help: "Failed connections to vCenter during snapshot runs.",
+	}, []string{"vcenter"})
+
+	VcenterSnapshotDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
+		Name:    "vctp_vcenter_snapshot_duration_seconds",
+		Help:    "Duration of per-vCenter hourly snapshot jobs.",
+		Buckets: prometheus.ExponentialBuckets(0.5, 2, 10),
+	}, []string{"vcenter"})
+
+	VcenterInventorySize = prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "vctp_vcenter_inventory_size",
+		Help: "Number of VMs seen in the last successful snapshot per vCenter.",
+	}, []string{"vcenter"})
+)
+
+func init() {
+	registry.MustRegister(
+		HourlySnapshotTotal,
+		HourlySnapshotFailures,
+		HourlySnapshotLast,
+		HourlySnapshotRows,
+		DailyAggregationsTotal,
+		DailyAggregationFailures,
+		DailyAggregationDuration,
+		MonthlyAggregationsTotal,
+		MonthlyAggregationFailures,
+		MonthlyAggregationDuration,
+		ReportsAvailable,
+		VcenterConnectFailures,
+		VcenterSnapshotDuration,
+		VcenterInventorySize,
+	)
+}
+
+// Handler returns an http.Handler that serves Prometheus metrics.
+func Handler() http.Handler {
+	return promhttp.HandlerFor(registry, promhttp.HandlerOpts{})
+}
+
+// RecordVcenterSnapshot logs per-vCenter snapshot metrics.
+func RecordVcenterSnapshot(vcenter string, duration time.Duration, vmCount int64, err error) {
+	VcenterSnapshotDuration.WithLabelValues(vcenter).Observe(duration.Seconds())
+	if err != nil {
+		VcenterConnectFailures.WithLabelValues(vcenter).Inc()
+		return
+	}
+	VcenterInventorySize.WithLabelValues(vcenter).Set(float64(vmCount))
+}
+
+// RecordHourlySnapshot logs aggregate hourly snapshot results.
+func RecordHourlySnapshot(start time.Time, rows int64, err error) {
+	HourlySnapshotLast.Set(float64(start.Unix()))
+	HourlySnapshotRows.Set(float64(rows))
+	if err != nil {
+		HourlySnapshotFailures.Inc()
+		return
+	}
+	HourlySnapshotTotal.Inc()
+}
+
+// RecordDailyAggregation logs daily aggregation metrics.
+func RecordDailyAggregation(duration time.Duration, err error) {
+	DailyAggregationDuration.Observe(duration.Seconds())
+	if err != nil {
+		DailyAggregationFailures.Inc()
+		return
+	}
+	DailyAggregationsTotal.Inc()
+}
+
+// RecordMonthlyAggregation logs monthly aggregation metrics.
+func RecordMonthlyAggregation(duration time.Duration, err error) {
+	MonthlyAggregationDuration.Observe(duration.Seconds())
+	if err != nil {
+		MonthlyAggregationFailures.Inc()
+		return
+	}
+	MonthlyAggregationsTotal.Inc()
+}
+
+// SetReportsAvailable updates the gauge for report files found on disk.
+func SetReportsAvailable(count int) {
+	ReportsAvailable.Set(float64(count))
+}