EMS/internal/metrics/metrics.go

package metrics

import (
	"github.com/prometheus/client_golang/prometheus"
	"github.com/tb/ems/internal/engine"
)

// Metrics holds all EMS Prometheus metrics.
type Metrics struct {
	// Current system state
	GridPowerW    prometheus.Gauge
	BatterySOC    prometheus.Gauge
	PVProductionW prometheus.Gauge

	// Consumer states
	ConsumerActive *prometheus.GaugeVec

	// Decision tracking
	SwitchCyclesTotal *prometheus.CounterVec
	DecisionDuration  prometheus.Histogram

	// Actuator health
	ActuatorErrors *prometheus.CounterVec
	APILatency     *prometheus.HistogramVec

	// Cumulative
	SelfConsumedKWh prometheus.Counter
	GridExportKWh   prometheus.Counter
	GridImportKWh   prometheus.Counter
}

// NewMetrics creates and registers all EMS metrics.
func NewMetrics(reg prometheus.Registerer) *Metrics {
	m := &Metrics{
		GridPowerW: prometheus.NewGauge(prometheus.GaugeOpts{
			Name: "ems_grid_power_watts",
			Help: "Current grid power exchange in watts (positive=import, negative=export)",
		}),
		BatterySOC: prometheus.NewGauge(prometheus.GaugeOpts{
			Name: "ems_battery_soc_percent",
			Help: "Current battery state of charge in percent",
		}),
		PVProductionW: prometheus.NewGauge(prometheus.GaugeOpts{
			Name: "ems_pv_production_watts",
			Help: "Current PV production in watts",
		}),
		ConsumerActive: prometheus.NewGaugeVec(prometheus.GaugeOpts{
			Name: "ems_consumer_active",
			Help: "Whether a consumer is currently active (1=on, 0=off)",
		}, []string{"consumer"}),
		SwitchCyclesTotal: prometheus.NewCounterVec(prometheus.CounterOpts{
			Name: "ems_switch_cycles_total",
			Help: "Total number of switch cycles per consumer",
		}, []string{"consumer", "action"}),
		DecisionDuration: prometheus.NewHistogram(prometheus.HistogramOpts{
			Name:    "ems_decision_duration_seconds",
			Help:    "Time taken for the decision engine to evaluate",
			Buckets: prometheus.DefBuckets,
		}),
		ActuatorErrors: prometheus.NewCounterVec(prometheus.CounterOpts{
			Name: "ems_actuator_errors_total",
			Help: "Total actuator errors per consumer",
		}, []string{"consumer"}),
		APILatency: prometheus.NewHistogramVec(prometheus.HistogramOpts{
			Name:    "ems_api_latency_seconds",
			Help:    "API call latency per source",
			Buckets: []float64{0.01, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10},
		}, []string{"source"}),
		SelfConsumedKWh: prometheus.NewCounter(prometheus.CounterOpts{
			Name: "ems_self_consumed_kwh_total",
			Help: "Total energy self-consumed in kWh (estimated)",
		}),
		GridExportKWh: prometheus.NewCounter(prometheus.CounterOpts{
			Name: "ems_grid_export_kwh_total",
			Help: "Total energy exported to grid in kWh (estimated)",
		}),
		GridImportKWh: prometheus.NewCounter(prometheus.CounterOpts{
			Name: "ems_grid_import_kwh_total",
			Help: "Total energy imported from grid in kWh (estimated)",
		}),
	}

	reg.MustRegister(
		m.GridPowerW,
		m.BatterySOC,
		m.PVProductionW,
		m.ConsumerActive,
		m.SwitchCyclesTotal,
		m.DecisionDuration,
		m.ActuatorErrors,
		m.APILatency,
		m.SelfConsumedKWh,
		m.GridExportKWh,
		m.GridImportKWh,
	)

	return m
}

// RecordActions records switching actions in the metrics.
func (m *Metrics) RecordActions(actions []engine.Action) {
	for _, a := range actions {
		action := "on"
		if !a.TurnOn {
			action = "off"
		}
		m.SwitchCyclesTotal.WithLabelValues(a.Consumer.String(), action).Inc()
	}
}

// UpdateConsumerStates updates the consumer active gauges.
func (m *Metrics) UpdateConsumerStates(states map[engine.Consumer]bool) {
	for c, active := range states {
		val := 0.0
		if active {
			val = 1.0
		}
		m.ConsumerActive.WithLabelValues(c.String()).Set(val)
	}
}