ananke/internal/metrics/exporter.go

package metrics

import (
	"fmt"
	"net/http"
	"os"
	"sort"
	"strings"
	"sync"
	"time"
)

type Sample struct {
	Name          string
	Target        string
	OnBattery     bool
	LowBattery    bool
	RuntimeSecond int
	BatteryCharge float64
	LoadPercent   float64
	PowerNominalW float64
	ThresholdSec  int
	Trigger       bool
	BreachCount   int
	Status        string
	LastError     string
	UpdatedAt     time.Time
}

type Exporter struct {
	mu                 sync.RWMutex
	shutdownBudgetSec  int
	shutdownTriggers   int
	lastShutdownReason string
	lastShutdownAt     time.Time
	samples            map[string]Sample
}

// New runs one orchestration or CLI step.
// Signature: New() *Exporter.
// Why: keeps behavior explicit so startup/shutdown workflows remain maintainable as services evolve.
func New() *Exporter {
	return &Exporter{
		samples: make(map[string]Sample),
	}
}

// UpdateBudget runs one orchestration or CLI step.
// Signature: (e *Exporter) UpdateBudget(seconds int).
// Why: keeps behavior explicit so startup/shutdown workflows remain maintainable as services evolve.
func (e *Exporter) UpdateBudget(seconds int) {
	e.mu.Lock()
	defer e.mu.Unlock()
	e.shutdownBudgetSec = seconds
}

// UpdateSample runs one orchestration or CLI step.
// Signature: (e *Exporter) UpdateSample(s Sample).
// Why: keeps behavior explicit so startup/shutdown workflows remain maintainable as services evolve.
func (e *Exporter) UpdateSample(s Sample) {
	e.mu.Lock()
	defer e.mu.Unlock()
	if s.UpdatedAt.IsZero() {
		s.UpdatedAt = time.Now().UTC()
	}
	e.samples[s.Name] = s
}

// MarkShutdown runs one orchestration or CLI step.
// Signature: (e *Exporter) MarkShutdown(reason string).
// Why: keeps behavior explicit so startup/shutdown workflows remain maintainable as services evolve.
func (e *Exporter) MarkShutdown(reason string) {
	e.mu.Lock()
	defer e.mu.Unlock()
	e.shutdownTriggers++
	e.lastShutdownReason = reason
	e.lastShutdownAt = time.Now().UTC()
}

// Handler runs one orchestration or CLI step.
// Signature: (e *Exporter) Handler(path string) http.Handler.
// Why: keeps behavior explicit so startup/shutdown workflows remain maintainable as services evolve.
func (e *Exporter) Handler(path string) http.Handler {
	mux := http.NewServeMux()
	metricsPath := path
	if metricsPath == "" {
		metricsPath = "/metrics"
	}
	mux.HandleFunc(metricsPath, e.serveMetrics)
	mux.HandleFunc("/healthz", func(w http.ResponseWriter, _ *http.Request) {
		w.WriteHeader(http.StatusOK)
		_, _ = w.Write([]byte("ok\n"))
	})
	return mux
}

// serveMetrics runs one orchestration or CLI step.
// Signature: (e *Exporter) serveMetrics(w http.ResponseWriter, _ *http.Request).
// Why: keeps behavior explicit so startup/shutdown workflows remain maintainable as services evolve.
func (e *Exporter) serveMetrics(w http.ResponseWriter, _ *http.Request) {
	e.mu.RLock()
	defer e.mu.RUnlock()

	w.Header().Set("Content-Type", "text/plain; version=0.0.4; charset=utf-8")
	var b strings.Builder
	b.WriteString("# HELP ananke_shutdown_budget_seconds Estimated graceful shutdown budget in seconds (p95).\n")
	b.WriteString("# TYPE ananke_shutdown_budget_seconds gauge\n")
	b.WriteString(fmt.Sprintf("ananke_shutdown_budget_seconds %d\n", e.shutdownBudgetSec))
	b.WriteString("# HELP ananke_shutdown_triggers_total Total number of shutdown triggers issued by this instance.\n")
	b.WriteString("# TYPE ananke_shutdown_triggers_total counter\n")
	b.WriteString(fmt.Sprintf("ananke_shutdown_triggers_total %d\n", e.shutdownTriggers))
	b.WriteString("# HELP ananke_shutdown_last_trigger_timestamp_seconds Unix timestamp of last shutdown trigger.\n")
	b.WriteString("# TYPE ananke_shutdown_last_trigger_timestamp_seconds gauge\n")
	if e.lastShutdownAt.IsZero() {
		b.WriteString("ananke_shutdown_last_trigger_timestamp_seconds 0\n")
	} else {
		b.WriteString(fmt.Sprintf("ananke_shutdown_last_trigger_timestamp_seconds %d\n", e.lastShutdownAt.Unix()))
	}
	b.WriteString("# HELP ananke_ups_on_battery Whether a UPS source is currently on battery.\n")
	b.WriteString("# TYPE ananke_ups_on_battery gauge\n")
	b.WriteString("# HELP ananke_ups_low_battery Whether a UPS source currently reports low battery.\n")
	b.WriteString("# TYPE ananke_ups_low_battery gauge\n")
	b.WriteString("# HELP ananke_ups_runtime_seconds Battery runtime remaining reported by UPS.\n")
	b.WriteString("# TYPE ananke_ups_runtime_seconds gauge\n")
	b.WriteString("# HELP ananke_ups_battery_charge_percent Battery charge percentage reported by UPS.\n")
	b.WriteString("# TYPE ananke_ups_battery_charge_percent gauge\n")
	b.WriteString("# HELP ananke_ups_load_percent UPS output load percentage.\n")
	b.WriteString("# TYPE ananke_ups_load_percent gauge\n")
	b.WriteString("# HELP ananke_ups_power_nominal_watts UPS nominal power rating in watts.\n")
	b.WriteString("# TYPE ananke_ups_power_nominal_watts gauge\n")
	b.WriteString("# HELP ananke_ups_threshold_seconds Red-line threshold for runtime-based shutdown.\n")
	b.WriteString("# TYPE ananke_ups_threshold_seconds gauge\n")
	b.WriteString("# HELP ananke_ups_trigger_active Whether this UPS source currently breaches shutdown trigger conditions.\n")
	b.WriteString("# TYPE ananke_ups_trigger_active gauge\n")
	b.WriteString("# HELP ananke_ups_breach_count Current debounce breach count for this UPS source.\n")
	b.WriteString("# TYPE ananke_ups_breach_count gauge\n")
	b.WriteString("# HELP ananke_ups_last_sample_timestamp_seconds Unix timestamp of most recent sample.\n")
	b.WriteString("# TYPE ananke_ups_last_sample_timestamp_seconds gauge\n")
	b.WriteString("# HELP ananke_ups_error Whether the last sample had an error.\n")
	b.WriteString("# TYPE ananke_ups_error gauge\n")

	names := make([]string, 0, len(e.samples))
	for name := range e.samples {
		names = append(names, name)
	}
	sort.Strings(names)
	for _, name := range names {
		s := e.samples[name]
		labels := fmt.Sprintf("{source=%q,target=%q,status=%q,last_reason=%q}",
			safe(name), safe(s.Target), safe(s.Status), safe(e.lastShutdownReason))
		b.WriteString(fmt.Sprintf("ananke_ups_on_battery%s %d\n", labels, boolNum(s.OnBattery)))
		b.WriteString(fmt.Sprintf("ananke_ups_low_battery%s %d\n", labels, boolNum(s.LowBattery)))
		b.WriteString(fmt.Sprintf("ananke_ups_runtime_seconds%s %d\n", labels, s.RuntimeSecond))
		b.WriteString(fmt.Sprintf("ananke_ups_battery_charge_percent%s %.2f\n", labels, s.BatteryCharge))
		b.WriteString(fmt.Sprintf("ananke_ups_load_percent%s %.2f\n", labels, s.LoadPercent))
		b.WriteString(fmt.Sprintf("ananke_ups_power_nominal_watts%s %.2f\n", labels, s.PowerNominalW))
		b.WriteString(fmt.Sprintf("ananke_ups_threshold_seconds%s %d\n", labels, s.ThresholdSec))
		b.WriteString(fmt.Sprintf("ananke_ups_trigger_active%s %d\n", labels, boolNum(s.Trigger)))
		b.WriteString(fmt.Sprintf("ananke_ups_breach_count%s %d\n", labels, s.BreachCount))
		if s.UpdatedAt.IsZero() {
			b.WriteString(fmt.Sprintf("ananke_ups_last_sample_timestamp_seconds%s 0\n", labels))
		} else {
			b.WriteString(fmt.Sprintf("ananke_ups_last_sample_timestamp_seconds%s %d\n", labels, s.UpdatedAt.Unix()))
		}
		b.WriteString(fmt.Sprintf("ananke_ups_error%s %d\n", labels, boolNum(s.LastError != "")))
	}
	appendQualityGateMetrics(&b)

	_, _ = w.Write([]byte(b.String()))
}

// appendQualityGateMetrics runs one orchestration or CLI step.
// Signature: appendQualityGateMetrics(dst *strings.Builder).
// Why: quality-gate pass/fail telemetry should appear alongside UPS metrics so
// Grafana can track Ananke suite health over time.
func appendQualityGateMetrics(dst *strings.Builder) {
	path := strings.TrimSpace(os.Getenv("ANANKE_QUALITY_METRICS_FILE"))
	if path == "" {
		path = "/var/lib/ananke/quality-gate.prom"
	}
	raw, err := os.ReadFile(path)
	if err != nil {
		return
	}
	text := strings.TrimSpace(string(raw))
	if text == "" {
		return
	}
	if dst.Len() > 0 {
		dst.WriteString("\n")
	}
	dst.WriteString(text)
	if !strings.HasSuffix(text, "\n") {
		dst.WriteString("\n")
	}
}

// boolNum runs one orchestration or CLI step.
// Signature: boolNum(v bool) int.
// Why: keeps behavior explicit so startup/shutdown workflows remain maintainable as services evolve.
func boolNum(v bool) int {
	if v {
		return 1
	}
	return 0
}

// safe runs one orchestration or CLI step.
// Signature: safe(in string) string.
// Why: keeps behavior explicit so startup/shutdown workflows remain maintainable as services evolve.
func safe(in string) string {
	out := strings.ReplaceAll(in, "\\", "\\\\")
	return strings.ReplaceAll(out, "\"", "\\\"")
}