metis/pkg/service/metrics.go

package service

import (
	"fmt"
	"io"
	"sort"
	"strings"
	"sync"
	"time"

	"metis/pkg/facts"
)

// Metrics captures the small Prometheus surface exported by Metis.
type Metrics struct {
	mu sync.RWMutex

	builds              map[string]int
	flashes             map[string]int
	snapshots           map[string]int
	lastSnapshotUnix    map[string]float64
	watches             map[string]int
	lastWatchSuccess    float64
	classDriftCounts    map[string]int
	lastWatchChangeSize float64
}

// NewMetrics builds a zero-value metrics registry.
func NewMetrics() *Metrics {
	return &Metrics{
		builds:           map[string]int{},
		flashes:          map[string]int{},
		snapshots:        map[string]int{},
		lastSnapshotUnix: map[string]float64{},
		watches:          map[string]int{},
		classDriftCounts: map[string]int{},
	}
}

// RecordBuild increments the per-node build counter because the UI and
// Prometheus graphs need a stable view of build outcomes by node.
func (m *Metrics) RecordBuild(node, status string) {
	m.mu.Lock()
	defer m.mu.Unlock()
	m.builds[counterKey(node, status)]++
}

// RecordFlash increments the per-node and per-host flash counter because the
// replacement workflow needs separate visibility for build and burn stages.
func (m *Metrics) RecordFlash(node, host, status string) {
	m.mu.Lock()
	defer m.mu.Unlock()
	m.flashes[counterKey(node, host, status)]++
}

// RecordSnapshot tracks accepted sentinel snapshots because drift detection
// depends on the last successful push per node.
func (m *Metrics) RecordSnapshot(node, status string, ts time.Time) {
	m.mu.Lock()
	defer m.mu.Unlock()
	m.snapshots[counterKey(node, status)]++
	if !ts.IsZero() {
		m.lastSnapshotUnix[node] = float64(ts.Unix())
	}
}

// RecordWatch increments the sentinel watch outcome counter because the
// dashboard needs to show whether the latest reconciliation succeeded.
func (m *Metrics) RecordWatch(status string) {
	m.mu.Lock()
	defer m.mu.Unlock()
	m.watches[counterKey(status)]++
	if status == "ok" {
		m.lastWatchSuccess = float64(time.Now().UTC().Unix())
	}
}

// SetDriftTargets refreshes the target-count gauge because the UI exposes how
// much class configuration is already populated versus still missing.
func (m *Metrics) SetDriftTargets(targets map[string]facts.Targets, changed int) {
	m.mu.Lock()
	defer m.mu.Unlock()
	m.classDriftCounts = map[string]int{}
	for class, target := range targets {
		count := 0
		if strings.TrimSpace(target.Kernel) != "" {
			count++
		}
		if strings.TrimSpace(target.OSImage) != "" {
			count++
		}
		if strings.TrimSpace(target.Containerd) != "" {
			count++
		}
		if strings.TrimSpace(target.K3sVersion) != "" {
			count++
		}
		count += len(target.Packages)
		m.classDriftCounts[class] = count
	}
	m.lastWatchChangeSize = float64(changed)
}

// Render writes a Prometheus text exposition response.
func (m *Metrics) Render(w io.Writer) {
	m.mu.RLock()
	defer m.mu.RUnlock()

	fmt.Fprintln(w, "# HELP metis_builds_total Replacement image builds by node and status")
	fmt.Fprintln(w, "# TYPE metis_builds_total counter")
	for _, key := range sortedKeys(m.builds) {
		parts := splitKey(key, 2)
		node, status := parts[0], parts[1]
		fmt.Fprintf(w, "metis_builds_total{node=%q,status=%q} %d\n", node, status, m.builds[key])
	}

	fmt.Fprintln(w, "# HELP metis_flashes_total Replacement flashes by node, host, and status")
	fmt.Fprintln(w, "# TYPE metis_flashes_total counter")
	for _, key := range sortedKeys(m.flashes) {
		parts := splitKey(key, 3)
		node, host, status := parts[0], parts[1], parts[2]
		fmt.Fprintf(w, "metis_flashes_total{node=%q,host=%q,status=%q} %d\n", node, host, status, m.flashes[key])
	}

	fmt.Fprintln(w, "# HELP metis_sentinel_snapshots_total Sentinel snapshots accepted by node and status")
	fmt.Fprintln(w, "# TYPE metis_sentinel_snapshots_total counter")
	for _, key := range sortedKeys(m.snapshots) {
		parts := splitKey(key, 2)
		node, status := parts[0], parts[1]
		fmt.Fprintf(w, "metis_sentinel_snapshots_total{node=%q,status=%q} %d\n", node, status, m.snapshots[key])
	}

	fmt.Fprintln(w, "# HELP metis_sentinel_snapshot_timestamp_seconds Last accepted sentinel snapshot timestamp by node")
	fmt.Fprintln(w, "# TYPE metis_sentinel_snapshot_timestamp_seconds gauge")
	for _, node := range sortedFloatKeys(m.lastSnapshotUnix) {
		fmt.Fprintf(w, "metis_sentinel_snapshot_timestamp_seconds{node=%q} %.0f\n", node, m.lastSnapshotUnix[node])
	}

	fmt.Fprintln(w, "# HELP metis_sentinel_watch_total Sentinel watch runs by status")
	fmt.Fprintln(w, "# TYPE metis_sentinel_watch_total counter")
	for _, key := range sortedKeys(m.watches) {
		status := splitKey(key, 1)[0]
		fmt.Fprintf(w, "metis_sentinel_watch_total{status=%q} %d\n", status, m.watches[key])
	}

	fmt.Fprintln(w, "# HELP metis_sentinel_watch_last_success_timestamp_seconds Last successful sentinel watch timestamp")
	fmt.Fprintln(w, "# TYPE metis_sentinel_watch_last_success_timestamp_seconds gauge")
	fmt.Fprintf(w, "metis_sentinel_watch_last_success_timestamp_seconds %.0f\n", m.lastWatchSuccess)

	fmt.Fprintln(w, "# HELP metis_sentinel_watch_changed_classes Number of class target sets changed by the last watch")
	fmt.Fprintln(w, "# TYPE metis_sentinel_watch_changed_classes gauge")
	fmt.Fprintf(w, "metis_sentinel_watch_changed_classes %.0f\n", m.lastWatchChangeSize)

	fmt.Fprintln(w, "# HELP metis_class_target_fields Count of populated target fields per class")
	fmt.Fprintln(w, "# TYPE metis_class_target_fields gauge")
	for _, class := range sortedFloatKeysInt(m.classDriftCounts) {
		fmt.Fprintf(w, "metis_class_target_fields{class=%q} %d\n", class, m.classDriftCounts[class])
	}
}

func counterKey(parts ...string) string {
	return strings.Join(parts, "\x00")
}

func splitKey(key string, want int) []string {
	parts := strings.Split(key, "\x00")
	for len(parts) < want {
		parts = append(parts, "")
	}
	return parts
}

func sortedKeys[T any](m map[string]T) []string {
	keys := make([]string, 0, len(m))
	for key := range m {
		keys = append(keys, key)
	}
	sort.Strings(keys)
	return keys
}

func sortedFloatKeys(m map[string]float64) []string {
	keys := make([]string, 0, len(m))
	for key := range m {
		keys = append(keys, key)
	}
	sort.Strings(keys)
	return keys
}

func sortedFloatKeysInt(m map[string]int) []string {
	keys := make([]string, 0, len(m))
	for key := range m {
		keys = append(keys, key)
	}
	sort.Strings(keys)
	return keys
}