nixos-exporter/collector/generation.go

package collector

import (
	"log/slog"
	"os"
	"path/filepath"
	"regexp"
	"strconv"
	"time"

	"github.com/prometheus/client_golang/prometheus"
)

const (
	profileDir       = "/nix/var/nix/profiles"
	currentSystemDir = "/run/current-system"
	bootedSystemDir  = "/run/booted-system"
)

var generationPattern = regexp.MustCompile(`^system-(\d+)-link$`)

type GenerationCollector struct {
	generationCount *prometheus.Desc
	currentGen      *prometheus.Desc
	bootedGen       *prometheus.Desc
	generationAge   *prometheus.Desc
	configMismatch  *prometheus.Desc
}

func NewGenerationCollector() *GenerationCollector {
	return &GenerationCollector{
		generationCount: prometheus.NewDesc(
			"nixos_generation_count",
			"Total number of system generations",
			nil, nil,
		),
		currentGen: prometheus.NewDesc(
			"nixos_current_generation",
			"Currently active generation number",
			nil, nil,
		),
		bootedGen: prometheus.NewDesc(
			"nixos_booted_generation",
			"Generation that was booted",
			nil, nil,
		),
		generationAge: prometheus.NewDesc(
			"nixos_generation_age_seconds",
			"Age of current generation in seconds",
			nil, nil,
		),
		configMismatch: prometheus.NewDesc(
			"nixos_config_mismatch",
			"1 if booted generation differs from current",
			nil, nil,
		),
	}
}

func (c *GenerationCollector) Describe(ch chan<- *prometheus.Desc) {
	ch <- c.generationCount
	ch <- c.currentGen
	ch <- c.bootedGen
	ch <- c.generationAge
	ch <- c.configMismatch
}

func (c *GenerationCollector) Collect(ch chan<- prometheus.Metric) {
	c.collectGenerationCount(ch)
	c.collectCurrentGeneration(ch)
	c.collectBootedGeneration(ch)
	c.collectGenerationAge(ch)
	c.collectConfigMismatch(ch)
}

func (c *GenerationCollector) collectGenerationCount(ch chan<- prometheus.Metric) {
	count, err := countGenerations(profileDir)
	if err != nil {
		slog.Error("Failed to count generations", "error", err)
		return
	}
	ch <- prometheus.MustNewConstMetric(c.generationCount, prometheus.GaugeValue, float64(count))
}

func (c *GenerationCollector) collectCurrentGeneration(ch chan<- prometheus.Metric) {
	gen, err := getCurrentGeneration(profileDir)
	if err != nil {
		slog.Error("Failed to get current generation", "error", err)
		return
	}
	ch <- prometheus.MustNewConstMetric(c.currentGen, prometheus.GaugeValue, float64(gen))
}

func (c *GenerationCollector) collectBootedGeneration(ch chan<- prometheus.Metric) {
	gen, err := getBootedGeneration(profileDir, bootedSystemDir)
	if err != nil {
		slog.Error("Failed to get booted generation", "error", err)
		return
	}
	ch <- prometheus.MustNewConstMetric(c.bootedGen, prometheus.GaugeValue, float64(gen))
}

func (c *GenerationCollector) collectGenerationAge(ch chan<- prometheus.Metric) {
	age, err := getGenerationAge(profileDir)
	if err != nil {
		slog.Error("Failed to get generation age", "error", err)
		return
	}
	ch <- prometheus.MustNewConstMetric(c.generationAge, prometheus.GaugeValue, age)
}

func (c *GenerationCollector) collectConfigMismatch(ch chan<- prometheus.Metric) {
	mismatch, err := checkConfigMismatch(currentSystemDir, bootedSystemDir)
	if err != nil {
		slog.Error("Failed to check config mismatch", "error", err)
		return
	}
	value := 0.0
	if mismatch {
		value = 1.0
	}
	ch <- prometheus.MustNewConstMetric(c.configMismatch, prometheus.GaugeValue, value)
}

// countGenerations counts system-*-link entries in the profile directory.
func countGenerations(profileDir string) (int, error) {
	entries, err := os.ReadDir(profileDir)
	if err != nil {
		return 0, err
	}

	count := 0
	for _, entry := range entries {
		if generationPattern.MatchString(entry.Name()) {
			count++
		}
	}
	return count, nil
}

// getCurrentGeneration parses the generation number from the system symlink.
func getCurrentGeneration(profileDir string) (int, error) {
	systemLink := filepath.Join(profileDir, "system")
	target, err := os.Readlink(systemLink)
	if err != nil {
		return 0, err
	}

	// Target is relative like "system-123-link"
	base := filepath.Base(target)
	matches := generationPattern.FindStringSubmatch(base)
	if matches == nil {
		return 0, nil
	}

	return strconv.Atoi(matches[1])
}

// getBootedGeneration finds the generation that matches /run/booted-system.
func getBootedGeneration(profileDir, bootedSystemDir string) (int, error) {
	bootedTarget, err := os.Readlink(bootedSystemDir)
	if err != nil {
		return 0, err
	}

	entries, err := os.ReadDir(profileDir)
	if err != nil {
		return 0, err
	}

	for _, entry := range entries {
		if !generationPattern.MatchString(entry.Name()) {
			continue
		}

		linkPath := filepath.Join(profileDir, entry.Name())
		target, err := os.Readlink(linkPath)
		if err != nil {
			continue
		}

		if target == bootedTarget {
			matches := generationPattern.FindStringSubmatch(entry.Name())
			if matches != nil {
				return strconv.Atoi(matches[1])
			}
		}
	}

	return 0, nil
}

// getGenerationAge returns the age of the current system profile in seconds.
func getGenerationAge(profileDir string) (float64, error) {
	systemLink := filepath.Join(profileDir, "system")
	info, err := os.Lstat(systemLink)
	if err != nil {
		return 0, err
	}

	age := time.Since(info.ModTime()).Seconds()
	return age, nil
}

// checkConfigMismatch compares /run/current-system and /run/booted-system targets.
func checkConfigMismatch(currentSystemDir, bootedSystemDir string) (bool, error) {
	currentTarget, err := os.Readlink(currentSystemDir)
	if err != nil {
		return false, err
	}

	bootedTarget, err := os.Readlink(bootedSystemDir)
	if err != nil {
		return false, err
	}

	return currentTarget != bootedTarget, nil
}