nixos-exporter/collector/flake.go

package collector

import (
	"context"
	"encoding/json"
	"fmt"
	"log/slog"
	"os"
	"os/exec"
	"strings"
	"sync"
	"time"

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

const configRevisionPath = "/run/current-system/configuration-revision"

type FlakeCollector struct {
	flakeURL      string
	checkInterval time.Duration

	inputAge       *prometheus.Desc
	inputInfo      *prometheus.Desc
	flakeInfo      *prometheus.Desc
	revisionBehind *prometheus.Desc

	mu          sync.RWMutex
	cachedData  *flakeMetadata
	lastFetch   time.Time
	fetchError  error
}

type flakeMetadata struct {
	Revision string                       `json:"revision"`
	Locks    flakeLocks                   `json:"locks"`
}

type flakeLocks struct {
	Nodes map[string]flakeLockNode `json:"nodes"`
	Root  string                   `json:"root"`
}

type flakeLockNode struct {
	Inputs   map[string]interface{} `json:"inputs,omitempty"`
	Locked   *lockedInfo            `json:"locked,omitempty"`
	Original *originalInfo          `json:"original,omitempty"`
}

type lockedInfo struct {
	LastModified int64  `json:"lastModified"`
	Rev          string `json:"rev"`
	Type         string `json:"type"`
}

type originalInfo struct {
	Type string `json:"type"`
}

func NewFlakeCollector(flakeURL string, checkInterval time.Duration) *FlakeCollector {
	return &FlakeCollector{
		flakeURL:      flakeURL,
		checkInterval: checkInterval,
		inputAge: prometheus.NewDesc(
			"nixos_flake_input_age_seconds",
			"Age of flake input in seconds",
			[]string{"input"}, nil,
		),
		inputInfo: prometheus.NewDesc(
			"nixos_flake_input_info",
			"Info gauge with revision and type labels",
			[]string{"input", "rev", "type"}, nil,
		),
		flakeInfo: prometheus.NewDesc(
			"nixos_flake_info",
			"Info gauge with current and remote flake revisions",
			[]string{"current_rev", "remote_rev"}, nil,
		),
		revisionBehind: prometheus.NewDesc(
			"nixos_flake_revision_behind",
			"1 if current system revision differs from remote latest, 0 if match",
			nil, nil,
		),
	}
}

func (c *FlakeCollector) Describe(ch chan<- *prometheus.Desc) {
	ch <- c.inputAge
	ch <- c.inputInfo
	ch <- c.flakeInfo
	ch <- c.revisionBehind
}

func (c *FlakeCollector) Collect(ch chan<- prometheus.Metric) {
	data, err := c.getFlakeData()
	if err != nil {
		slog.Error("Failed to get flake data", "error", err)
		return
	}

	c.collectInputMetrics(ch, data)
	c.collectRevisionBehind(ch, data)
}

func (c *FlakeCollector) getFlakeData() (*flakeMetadata, error) {
	c.mu.RLock()
	if c.cachedData != nil && time.Since(c.lastFetch) < c.checkInterval {
		data := c.cachedData
		c.mu.RUnlock()
		return data, nil
	}
	c.mu.RUnlock()

	c.mu.Lock()
	defer c.mu.Unlock()

	// Double-check after acquiring write lock
	if c.cachedData != nil && time.Since(c.lastFetch) < c.checkInterval {
		return c.cachedData, nil
	}

	data, err := fetchFlakeMetadata(c.flakeURL)
	if err != nil {
		c.fetchError = err
		// Return cached data if available, even if stale
		if c.cachedData != nil {
			slog.Warn("Using stale flake data due to fetch error", "error", err)
			return c.cachedData, nil
		}
		return nil, err
	}

	c.cachedData = data
	c.lastFetch = time.Now()
	c.fetchError = nil
	return data, nil
}

func (c *FlakeCollector) collectInputMetrics(ch chan<- prometheus.Metric, data *flakeMetadata) {
	now := time.Now().Unix()

	for name, node := range data.Locks.Nodes {
		// Skip the root node
		if name == "root" {
			continue
		}

		if node.Locked == nil {
			continue
		}

		// Input age
		age := float64(now - node.Locked.LastModified)
		ch <- prometheus.MustNewConstMetric(c.inputAge, prometheus.GaugeValue, age, name)

		// Input info
		rev := node.Locked.Rev
		if len(rev) > 7 {
			rev = rev[:7]
		}
		inputType := node.Locked.Type
		ch <- prometheus.MustNewConstMetric(c.inputInfo, prometheus.GaugeValue, 1, name, rev, inputType)
	}
}

func (c *FlakeCollector) collectRevisionBehind(ch chan<- prometheus.Metric, data *flakeMetadata) {
	currentRev, err := getCurrentSystemRevision()
	if err != nil {
		slog.Error("Failed to get current system revision", "error", err)
		return
	}

	remoteRev := data.Revision
	if len(remoteRev) > 7 {
		remoteRev = remoteRev[:7]
	}

	// Emit flake info metric with revisions
	ch <- prometheus.MustNewConstMetric(c.flakeInfo, prometheus.GaugeValue, 1, currentRev, remoteRev)

	behind := 0.0
	if currentRev != "" && data.Revision != "" {
		if currentRev != remoteRev && !strings.HasPrefix(data.Revision, currentRev) {
			behind = 1.0
		}
	}

	ch <- prometheus.MustNewConstMetric(c.revisionBehind, prometheus.GaugeValue, behind)
}

func fetchFlakeMetadata(flakeURL string) (*flakeMetadata, error) {
	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
	defer cancel()

	cmd := exec.CommandContext(ctx, "nix", "flake", "metadata", "--json", flakeURL)
	output, err := cmd.Output()
	if err != nil {
		if exitErr, ok := err.(*exec.ExitError); ok {
			return nil, fmt.Errorf("nix flake metadata failed: %s", strings.TrimSpace(string(exitErr.Stderr)))
		}
		return nil, fmt.Errorf("nix flake metadata failed: %w", err)
	}

	var data flakeMetadata
	if err := json.Unmarshal(output, &data); err != nil {
		return nil, fmt.Errorf("failed to parse flake metadata: %w", err)
	}

	return &data, nil
}

func getCurrentSystemRevision() (string, error) {
	data, err := os.ReadFile(configRevisionPath)
	if err != nil {
		if os.IsNotExist(err) {
			// configuration-revision doesn't exist; user hasn't set system.configurationRevision
			return "", nil
		}
		return "", err
	}

	rev := strings.TrimSpace(string(data))
	if len(rev) > 7 {
		rev = rev[:7]
	}
	return rev, nil
}