sys-exporter/collector/systemd/systemd.go

package systemd

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

	"git.faercol.me/monitoring/sys-exporter/registry"
	"github.com/prometheus/client_golang/prometheus"
	"go.uber.org/zap"
)

type serviceStatus struct {
	Unit        string `json:"unit"`
	Load        string `json:"load"`
	Active      string `json:"active"`
	Sub         string `json:"sub"`
	Description string `json:"description"`
}

func (s serviceStatus) unitType() string {
	split := strings.Split(s.Unit, ".")
	if len(split) < 2 {
		return ""
	}
	return split[len(split)-1]
}

type SystemdCollector struct {
	promExporter *prometheus.GaugeVec
	updateFreq   time.Duration

	l *zap.SugaredLogger
}

func (c *SystemdCollector) Collect() interface{} {
	return nil
}

func (c *SystemdCollector) getServicesStatus() ([]serviceStatus, error) {
	out, err := exec.Command("/sbin/systemctl", "list-units", "--output", "json").CombinedOutput()
	if err != nil {
		return nil, fmt.Errorf("failed to run systemd command: %w", err)
	}

	res := []serviceStatus{}
	if err := json.Unmarshal(out, &res); err != nil {
		return nil, fmt.Errorf("failed to parse systemd output: %w", err)
	}

	return res, nil
}

func (c *SystemdCollector) update() error {
	systemd_output, err := c.getServicesStatus()
	if err != nil {
		return fmt.Errorf("failed to get services status: %w", err)
	}

	mappingPerType := map[string][]serviceStatus{}
	for _, s := range systemd_output {
		mappingPerType[s.unitType()] = append(mappingPerType[s.unitType()], s)
	}

	totalUnits := 0
	runningUnits := 0
	failedUnits := 0

	for unitType, units := range mappingPerType {

		unitsForType := 0
		healthyForType := 0
		failedForType := 0

		for _, u := range units {
			unitsForType++
			totalUnits++
			if u.Active == "active" {
				runningUnits++
				healthyForType++
			}
			if u.Active == "failed" {
				failedUnits++
				failedForType++
			}
		}

		c.promExporter.WithLabelValues(unitType, "total").Set(float64(unitsForType))
		c.promExporter.WithLabelValues(unitType, "running").Set(float64(healthyForType))
		c.promExporter.WithLabelValues(unitType, "failed").Set(float64(failedForType))
	}

	c.promExporter.WithLabelValues("total", "total").Set(float64(totalUnits))
	c.promExporter.WithLabelValues("total", "running").Set(float64(runningUnits))
	c.promExporter.WithLabelValues("total", "failed").Set(float64(failedUnits))

	return nil
}

func (c *SystemdCollector) Run(ctx context.Context, l *zap.SugaredLogger) {
	c.l = l

	c.l.Debug("Initializing collector")
	if err := c.update(); err != nil {
		c.l.Errorf("Failed to init collector: %s\n", err)
	}

	for {
		select {
		case <-ctx.Done():
			c.l.Debug("Stopping collector")
			return
		case <-time.After(c.updateFreq):
			c.l.Debug("Updating collector")
			if err := c.update(); err != nil {
				c.l.Errorf("Failed to update collector: %s\n", err)
			}
		}
	}
}

func (c *SystemdCollector) PromCollector() prometheus.Collector {
	return c.promExporter
}

func New() *SystemdCollector {
	return &SystemdCollector{
		updateFreq:   30 * time.Second,
		promExporter: prometheus.NewGaugeVec(prometheus.GaugeOpts{Namespace: "services", Subsystem: "systemd", Name: "running_units", Help: "Count of running services for systemd"}, []string{"type", "status"}),
	}
}

func init() {
	registry.R.MustRegisterCollector("services.systemd", New())
}