Merge pull request #15 from utilitywarehouse/init-counters

metrics: initialize counters

main.go

@@ -9,10 +9,8 @@ import (
 	"regexp"
 	"strings"
 
-	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
 	"golang.zx2c4.com/wireguard/wgctrl"
-	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
 
 	"github.com/utilitywarehouse/semaphore-wireguard/kube"
 	"github.com/utilitywarehouse/semaphore-wireguard/log"
@@ -107,7 +105,7 @@ func main() {
 		}
 	}()
 
-	makeMetricsCollector(wgMetricsClient, wgDeviceNames)
+	registerMetrics(wgMetricsClient, wgDeviceNames)
 	listenAndServe(runners)
 
 	// Stop runners before finishing
@@ -158,22 +156,6 @@ func makeRunner(homeClient kubernetes.Interface, localName string, rConf *remote
 	return r, wgDeviceName, nil
 }
 
-func makeMetricsCollector(wgMetricsClient *wgctrl.Client, wgDeviceNames []string) {
-	mc := newMetricsCollector(func() ([]*wgtypes.Device, error) {
-		var devices []*wgtypes.Device
-		for _, name := range wgDeviceNames {
-			device, err := wgMetricsClient.Device(name)
-			if err != nil {
-				return nil, err
-			}
-			devices = append(devices, device)
-		}
-		return devices, nil
-	})
-	prometheus.MustRegister(mc)
-
-}
-
 func listenAndServe(runners []*Runner) {
 	mux := http.NewServeMux()
 	mux.Handle("/metrics", promhttp.Handler())

metrics.go

@@ -3,33 +3,65 @@ package main
 import (
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/utilitywarehouse/semaphore-wireguard/log"
+	"golang.zx2c4.com/wireguard/wgctrl"
 	"golang.zx2c4.com/wireguard/wgctrl/wgtypes"
 )
 
 var (
 	syncPeersAttempt = prometheus.NewCounterVec(
 		prometheus.CounterOpts{
-			Name: "semaphore_wg_sync_peers",
+			Name: "semaphore_wg_sync_peers_total",
 			Help: "Counts runners' attempts to sync peers.",
 		},
 		[]string{"device", "success"},
 	)
 	syncQueueFullFailures = prometheus.NewCounterVec(
 		prometheus.CounterOpts{
-			Name: "semaphore_wg_sync_queue_full_failures",
+			Name: "semaphore_wg_sync_queue_full_failures_total",
 			Help: "Number of times a sync task was not added to the sync queue in time because the queue was full.",
 		},
 		[]string{"device"},
 	)
 	syncRequeue = prometheus.NewCounterVec(
 		prometheus.CounterOpts{
-			Name: "semaphore_wg_sync_requeue",
+			Name: "semaphore_wg_sync_requeue_total",
 			Help: "Number of attempts to requeue a sync.",
 		},
 		[]string{"device"},
 	)
 )
 
+// registerMetrics registers all the prometheus collectors for this package
+func registerMetrics(wgMetricsClient *wgctrl.Client, wgDeviceNames []string) {
+	// Initialize counters with a value of 0
+	for _, d := range wgDeviceNames {
+		for _, s := range []string{"0", "1"} {
+			syncPeersAttempt.With(prometheus.Labels{"device": d, "success": s})
+		}
+		syncQueueFullFailures.With(prometheus.Labels{"device": d})
+		syncRequeue.With(prometheus.Labels{"device": d})
+	}
+
+	mc := newMetricsCollector(func() ([]*wgtypes.Device, error) {
+		var devices []*wgtypes.Device
+		for _, name := range wgDeviceNames {
+			device, err := wgMetricsClient.Device(name)
+			if err != nil {
+				return nil, err
+			}
+			devices = append(devices, device)
+		}
+		return devices, nil
+	})
+
+	prometheus.MustRegister(
+		mc,
+		syncPeersAttempt,
+		syncQueueFullFailures,
+		syncRequeue,
+	)
+}
+
 // A collector is a prometheus.Collector for a WireGuard device.
 type collector struct {
 	DeviceInfo         *prometheus.Desc
@@ -42,13 +74,7 @@ type collector struct {
 	devices func() ([]*wgtypes.Device, error) // to allow testing
 }
 
-func init() {
-	prometheus.MustRegister(syncPeersAttempt)
-	prometheus.MustRegister(syncQueueFullFailures)
-	prometheus.MustRegister(syncRequeue)
-}
-
-// NewMetricsCollector constructs a prometheus.Collector to collect metrics for
+// newMetricsCollector constructs a prometheus.Collector to collect metrics for
 // all present wg devices and correlate with user if possible
 func newMetricsCollector(devices func() ([]*wgtypes.Device, error)) prometheus.Collector {
 	// common labels for all metrics
@@ -182,7 +208,7 @@ func (c *collector) Collect(ch chan<- prometheus.Metric) {
 	}
 }
 
-func MetricsSyncPeerAttempt(device string, err error) {
+func metricsSyncPeerAttempt(device string, err error) {
 	s := "1"
 	if err != nil {
 		s = "0"
@@ -193,13 +219,13 @@ func MetricsSyncPeerAttempt(device string, err error) {
 	}).Inc()
 }
 
-func MetricsIncSyncQueueFullFailures(device string) {
+func metricsIncSyncQueueFullFailures(device string) {
 	syncQueueFullFailures.With(prometheus.Labels{
 		"device": device,
 	}).Inc()
 }
 
-func MetricsIncSyncRequeue(device string) {
+func metricsIncSyncRequeue(device string) {
 	syncRequeue.With(prometheus.Labels{
 		"device": device,
 	}).Inc()

runner.go

@@ -123,7 +123,7 @@ func (r *Runner) syncLoop() {
 				continue
 			}
 			err := r.syncPeers()
-			MetricsSyncPeerAttempt(r.device.Name(), err)
+			metricsSyncPeerAttempt(r.device.Name(), err)
 			if err != nil {
 				log.Logger.Warn("Failed to sync wg peers", "err", err)
 				r.requeuePeersSync()
@@ -165,7 +165,7 @@ func (r *Runner) enqueuePeersSync() {
 		log.Logger.Debug("Sync task queued")
 	case <-time.After(5 * time.Second):
 		log.Logger.Error("Timed out trying to queue a sync action for netset, sync queue is full")
-		MetricsIncSyncQueueFullFailures(r.device.Name())
+		metricsIncSyncQueueFullFailures(r.device.Name())
 		r.requeuePeersSync()
 	}
 }
@@ -174,7 +174,7 @@ func (r *Runner) requeuePeersSync() {
 	log.Logger.Debug("Requeueing peers sync task")
 	go func() {
 		time.Sleep(1)
-		MetricsIncSyncRequeue(r.device.Name())
+		metricsIncSyncRequeue(r.device.Name())
 		r.enqueuePeersSync()
 	}()
 }