Live: collect more usage stats – enabled, ha enabled, num nodes, channels, subs (#47378)

pkg/services/live/live.go

@@ -9,6 +9,7 @@ import (
 	"net/http"
 	"net/url"
 	"os"
+	"strconv"
 	"strings"
 	"sync"
 	"time"
@@ -1365,6 +1366,11 @@ func handleLog(msg centrifuge.LogEntry) {
 func (g *GrafanaLive) sampleLiveStats() {
 	numClients := g.node.Hub().NumClients()
 	numUsers := g.node.Hub().NumUsers()
+	numChannels := g.node.Hub().NumChannels()
+	var numNodes int
+	if info, err := g.node.Info(); err == nil {
+		numNodes = len(info.Nodes)
+	}
 
 	g.usageStats.sampleCount++
 	g.usageStats.numClientsSum += numClients
@@ -1374,16 +1380,16 @@ func (g *GrafanaLive) sampleLiveStats() {
 		g.usageStats.numClientsMax = numClients
 	}
 
-	if numClients < g.usageStats.numClientsMin {
-		g.usageStats.numClientsMin = numClients
-	}
-
 	if numUsers > g.usageStats.numUsersMax {
 		g.usageStats.numUsersMax = numUsers
 	}
 
-	if numUsers < g.usageStats.numUsersMin {
+	if numNodes > g.usageStats.numNodesMax {
-		g.usageStats.numUsersMin = numUsers
+		g.usageStats.numNodesMax = numNodes
+	}
+
+	if numChannels > g.usageStats.numChannelsMax {
+		g.usageStats.numChannelsMax = numChannels
 	}
 }
 
@@ -1391,38 +1397,65 @@ func (g *GrafanaLive) resetLiveStats() {
 	g.usageStats = usageStats{}
 }
 
+func getHistogramMetric(val int, bounds []int, metricPrefix string) string {
+	for _, bound := range bounds {
+		if val <= bound {
+			return metricPrefix + "le_" + strconv.Itoa(bound)
+		}
+	}
+	return metricPrefix + "le_inf"
+}
+
+func (g *GrafanaLive) collectLiveStats(_ context.Context) (map[string]interface{}, error) {
+	liveUsersAvg := 0
+	liveClientsAvg := 0
+
+	if g.usageStats.sampleCount > 0 {
+		liveUsersAvg = g.usageStats.numUsersSum / g.usageStats.sampleCount
+		liveClientsAvg = g.usageStats.numClientsSum / g.usageStats.sampleCount
+	}
+
+	var liveEnabled int
+	if g.Cfg.LiveMaxConnections != 0 {
+		liveEnabled = 1
+	}
+
+	var liveHAEnabled int
+	if g.Cfg.LiveHAEngine != "" {
+		liveHAEnabled = 1
+	}
+
+	metrics := map[string]interface{}{
+		"stats.live_enabled.count":      liveEnabled,
+		"stats.live_ha_enabled.count":   liveHAEnabled,
+		"stats.live_samples.count":      g.usageStats.sampleCount,
+		"stats.live_users_max.count":    g.usageStats.numUsersMax,
+		"stats.live_users_avg.count":    liveUsersAvg,
+		"stats.live_clients_max.count":  g.usageStats.numClientsMax,
+		"stats.live_clients_avg.count":  liveClientsAvg,
+		"stats.live_channels_max.count": g.usageStats.numChannelsMax,
+		"stats.live_nodes_max.count":    g.usageStats.numNodesMax,
+	}
+
+	metrics[getHistogramMetric(g.usageStats.numClientsMax, []int{0, 10, 100, 1000, 10000, 100000}, "stats.live_clients_")] = 1
+	metrics[getHistogramMetric(g.usageStats.numUsersMax, []int{0, 10, 100, 1000, 10000, 100000}, "stats.live_users_")] = 1
+	metrics[getHistogramMetric(g.usageStats.numChannelsMax, []int{0, 10, 100, 1000, 10000, 100000}, "stats.live_channels_")] = 1
+	metrics[getHistogramMetric(g.usageStats.numNodesMax, []int{1, 3, 9}, "stats.live_nodes_")] = 1
+
+	return metrics, nil
+}
+
 func (g *GrafanaLive) registerUsageMetrics() {
 	g.usageStatsService.RegisterSendReportCallback(g.resetLiveStats)
-
+	g.usageStatsService.RegisterMetricsFunc(g.collectLiveStats)
-	g.usageStatsService.RegisterMetricsFunc(func(context.Context) (map[string]interface{}, error) {
-		liveUsersAvg := 0
-		liveClientsAvg := 0
-
-		if g.usageStats.sampleCount > 0 {
-			liveUsersAvg = g.usageStats.numUsersSum / g.usageStats.sampleCount
-			liveClientsAvg = g.usageStats.numClientsSum / g.usageStats.sampleCount
-		}
-
-		metrics := map[string]interface{}{
-			"stats.live_samples.count":     g.usageStats.sampleCount,
-			"stats.live_users_max.count":   g.usageStats.numUsersMax,
-			"stats.live_users_min.count":   g.usageStats.numUsersMin,
-			"stats.live_users_avg.count":   liveUsersAvg,
-			"stats.live_clients_max.count": g.usageStats.numClientsMax,
-			"stats.live_clients_min.count": g.usageStats.numClientsMin,
-			"stats.live_clients_avg.count": liveClientsAvg,
-		}
-
-		return metrics, nil
-	})
 }
 
 type usageStats struct {
-	numClientsMax int
+	numClientsMax  int
-	numClientsMin int
+	numClientsSum  int
-	numClientsSum int
+	numUsersMax    int
-	numUsersMax   int
+	numUsersSum    int
-	numUsersMin   int
+	sampleCount    int
-	numUsersSum   int
+	numNodesMax    int
-	sampleCount   int
+	numChannelsMax int
 }

pkg/services/live/live_test.go

@@ -157,3 +157,45 @@ func TestCheckOrigin(t *testing.T) {
 		})
 	}
 }
+
+func Test_getHistogramMetric(t *testing.T) {
+	type args struct {
+		val          int
+		bounds       []int
+		metricPrefix string
+	}
+
+	tests := []struct {
+		name string
+		args args
+		want string
+	}{
+		{
+			"zero",
+			args{0, []int{0, 10, 100, 1000, 10000, 100000}, "live_users_"},
+			"live_users_le_0",
+		},
+		{
+			"equal_to_bound",
+			args{10, []int{0, 10, 100, 1000, 10000, 100000}, "live_users_"},
+			"live_users_le_10",
+		},
+		{
+			"in_the_middle",
+			args{30000, []int{0, 10, 100, 1000, 10000, 100000}, "live_users_"},
+			"live_users_le_100000",
+		},
+		{
+			"more_than_upper_bound",
+			args{300000, []int{0, 10, 100, 1000, 10000, 100000}, "live_users_"},
+			"live_users_le_inf",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := getHistogramMetric(tt.args.val, tt.args.bounds, tt.args.metricPrefix); got != tt.want {
+				t.Errorf("getHistogramMetric() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}