swarm: add a peer dial latency metric (#2959)

libp2p · Oct 17, 2024 · af2042c · af2042c
1 parent 26c9014
commit af2042c
Show file tree

Hide file tree

Showing 4 changed files with 137 additions and 4 deletions.
diff --git a/dashboards/swarm/swarm.json b/dashboards/swarm/swarm.json
@@ -2598,6 +2598,128 @@
       ],
       "title": "Dial Success Rates",
       "type": "timeseries"
+    },    
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "${DS_PROMETHEUS}"
+      },
+      "fieldConfig": {
+        "defaults": {
+          "color": {
+            "mode": "palette-classic"
+          },
+          "custom": {
+            "axisBorderShow": false,
+            "axisCenteredZero": false,
+            "axisColorMode": "text",
+            "axisLabel": "",
+            "axisPlacement": "auto",
+            "barAlignment": 0,
+            "barWidthFactor": 0.6,
+            "drawStyle": "line",
+            "fillOpacity": 0,
+            "gradientMode": "none",
+            "hideFrom": {
+              "legend": false,
+              "tooltip": false,
+              "viz": false
+            },
+            "insertNulls": false,
+            "lineInterpolation": "linear",
+            "lineStyle": {
+              "fill": "solid"
+            },
+            "lineWidth": 1,
+            "pointSize": 5,
+            "scaleDistribution": {
+              "log": 2,
+              "type": "log"
+            },
+            "showPoints": "auto",
+            "spanNulls": false,
+            "stacking": {
+              "group": "A",
+              "mode": "none"
+            },
+            "thresholdsStyle": {
+              "mode": "off"
+            }
+          },
+          "mappings": [],
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              {
+                "color": "green",
+                "value": null
+              }
+            ]
+          },
+          "unit": "s"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 8,
+        "w": 12,
+        "x": 12,
+        "y": 67
+      },
+      "id": 50,
+      "options": {
+        "legend": {
+          "calcs": [],
+          "displayMode": "list",
+          "placement": "bottom",
+          "showLegend": true
+        },
+        "tooltip": {
+          "mode": "single",
+          "sort": "none"
+        }
+      },
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "${DS_PROMETHEUS}"
+          },
+          "editorMode": "code",
+          "exemplar": false,
+          "expr": "histogram_quantile(0.5, sum(rate(libp2p_swarm_dial_latency_seconds_bucket{outcome=\"success\", instance=~\"$instance\"}[$__rate_interval])) by (le))",
+          "instant": false,
+          "legendFormat": "50th percentile",
+          "range": true,
+          "refId": "A"
+        },
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "${DS_PROMETHEUS}"
+          },
+          "editorMode": "code",
+          "expr": "histogram_quantile(0.9, sum(rate(libp2p_swarm_dial_latency_seconds_bucket{outcome=\"success\", instance=~\"$instance\"}[$__rate_interval])) by (le))",
+          "hide": false,
+          "legendFormat": "90th percentile",
+          "range": true,
+          "refId": "B"
+        },
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "${DS_PROMETHEUS}"
+          },
+          "editorMode": "code",
+          "expr": "histogram_quantile(0.95, sum(rate(libp2p_swarm_dial_latency_seconds_bucket{outcome=\"success\", instance=~\"$instance\"}[$__rate_interval])) by (le))",
+          "hide": false,
+          "legendFormat": "95th percentile",
+          "range": true,
+          "refId": "C"
+        }
+      ],
+      "title": "Peer Dial Latency (Seconds)",
+      "type": "timeseries"
     },
     {
       "datasource": {

diff --git a/p2p/net/swarm/dial_worker.go b/p2p/net/swarm/dial_worker.go
@@ -162,7 +162,7 @@ loop:
 		case req, ok := <-w.reqch:
 			if !ok {
 				if w.s.metricsTracer != nil {
-					w.s.metricsTracer.DialCompleted(w.connected, totalDials)
+					w.s.metricsTracer.DialCompleted(w.connected, totalDials, time.Since(startTime))
 				}
 				return
 			}

diff --git a/p2p/net/swarm/swarm_metrics.go b/p2p/net/swarm/swarm_metrics.go
@@ -77,6 +77,15 @@ var (
 		},
 		[]string{"outcome", "num_dials"},
 	)
+	dialLatency = prometheus.NewHistogramVec(
+		prometheus.HistogramOpts{
+			Namespace: metricNamespace,
+			Name:      "dial_latency_seconds",
+			Help:      "time taken to establish connection with the peer",
+			Buckets:   []float64{0.001, 1.3, 35},
+		},
+		[]string{"outcome", "num_dials"},
+	)
 	dialRankingDelay = prometheus.NewHistogram(
 		prometheus.HistogramOpts{
 			Namespace: metricNamespace,
@@ -118,6 +127,7 @@ var (
 		connHandshakeLatency,
 		dialsPerPeer,
 		dialRankingDelay,
+		dialLatency,
 		blackHoleSuccessCounterSuccessFraction,
 		blackHoleSuccessCounterState,
 		blackHoleSuccessCounterNextRequestAllowedAfter,
@@ -129,7 +139,7 @@ type MetricsTracer interface {
 	ClosedConnection(network.Direction, time.Duration, network.ConnectionState, ma.Multiaddr)
 	CompletedHandshake(time.Duration, network.ConnectionState, ma.Multiaddr)
 	FailedDialing(ma.Multiaddr, error, error)
-	DialCompleted(success bool, totalDials int)
+	DialCompleted(success bool, totalDials int, latency time.Duration)
 	DialRankingDelay(d time.Duration)
 	UpdatedBlackHoleSuccessCounter(name string, state BlackHoleState, nextProbeAfter int, successFraction float64)
 }
@@ -250,7 +260,7 @@ func (m *metricsTracer) FailedDialing(addr ma.Multiaddr, dialErr error, cause er
 	dialError.WithLabelValues(*tags...).Inc()
 }
 
-func (m *metricsTracer) DialCompleted(success bool, totalDials int) {
+func (m *metricsTracer) DialCompleted(success bool, totalDials int, latency time.Duration) {
 	tags := metricshelper.GetStringSlice()
 	defer metricshelper.PutStringSlice(tags)
 	if success {
@@ -268,6 +278,7 @@ func (m *metricsTracer) DialCompleted(success bool, totalDials int) {
 	}
 	*tags = append(*tags, numDials)
 	dialsPerPeer.WithLabelValues(*tags...).Inc()
+	dialLatency.WithLabelValues(*tags...).Observe(latency.Seconds())
 }
 
 func (m *metricsTracer) DialRankingDelay(d time.Duration) {

diff --git a/p2p/net/swarm/swarm_metrics_test.go b/p2p/net/swarm/swarm_metrics_test.go
@@ -92,7 +92,7 @@ func TestMetricsNoAllocNoCover(t *testing.T) {
 			mt.CompletedHandshake(time.Duration(mrand.Intn(100))*time.Second, randItem(connections), randItem(addrs))
 		},
 		"FailedDialing":    func() { mt.FailedDialing(randItem(addrs), randItem(errors), randItem(errors)) },
-		"DialCompleted":    func() { mt.DialCompleted(mrand.Intn(2) == 1, mrand.Intn(10)) },
+		"DialCompleted":    func() { mt.DialCompleted(mrand.Intn(2) == 1, mrand.Intn(10), time.Duration(mrand.Intn(1000_000_000))) },
 		"DialRankingDelay": func() { mt.DialRankingDelay(time.Duration(mrand.Intn(1e10))) },
 		"UpdatedBlackHoleSuccessCounter": func() {
 			mt.UpdatedBlackHoleSuccessCounter(