Merge pull request #398 from ava-labs/get-ancestors-metrics

Add metrics to track the number of vtx/blocks put into MultiPut msgs
ava-labs · Sep 13, 2020 · 3c73ab9 · 3c73ab9
2 parents 41e3ced + e6b8470
commit 3c73ab9
Show file tree

Hide file tree

Showing 7 changed files with 55 additions and 8 deletions.
diff --git a/network/network.go b/network/network.go
@@ -281,6 +281,10 @@ func (n *network) GetAcceptedFrontier(validatorIDs ids.ShortSet, chainID ids.ID,
 			sent = peer.send(msg)
 		}
 		if !sent {
+			n.log.Debug("failed to send GetAcceptedFrontier(%s, %s, %d)",
+				vID,
+				chainID,
+				requestID)
 			n.executor.Add(func() { n.router.GetAcceptedFrontierFailed(vID, chainID, requestID) })
 			n.getAcceptedFrontier.numFailed.Inc()
 		} else {

diff --git a/snow/engine/avalanche/metrics.go b/snow/engine/avalanche/metrics.go
@@ -11,6 +11,7 @@ import (
 
 type metrics struct {
 	numVtxRequests, numPendingVts, numMissingTxs prometheus.Gauge
+	getAncestorsVtxs                             prometheus.Histogram
 }
 
 // Initialize implements the Engine interface
@@ -30,12 +31,29 @@ func (m *metrics) Initialize(namespace string, registerer prometheus.Registerer)
 		Name:      "missing_txs",
 		Help:      "Number of missing transactions",
 	})
+	m.getAncestorsVtxs = prometheus.NewHistogram(prometheus.HistogramOpts{
+		Namespace: namespace,
+		Name:      "get_ancestors_vtxs",
+		Help:      "The number of vertices fetched in a call to GetAncestors",
+		Buckets: []float64{
+			0,
+			1,
+			5,
+			10,
+			100,
+			500,
+			1000,
+			1500,
+			2000,
+		},
+	})
 
 	errs := wrappers.Errs{}
 	errs.Add(
 		registerer.Register(m.numVtxRequests),
 		registerer.Register(m.numPendingVts),
 		registerer.Register(m.numMissingTxs),
+		registerer.Register(m.getAncestorsVtxs),
 	)
 	return errs.Err
 }
diff --git a/snow/engine/avalanche/transitive.go b/snow/engine/avalanche/transitive.go
@@ -186,6 +186,7 @@ func (t *Transitive) GetAncestors(vdr ids.ShortID, requestID uint32, vtxID ids.I
 		}
 	}
 
+	t.metrics.getAncestorsVtxs.Observe(float64(len(ancestorsBytes)))
 	t.Sender.MultiPut(vdr, requestID, ancestorsBytes)
 	return nil
 }

diff --git a/snow/engine/snowman/metrics.go b/snow/engine/snowman/metrics.go
@@ -11,6 +11,7 @@ import (
 
 type metrics struct {
 	numRequests, numBlocked prometheus.Gauge
+	getAncestorsBlks        prometheus.Histogram
 }
 
 // Initialize the metrics
@@ -25,11 +26,28 @@ func (m *metrics) Initialize(namespace string, registerer prometheus.Registerer)
 		Name:      "blocked",
 		Help:      "Number of blocks that are pending issuance",
 	})
+	m.getAncestorsBlks = prometheus.NewHistogram(prometheus.HistogramOpts{
+		Namespace: namespace,
+		Name:      "get_ancestors_blks",
+		Help:      "The number of blocks fetched in a call to GetAncestors",
+		Buckets: []float64{
+			0,
+			1,
+			5,
+			10,
+			100,
+			500,
+			1000,
+			1500,
+			2000,
+		},
+	})
 
 	errs := wrappers.Errs{}
 	errs.Add(
 		registerer.Register(m.numRequests),
 		registerer.Register(m.numBlocked),
+		registerer.Register(m.getAncestorsBlks),
 	)
 	return errs.Err
 }
diff --git a/snow/engine/snowman/transitive.go b/snow/engine/snowman/transitive.go
@@ -179,6 +179,7 @@ func (t *Transitive) GetAncestors(vdr ids.ShortID, requestID uint32, blkID ids.I
 		}
 	}
 
+	t.metrics.getAncestorsBlks.Observe(float64(len(ancestorsBytes)))
 	t.Sender.MultiPut(vdr, requestID, ancestorsBytes)
 	return nil
 }

diff --git a/snow/networking/router/handler.go b/snow/networking/router/handler.go
@@ -394,26 +394,26 @@ func (h *Handler) GetFailed(validatorID ids.ShortID, requestID uint32) {
 }
 
 // PushQuery passes a PushQuery message received from the network to the consensus engine.
-func (h *Handler) PushQuery(validatorID ids.ShortID, requestID uint32, deadline time.Time, blockID ids.ID, block []byte) bool {
+func (h *Handler) PushQuery(validatorID ids.ShortID, requestID uint32, deadline time.Time, containerID ids.ID, container []byte) bool {
 	return h.serviceQueue.PushMessage(message{
 		messageType: pushQueryMsg,
 		validatorID: validatorID,
 		requestID:   requestID,
 		deadline:    deadline,
-		containerID: blockID,
-		container:   block,
+		containerID: containerID,
+		container:   container,
 		received:    h.clock.Time(),
 	})
 }
 
 // PullQuery passes a PullQuery message received from the network to the consensus engine.
-func (h *Handler) PullQuery(validatorID ids.ShortID, requestID uint32, deadline time.Time, blockID ids.ID) bool {
+func (h *Handler) PullQuery(validatorID ids.ShortID, requestID uint32, deadline time.Time, containerID ids.ID) bool {
 	return h.serviceQueue.PushMessage(message{
 		messageType: pullQueryMsg,
 		validatorID: validatorID,
 		requestID:   requestID,
 		deadline:    deadline,
-		containerID: blockID,
+		containerID: containerID,
 		received:    h.clock.Time(),
 	})
 }

diff --git a/snow/networking/router/message.go b/snow/networking/router/message.go
@@ -62,9 +62,14 @@ func (m message) String() string {
 	sb.WriteString(fmt.Sprintf("\n    messageType: %s", m.messageType))
 	sb.WriteString(fmt.Sprintf("\n    validatorID: %s", m.validatorID))
 	sb.WriteString(fmt.Sprintf("\n    requestID: %d", m.requestID))
-	sb.WriteString(fmt.Sprintf("\n    containerID: %s", m.containerID))
-	sb.WriteString(fmt.Sprintf("\n    containerIDs: %s", m.containerIDs))
-	if m.messageType == notifyMsg {
+	switch m.messageType {
+	case getAcceptedMsg, acceptedMsg, chitsMsg:
+		sb.WriteString(fmt.Sprintf("\n    containerIDs: %s", m.containerIDs))
+	case getMsg, getAncestorsMsg, putMsg, pushQueryMsg, pullQueryMsg:
+		sb.WriteString(fmt.Sprintf("\n    containerID: %s", m.containerID))
+	case multiPutMsg:
+		sb.WriteString(fmt.Sprintf("\n    numContainers: %d", len(m.containers)))
+	case notifyMsg:
 		sb.WriteString(fmt.Sprintf("\n    notification: %s", m.notification))
 	}
 	if !m.deadline.IsZero() {
-Original file line number
+Diff line change
@@ Expand Up @@
     		}
     	}
+    	t.metrics.getAncestorsVtxs.Observe(float64(len(ancestorsBytes)))
     	t.Sender.MultiPut(vdr, requestID, ancestorsBytes)
     	return nil
     }
@@ Expand Down @@