Merge pull request #1328 from filecoin-project/metrics/rework-default-view

remove high cardinality tags from metrics

chain/sub/incoming.go

@@ -2,7 +2,6 @@ package sub
 
 import (
 	"context"
-	"fmt"
 	"time"
 
 	lru "github.com/hashicorp/golang-lru"
@@ -112,11 +111,6 @@ func (bv *BlockValidator) flagPeer(p peer.ID) {
 
 func (bv *BlockValidator) Validate(ctx context.Context, pid peer.ID, msg *pubsub.Message) bool {
 	stats.Record(ctx, metrics.BlockReceived.M(1))
-	ctx, _ = tag.New(
-		ctx,
-		tag.Insert(metrics.PeerID, pid.String()),
-		tag.Insert(metrics.ReceivedFrom, msg.ReceivedFrom.String()),
-	)
 	blk, err := types.DecodeBlockMsg(msg.GetData())
 	if err != nil {
 		log.Error("got invalid block over pubsub: ", err)
@@ -178,7 +172,6 @@ func NewMessageValidator(mp *messagepool.MessagePool) *MessageValidator {
 
 func (mv *MessageValidator) Validate(ctx context.Context, pid peer.ID, msg *pubsub.Message) bool {
 	stats.Record(ctx, metrics.MessageReceived.M(1))
-	ctx, _ = tag.New(ctx, tag.Insert(metrics.PeerID, pid.String()))
 	m, err := types.DecodeSignedMessage(msg.Message.GetData())
 	if err != nil {
 		log.Warnf("failed to decode incoming message: %s", err)
@@ -191,9 +184,6 @@ func (mv *MessageValidator) Validate(ctx context.Context, pid peer.ID, msg *pubs
 		log.Warnf("failed to add message from network to message pool (From: %s, To: %s, Nonce: %d, Value: %s): %s", m.Message.From, m.Message.To, m.Message.Nonce, types.FIL(m.Message.Value), err)
 		ctx, _ = tag.New(
 			ctx,
-			tag.Insert(metrics.MessageFrom, m.Message.From.String()),
-			tag.Insert(metrics.MessageTo, m.Message.To.String()),
-			tag.Insert(metrics.MessageNonce, fmt.Sprint(m.Message.Nonce)),
 			tag.Insert(metrics.FailureType, "add"),
 		)
 		stats.Record(ctx, metrics.MessageValidationFailure.M(1))

metrics/metrics.go

@@ -36,67 +36,83 @@ var (
 	RPCResponseError         = stats.Int64("rpc/response_error", "Total number of responses errors handled", stats.UnitDimensionless)
 )
 
-// DefaultViews is an array of Consensus views for metric gathering purposes
-var DefaultViews = []*view.View{
-	&view.View{
+var (
+	InfoView = &view.View{
 		Name:        "info",
 		Description: "Lotus node information",
 		Measure:     LotusInfo,
 		Aggregation: view.LastValue(),
 		TagKeys:     []tag.Key{Version, Commit},
-	},
-	&view.View{
+	}
+	ChainNodeHeightView = &view.View{
 		Measure:     ChainNodeHeight,
 		Aggregation: view.LastValue(),
-	},
-	&view.View{
+	}
+	ChainNodeWorkerHeightView = &view.View{
 		Measure:     ChainNodeWorkerHeight,
 		Aggregation: view.LastValue(),
-	},
-	&view.View{
+	}
+	BlockReceivedView = &view.View{
 		Measure:     BlockReceived,
 		Aggregation: view.Count(),
-	},
-	&view.View{
+	}
+	BlockValidationFailureView = &view.View{
 		Measure:     BlockValidationFailure,
 		Aggregation: view.Count(),
-		TagKeys:     []tag.Key{FailureType, PeerID, ReceivedFrom},
-	},
-	&view.View{
+		TagKeys:     []tag.Key{FailureType},
+	}
+	BlockValidationSuccessView = &view.View{
 		Measure:     BlockValidationSuccess,
 		Aggregation: view.Count(),
-	},
-	&view.View{
+	}
+	MessageReceivedView = &view.View{
 		Measure:     MessageReceived,
 		Aggregation: view.Count(),
-	},
-	&view.View{
+	}
+	MessageValidationFailureView = &view.View{
 		Measure:     MessageValidationFailure,
 		Aggregation: view.Count(),
-		TagKeys:     []tag.Key{FailureType, MessageFrom, MessageTo, MessageNonce},
-	},
-	&view.View{
+		TagKeys:     []tag.Key{FailureType},
+	}
+	MessageValidationSuccessView = &view.View{
 		Measure:     MessageValidationSuccess,
 		Aggregation: view.Count(),
-	},
-	&view.View{
+	}
+	PeerCountView = &view.View{
 		Measure:     PeerCount,
 		Aggregation: view.LastValue(),
-	},
+	}
 	// All RPC related metrics should at the very least tag the RPCMethod
-	&view.View{
+	RPCInvalidMethodView = &view.View{
 		Measure:     RPCInvalidMethod,
 		Aggregation: view.Count(),
 		TagKeys:     []tag.Key{RPCMethod},
-	},
-	&view.View{
+	}
+	RPCRequestErrorView = &view.View{
 		Measure:     RPCRequestError,
 		Aggregation: view.Count(),
 		TagKeys:     []tag.Key{RPCMethod},
-	},
-	&view.View{
+	}
+	RPCResponseErrorView = &view.View{
 		Measure:     RPCResponseError,
 		Aggregation: view.Count(),
 		TagKeys:     []tag.Key{RPCMethod},
-	},
+	}
+)
+
+// DefaultViews is an array of OpenCensus views for metric gathering purposes
+var DefaultViews = []*view.View{
+	InfoView,
+	ChainNodeHeightView,
+	ChainNodeWorkerHeightView,
+	BlockReceivedView,
+	BlockValidationFailureView,
+	BlockValidationSuccessView,
+	MessageReceivedView,
+	MessageValidationFailureView,
+	MessageValidationSuccessView,
+	PeerCountView,
+	RPCInvalidMethodView,
+	RPCRequestErrorView,
+	RPCResponseErrorView,
 }