Express block validation, cpu/mem usage via OpenCensus

chain/sync.go

@@ -528,8 +528,17 @@ func blockSanityChecks(h *types.BlockHeader) error {
 
 // Should match up with 'Semantical Validation' in validation.md in the spec
 func (syncer *Syncer) ValidateBlock(ctx context.Context, b *types.FullBlock) error {
+	validationStart := time.Now()
+	defer func() {
+		dur := time.Since(validationStart)
+		durMilli := dur.Seconds() * float64(1000)
+		stats.Record(ctx, metrics.BlockValidationDurationMilliseconds.M(durMilli))
+		log.Infow("block validation", "took", dur, "height", b.Header.Height)
+	}()
+
 	ctx, span := trace.StartSpan(ctx, "validateBlock")
 	defer span.End()
+
 	if build.InsecurePoStValidation {
 		log.Warn("insecure test validation is enabled, if you see this outside of a test, it is a severe bug!")
 	}

cmd/lotus/daemon.go

@@ -18,6 +18,7 @@ import (
 	blockstore "github.com/ipfs/go-ipfs-blockstore"
 	"github.com/mitchellh/go-homedir"
 	"github.com/multiformats/go-multiaddr"
+	"go.opencensus.io/plugin/runmetrics"
 	"go.opencensus.io/stats"
 	"go.opencensus.io/stats/view"
 	"go.opencensus.io/tag"
@@ -114,6 +115,13 @@ var DaemonCmd = &cli.Command{
 		},
 	},
 	Action: func(cctx *cli.Context) error {
+		err := runmetrics.Enable(runmetrics.RunMetricOptions{
+			EnableCPU:    true,
+			EnableMemory: true,
+		})
+		if err != nil {
+			return xerrors.Errorf("enabling runtime metrics: %w", err)
+		}
 		if prof := cctx.String("pprof"); prof != "" {
 			profile, err := os.Create(prof)
 			if err != nil {

metrics/metrics.go

@@ -22,16 +22,17 @@ var (
 
 // Measures
 var (
-	LotusInfo                = stats.Int64("info", "Arbitrary counter to tag lotus info to", stats.UnitDimensionless)
+	LotusInfo                           = stats.Int64("info", "Arbitrary counter to tag lotus info to", stats.UnitDimensionless)
-	ChainNodeHeight          = stats.Int64("chain/node_height", "Current Height of the node", stats.UnitDimensionless)
+	ChainNodeHeight                     = stats.Int64("chain/node_height", "Current Height of the node", stats.UnitDimensionless)
-	ChainNodeWorkerHeight    = stats.Int64("chain/node_worker_height", "Current Height of workers on the node", stats.UnitDimensionless)
+	ChainNodeWorkerHeight               = stats.Int64("chain/node_worker_height", "Current Height of workers on the node", stats.UnitDimensionless)
-	MessageReceived          = stats.Int64("message/received", "Counter for total received messages", stats.UnitDimensionless)
+	MessageReceived                     = stats.Int64("message/received", "Counter for total received messages", stats.UnitDimensionless)
-	MessageValidationFailure = stats.Int64("message/failure", "Counter for message validation failures", stats.UnitDimensionless)
+	MessageValidationFailure            = stats.Int64("message/failure", "Counter for message validation failures", stats.UnitDimensionless)
-	MessageValidationSuccess = stats.Int64("message/success", "Counter for message validation successes", stats.UnitDimensionless)
+	MessageValidationSuccess            = stats.Int64("message/success", "Counter for message validation successes", stats.UnitDimensionless)
-	BlockReceived            = stats.Int64("block/received", "Counter for total received blocks", stats.UnitDimensionless)
+	BlockReceived                       = stats.Int64("block/received", "Counter for total received blocks", stats.UnitDimensionless)
-	BlockValidationFailure   = stats.Int64("block/failure", "Counter for block validation failures", stats.UnitDimensionless)
+	BlockValidationFailure              = stats.Int64("block/failure", "Counter for block validation failures", stats.UnitDimensionless)
-	BlockValidationSuccess   = stats.Int64("block/success", "Counter for block validation successes", stats.UnitDimensionless)
+	BlockValidationSuccess              = stats.Int64("block/success", "Counter for block validation successes", stats.UnitDimensionless)
-	PeerCount                = stats.Int64("peer/count", "Current number of FIL peers", stats.UnitDimensionless)
+	BlockValidationDurationMilliseconds = stats.Float64("block/validation_ms", "Duration for Block Validation in ms", stats.UnitMilliseconds)
+	PeerCount                           = stats.Int64("peer/count", "Current number of FIL peers", stats.UnitDimensionless)
 )
 
 var (
@@ -63,6 +64,10 @@ var (
 		Measure:     BlockValidationSuccess,
 		Aggregation: view.Count(),
 	}
+	BlockValidationDurationView = &view.View{
+		Measure:     BlockValidationDurationMilliseconds,
+		Aggregation: view.Sum(),
+	}
 	MessageReceivedView = &view.View{
 		Measure:     MessageReceived,
 		Aggregation: view.Count(),
@@ -90,6 +95,7 @@ var DefaultViews = append([]*view.View{
 	BlockReceivedView,
 	BlockValidationFailureView,
 	BlockValidationSuccessView,
+	BlockValidationDurationView,
 	MessageReceivedView,
 	MessageValidationFailureView,
 	MessageValidationSuccessView,