Add metrics for VM flush duration and obj count

Signed-off-by: Jakub Sztandera <kubuxu@protocol.ai>

chain/vm/vm.go

@@ -9,6 +9,7 @@ import (
 	"time"
 
 	"github.com/filecoin-project/lotus/chain/actors/builtin"
+	"github.com/filecoin-project/lotus/metrics"
 
 	block "github.com/ipfs/go-block-format"
 	cid "github.com/ipfs/go-cid"
@@ -16,6 +17,7 @@ import (
 	logging "github.com/ipfs/go-log/v2"
 	mh "github.com/multiformats/go-multihash"
 	cbg "github.com/whyrusleeping/cbor-gen"
+	"go.opencensus.io/stats"
 	"go.opencensus.io/trace"
 	"golang.org/x/xerrors"
 
@@ -618,7 +620,7 @@ func (vm *VM) Flush(ctx context.Context) (cid.Cid, error) {
 		return cid.Undef, xerrors.Errorf("flushing vm: %w", err)
 	}
 
-	if err := Copy(ctx, from, to, root); err != nil {
+	if err := Copy(context.WithValue(ctx, "vm-flush", true), from, to, root); err != nil {
 		return cid.Undef, xerrors.Errorf("copying tree: %w", err)
 	}
 
@@ -675,6 +677,7 @@ func linksForObj(blk block.Block, cb func(cid.Cid)) error {
 func Copy(ctx context.Context, from, to blockstore.Blockstore, root cid.Cid) error {
 	ctx, span := trace.StartSpan(ctx, "vm.Copy") // nolint
 	defer span.End()
+	start := time.Now()
 
 	var numBlocks int
 	var totalCopySize int
@@ -708,6 +711,10 @@ func Copy(ctx context.Context, from, to blockstore.Blockstore, root cid.Cid) err
 		trace.Int64Attribute("numBlocks", int64(numBlocks)),
 		trace.Int64Attribute("copySize", int64(totalCopySize)),
 	)
+	if yes, ok := ctx.Value("vm-flush").(bool); yes && ok {
+		took := metrics.SinceInMilliseconds(start)
+		stats.Record(ctx, metrics.VMFlushCopyCount.M(int64(numBlocks)), metrics.VMFlushCopyDuration.M(took))
+	}
 
 	return nil
 }

metrics/metrics.go

@@ -53,6 +53,8 @@ var (
 	PubsubSendRPC                       = stats.Int64("pubsub/send_rpc", "Counter for total sent RPCs", stats.UnitDimensionless)
 	PubsubDropRPC                       = stats.Int64("pubsub/drop_rpc", "Counter for total dropped RPCs", stats.UnitDimensionless)
 	APIRequestDuration                  = stats.Float64("api/request_duration_ms", "Duration of API requests", stats.UnitMilliseconds)
+	VMFlushCopyDuration                 = stats.Float64("vm/flush_copy_ms", "Time spent in VM Flush Copy", stats.UnitMilliseconds)
+	VMFlushCopyCount                    = stats.Int64("vm/flush_copy_count", "Number of copied objects", stats.UnitDimensionless)
 )
 
 var (
@@ -146,6 +148,14 @@ var (
 		Aggregation: defaultMillisecondsDistribution,
 		TagKeys:     []tag.Key{APIInterface, Endpoint},
 	}
+	VMFlushCopyDurationView = &view.View{
+		Measure:     VMFlushCopyDuration,
+		Aggregation: view.Sum(),
+	}
+	VMFlushCopyCountView = &view.View{
+		Measure:     VMFlushCopyCount,
+		Aggregation: view.Sum(),
+	}
 )
 
 // DefaultViews is an array of OpenCensus views for metric gathering purposes
@@ -171,6 +181,8 @@ var DefaultViews = append([]*view.View{
 	PubsubSendRPCView,
 	PubsubDropRPCView,
 	APIRequestDurationView,
+	VMFlushCopyCountView,
+	VMFlushCopyDurationView,
 },
 	rpcmetrics.DefaultViews...)