Add proposervm slot metrics (#3048)

Signed-off-by: Tsachi Herman <24438559+tsachiherman@users.noreply.github.com>

vms/proposervm/block.go

@@ -369,7 +369,10 @@ func (p *postForkCommonComponents) verifyPostDurangoBlockDelay(
 		currentSlot  = proposer.TimeToSlot(parentTimestamp, blkTimestamp)
 		proposerID   = blk.Proposer()
 	)
+	// populate the slot for the block.
+	blk.slot = &currentSlot
 
+	// find the expected proposer
 	expectedProposerID, err := p.vm.Windower.ExpectedProposer(
 		ctx,
 		blkHeight,
@@ -452,6 +455,11 @@ func (p *postForkCommonComponents) shouldBuildSignedBlockPostDurango(
 		)
 		return false, err
 	}
+
+	// report the build slot to the metrics.
+	p.vm.proposerBuildSlotGauge.Set(float64(proposer.TimeToSlot(parentTimestamp, nextStartTime)))
+
+	// set the scheduler to let us know when the next block need to be built.
 	p.vm.Scheduler.SetBuildBlockTime(nextStartTime)
 
 	// In case the inner VM only issued one pendingTxs message, we should

vms/proposervm/block_test.go

@@ -396,8 +396,9 @@ func TestPostDurangoBuildChildResetScheduler(t *testing.T) {
 			ValidatorState: vdrState,
 			Log:            logging.NoLog{},
 		},
-		Windower:  windower,
-		Scheduler: scheduler,
+		Windower:               windower,
+		Scheduler:              scheduler,
+		proposerBuildSlotGauge: prometheus.NewGauge(prometheus.GaugeOpts{}),
 	}
 	vm.Clock.Set(now)
 

vms/proposervm/post_fork_block.go

@@ -17,6 +17,11 @@ var _ PostForkBlock = (*postForkBlock)(nil)
 type postForkBlock struct {
 	block.SignedBlock
 	postForkCommonComponents
+
+	// slot of the proposer that produced this block.
+	// It is populated in verifyPostDurangoBlockDelay.
+	// It is used to report metrics during Accept.
+	slot *uint64
 }
 
 // Accept:
@@ -27,7 +32,13 @@ func (b *postForkBlock) Accept(ctx context.Context) error {
 	if err := b.acceptOuterBlk(); err != nil {
 		return err
 	}
-	return b.acceptInnerBlk(ctx)
+	if err := b.acceptInnerBlk(ctx); err != nil {
+		return err
+	}
+	if b.slot != nil {
+		b.vm.acceptedBlocksSlotHistogram.Observe(float64(*b.slot))
+	}
+	return nil
 }
 
 func (b *postForkBlock) acceptOuterBlk() error {

vms/proposervm/vm.go

@@ -9,6 +9,7 @@ import (
 	"fmt"
 	"time"
 
+	"github.com/prometheus/client_golang/prometheus"
 	"go.uber.org/zap"
 
 	"github.com/ava-labs/avalanchego/cache"
@@ -22,6 +23,7 @@ import (
 	"github.com/ava-labs/avalanchego/snow/consensus/snowman"
 	"github.com/ava-labs/avalanchego/snow/engine/common"
 	"github.com/ava-labs/avalanchego/snow/engine/snowman/block"
+	"github.com/ava-labs/avalanchego/utils"
 	"github.com/ava-labs/avalanchego/utils/constants"
 	"github.com/ava-labs/avalanchego/utils/math"
 	"github.com/ava-labs/avalanchego/utils/timer/mockable"
@@ -97,6 +99,14 @@ type VM struct {
 
 	// lastAcceptedHeight is set to the last accepted PostForkBlock's height.
 	lastAcceptedHeight uint64
+
+	// proposerBuildSlotGauge reports the slot index when this node may attempt
+	// to build a block.
+	proposerBuildSlotGauge prometheus.Gauge
+
+	// acceptedBlocksSlotHistogram reports the slots that accepted blocks were
+	// proposed in.
+	acceptedBlocksSlotHistogram prometheus.Histogram
 }
 
 // New performs best when [minBlkDelay] is whole seconds. This is because block
@@ -206,7 +216,28 @@ func (vm *VM) Initialize(
 	default:
 		return err
 	}
-	return nil
+
+	vm.proposerBuildSlotGauge = prometheus.NewGauge(prometheus.GaugeOpts{
+		Name: "block_building_slot",
+		Help: "the slot that this node may attempt to build a block",
+	})
+	vm.acceptedBlocksSlotHistogram = prometheus.NewHistogram(prometheus.HistogramOpts{
+		Name: "accepted_blocks_slot",
+		Help: "the slot accepted blocks were proposed in",
+		// define the following ranges:
+		// (-inf, 0]
+		// (0, 1]
+		// (1, 2]
+		// (2, inf)
+		// the usage of ".5" before was to ensure we work around the limitation
+		// of comparing floating point of the same numerical value.
+		Buckets: []float64{0.5, 1.5, 2.5},
+	})
+
+	return utils.Err(
+		vm.Config.Registerer.Register(vm.proposerBuildSlotGauge),
+		vm.Config.Registerer.Register(vm.acceptedBlocksSlotHistogram),
+	)
 }
 
 // shutdown ops then propagate shutdown to innerVM
@@ -294,13 +325,15 @@ func (vm *VM) SetPreference(ctx context.Context, preferred ids.ID) error {
 	)
 	if vm.IsDurangoActivated(parentTimestamp) {
 		currentTime := vm.Clock.Time().Truncate(time.Second)
-		nextStartTime, err = vm.getPostDurangoSlotTime(
+		if nextStartTime, err = vm.getPostDurangoSlotTime(
 			ctx,
 			childBlockHeight,
 			pChainHeight,
 			proposer.TimeToSlot(parentTimestamp, currentTime),
 			parentTimestamp,
-		)
+		); err == nil {
+			vm.proposerBuildSlotGauge.Set(float64(proposer.TimeToSlot(parentTimestamp, nextStartTime)))
+		}
 	} else {
 		nextStartTime, err = vm.getPreDurangoSlotTime(
 			ctx,