GetValidators e2e benchmarks

scripts/warp_measures.sh

@@ -0,0 +1,62 @@
+#! /bin/bash
+set -eu
+
+# TODO:
+# Just make a bunch of calls, recording timestamp and height called
+# Calls may be alternated among height and getValidatorsAt
+# All the other data can be retrieved from prometheus and processed by hand.
+
+avalanche_node_addr="127.0.0.1:9650/"
+p_chain_endpoint=$avalanche_node_addr"ext/bc/P"
+
+temp_height_file="/tmp/warp_height_file.json"
+cat > "${temp_height_file}" << EOF
+{
+    "jsonrpc": "2.0",
+    "method": "platform.getHeight",
+    "params": {},
+    "id": 1
+}
+EOF
+
+timestamp=$(date +%s)
+echo $timestamp "START"
+
+for depth in {10000..1..50}; do
+height_call=$(curl $p_chain_endpoint -X POST -H 'content-type:application/json' --silent --data "@${temp_height_file}")
+if [ $? -ne 0 ]; then
+    echo "height call errored. Exiting"
+    exit 1
+fi
+
+current_height=$(echo $height_call | jq -r '."result"."height"')
+timestamp=$(date +%s)
+echo $timestamp "current  height" $current_height
+
+polled_height=$(($current_height - depth))
+
+temp_validatorsAt_file="/tmp/warp_validators_at_file.json"
+cat > "${temp_validatorsAt_file}" << EOF
+{
+    "jsonrpc": "2.0",
+    "method": "platform.getValidatorsAt",
+    "params": {
+        "height":$polled_height
+    },
+    "id": 1
+}
+EOF
+
+validators_call=$(curl $p_chain_endpoint -X POST -H 'content-type:application/json' --silent --data "@${temp_validatorsAt_file}")
+if [ $? -ne 0 ]; then
+    echo "validators at call errored. Exiting"
+    exit 1
+fi
+timestamp=$(date +%s)
+echo $timestamp "validtrs height" $polled_height
+
+sleep 30
+done
+
+timestamp=$(date +%s)
+echo $timestamp "DONE"

vms/platformvm/metrics/metrics.go

@@ -35,6 +35,8 @@ type Metrics interface {
 	// Mark that we computed a validator diff at a height with the given
 	// difference from the top.
 	AddValidatorSetsHeightDiff(uint64)
+
+	AddValidatorSetsOpsCountDiff(uint)
 	// Mark that this much stake is staked on the node.
 	SetLocalStake(uint64)
 	// Mark that this much stake is staked in the network.
@@ -122,6 +124,11 @@ func New(
 			Name:      "validator_sets_height_diff_sum",
 			Help:      "Total number of validator sets diffs applied for generating validator sets",
 		}),
+		validatorSetsOpsCountDiff: prometheus.NewGauge(prometheus.GaugeOpts{
+			Namespace: namespace,
+			Name:      "validator_sets_diffs_ops_count_sum",
+			Help:      "Total number of validator sets operations in diffs for generating validator sets",
+		}),
 		validatorSetsDuration: prometheus.NewGauge(prometheus.GaugeOpts{
 			Namespace: namespace,
 			Name:      "validator_sets_duration_sum",
@@ -148,6 +155,7 @@ func New(
 		registerer.Register(m.validatorSetsCreated),
 		registerer.Register(m.validatorSetsCached),
 		registerer.Register(m.validatorSetsHeightDiff),
+		registerer.Register(m.validatorSetsOpsCountDiff),
 		registerer.Register(m.validatorSetsDuration),
 	)
 
@@ -173,10 +181,11 @@ type metrics struct {
 
 	numVotesWon, numVotesLost prometheus.Counter
 
-	validatorSetsCached     prometheus.Counter
-	validatorSetsCreated    prometheus.Counter
-	validatorSetsHeightDiff prometheus.Gauge
-	validatorSetsDuration   prometheus.Gauge
+	validatorSetsCached       prometheus.Counter
+	validatorSetsCreated      prometheus.Counter
+	validatorSetsHeightDiff   prometheus.Gauge
+	validatorSetsOpsCountDiff prometheus.Gauge
+	validatorSetsDuration     prometheus.Gauge
 }
 
 func (m *metrics) MarkOptionVoteWon() {
@@ -207,6 +216,10 @@ func (m *metrics) AddValidatorSetsHeightDiff(d uint64) {
 	m.validatorSetsHeightDiff.Add(float64(d))
 }
 
+func (m *metrics) AddValidatorSetsOpsCountDiff(d uint) {
+	m.validatorSetsOpsCountDiff.Add(float64(d))
+}
+
 func (m *metrics) SetLocalStake(s uint64) {
 	m.localStake.Set(float64(s))
 }

vms/platformvm/metrics/no_op.go

@@ -39,6 +39,8 @@ func (noopMetrics) AddValidatorSetsDuration(time.Duration) {}
 
 func (noopMetrics) AddValidatorSetsHeightDiff(uint64) {}
 
+func (noopMetrics) AddValidatorSetsOpsCountDiff(uint) {}
+
 func (noopMetrics) SetLocalStake(uint64) {}
 
 func (noopMetrics) SetTotalStake(uint64) {}

vms/platformvm/validators/manager.go

@@ -9,7 +9,6 @@ import (
 	"fmt"
 	"time"
 
-	"github.com/ava-labs/avalanchego/cache"
 	"github.com/ava-labs/avalanchego/database"
 	"github.com/ava-labs/avalanchego/ids"
 	"github.com/ava-labs/avalanchego/snow/validators"
@@ -24,7 +23,6 @@ import (
 )
 
 const (
-	validatorSetsCacheSize        = 64
 	maxRecentlyAcceptedWindowSize = 256
 	recentlyAcceptedWindowTTL     = 5 * time.Minute
 )
@@ -56,7 +54,6 @@ func NewManager(
 		state:   state,
 		metrics: metrics,
 		clk:     clk,
-		caches:  make(map[ids.ID]cache.Cacher[uint64, map[ids.NodeID]*validators.GetValidatorOutput]),
 		recentlyAccepted: window.New[ids.ID](
 			window.Config{
 				Clock:   clk,
@@ -73,10 +70,7 @@ type manager struct {
 	metrics metrics.Metrics
 	clk     *mockable.Clock
 
-	// Maps caches for each subnet that is currently tracked.
-	// Key: Subnet ID
-	// Value: cache mapping height -> validator set map
-	caches map[ids.ID]cache.Cacher[uint64, map[ids.NodeID]*validators.GetValidatorOutput]
+	// dropped caches to simulate worst case scenario when serving validators set
 
 	// sliding window of blocks that were recently accepted
 	recentlyAccepted window.Window[ids.ID]
@@ -132,20 +126,6 @@ func (m *manager) GetCurrentHeight(context.Context) (uint64, error) {
 }
 
 func (m *manager) GetValidatorSet(ctx context.Context, height uint64, subnetID ids.ID) (map[ids.NodeID]*validators.GetValidatorOutput, error) {
-	validatorSetsCache, exists := m.caches[subnetID]
-	if !exists {
-		validatorSetsCache = &cache.LRU[uint64, map[ids.NodeID]*validators.GetValidatorOutput]{Size: validatorSetsCacheSize}
-		// Only cache tracked subnets
-		if subnetID == constants.PrimaryNetworkID || m.cfg.TrackedSubnets.Contains(subnetID) {
-			m.caches[subnetID] = validatorSetsCache
-		}
-	}
-
-	if validatorSet, ok := validatorSetsCache.Get(height); ok {
-		m.metrics.IncValidatorSetsCached()
-		return validatorSet, nil
-	}
-
 	lastAcceptedHeight, err := m.GetCurrentHeight(ctx)
 	if err != nil {
 		return nil, err
@@ -185,33 +165,34 @@ func (m *manager) GetValidatorSet(ctx context.Context, height uint64, subnetID i
 		}
 	}
 
+	totalOpsCount := uint(0)
 	for diffHeight := lastAcceptedHeight; diffHeight > height; diffHeight-- {
-		err := m.applyValidatorDiffs(vdrSet, subnetID, diffHeight)
+		opsCount, err := m.applyValidatorDiffs(vdrSet, subnetID, diffHeight)
 		if err != nil {
 			return nil, err
 		}
+		totalOpsCount += opsCount
 	}
 
-	// cache the validator set
-	validatorSetsCache.Put(height, vdrSet)
-
 	endTime := m.clk.Time()
 	m.metrics.IncValidatorSetsCreated()
 	m.metrics.AddValidatorSetsDuration(endTime.Sub(startTime))
 	m.metrics.AddValidatorSetsHeightDiff(lastAcceptedHeight - height)
+	m.metrics.AddValidatorSetsOpsCountDiff(totalOpsCount)
 	return vdrSet, nil
 }
 
 func (m *manager) applyValidatorDiffs(
 	vdrSet map[ids.NodeID]*validators.GetValidatorOutput,
 	subnetID ids.ID,
 	height uint64,
-) error {
+) (uint, error) {
 	weightDiffs, err := m.state.GetValidatorWeightDiffs(height, subnetID)
 	if err != nil {
-		return err
+		return 0, err
 	}
 
+	opsCount := uint(0)
 	for nodeID, weightDiff := range weightDiffs {
 		vdr, ok := vdrSet[nodeID]
 		if !ok {
@@ -220,6 +201,7 @@ func (m *manager) applyValidatorDiffs(
 				NodeID: nodeID,
 			}
 			vdrSet[nodeID] = vdr
+			opsCount++
 		}
 
 		// The weight of this node changed at this block.
@@ -232,20 +214,22 @@ func (m *manager) applyValidatorDiffs(
 			// prior block it was lower.
 			vdr.Weight, err = math.Sub(vdr.Weight, weightDiff.Amount)
 		}
+		opsCount++
 		if err != nil {
-			return err
+			return 0, err
 		}
 
 		if vdr.Weight == 0 {
 			// The validator's weight was 0 before this block so
 			// they weren't in the validator set.
 			delete(vdrSet, nodeID)
+			opsCount++
 		}
 	}
 
 	pkDiffs, err := m.state.GetValidatorPublicKeyDiffs(height)
 	if err != nil {
-		return err
+		return 0, err
 	}
 
 	for nodeID, pk := range pkDiffs {
@@ -256,9 +240,10 @@ func (m *manager) applyValidatorDiffs(
 			// The validator's public key was removed at this block, so it
 			// was in the validator set before.
 			vdr.PublicKey = pk
+			opsCount++
 		}
 	}
-	return nil
+	return opsCount, nil
 }
 
 func (m *manager) GetSubnetID(_ context.Context, chainID ids.ID) (ids.ID, error) {