Minimal primary sync (#1784)

StephenButtolph · web-flow · commit f181a25ccea2 · 2023-08-02T20:38:52.000-04:00
diff --git a/chains/manager.go b/chains/manager.go
@@ -87,10 +87,11 @@ var (
 	// Bootstrapping prefixes for ChainVMs
 	bootstrappingDB = []byte("bs")
 
-	errUnknownVMType          = errors.New("the vm should have type avalanche.DAGVM or snowman.ChainVM")
-	errCreatePlatformVM       = errors.New("attempted to create a chain running the PlatformVM")
-	errNotBootstrapped        = errors.New("subnets not bootstrapped")
-	errNoPrimaryNetworkConfig = errors.New("no subnet config for primary network found")
+	errUnknownVMType           = errors.New("the vm should have type avalanche.DAGVM or snowman.ChainVM")
+	errCreatePlatformVM        = errors.New("attempted to create a chain running the PlatformVM")
+	errNotBootstrapped         = errors.New("subnets not bootstrapped")
+	errNoPrimaryNetworkConfig  = errors.New("no subnet config for primary network found")
+	errNoLiteSyncForValidators = errors.New("lite sync cannot be configured for a validator")
 
 	_ Manager = (*manager)(nil)
 )
@@ -183,7 +184,7 @@ type ManagerConfig struct {
 	Router                      router.Router              // Routes incoming messages to the appropriate chain
 	Net                         network.Network            // Sends consensus messages to other validators
 	Validators                  validators.Manager         // Validators validating on this chain
-	ReducedMode                 bool
+	LiteSyncPrimaryNetwork      bool
 	NodeID                      ids.NodeID    // The ID of this node
 	NetworkID                   uint32        // ID of the network this node is connected to
 	Server                      server.Server // Handles HTTP API calls
@@ -870,7 +871,6 @@ func (m *manager) createAvalancheChain(
 		Validators:    vdrs,
 		Params:        consensusParams,
 		Consensus:     snowmanConsensus,
-		ReducedMode:   m.ReducedMode && snowmanCommonCfg.Ctx.ChainID == constants.PlatformChainID,
 	}
 	snowmanEngine, err := smeng.New(snowmanEngineConfig)
 	if err != nil {
@@ -1214,6 +1214,7 @@ func (m *manager) createSnowmanChain(
 		Validators:    vdrs,
 		Params:        consensusParams,
 		Consensus:     consensus,
+		LiteSync:      m.LiteSyncPrimaryNetwork && commonCfg.Ctx.ChainID == constants.PlatformChainID,
 	}
 	engine, err := smeng.New(engineConfig)
 	if err != nil {
@@ -1323,6 +1324,33 @@ func (m *manager) registerBootstrappedHealthChecks() error {
 	if err := m.Health.RegisterHealthCheck("bootstrapped", bootstrappedCheck, health.ApplicationTag); err != nil {
 		return fmt.Errorf("couldn't register bootstrapped health check: %w", err)
 	}
+
+	// If the node is not lite syncing the Primary Network, there is no reason
+	// to shutdown if it is a validator.
+	if !m.LiteSyncPrimaryNetwork {
+		return nil
+	}
+
+	liteSyncCheck := health.CheckerFunc(func(ctx context.Context) (interface{}, error) {
+		// Note: The health check is skipped during bootstrapping to allow a
+		// node to sync the network even if it was previously a validator.
+		if !m.IsBootstrapped(constants.PlatformChainID) {
+			return "node is currently bootstrapping", nil
+		}
+		if !validators.Contains(m.Validators, constants.PrimaryNetworkID, m.NodeID) {
+			return "node is not a primary network validator", nil
+		}
+
+		m.Log.Fatal("node is a primary network validator. Shutting down...",
+			zap.Error(errNoLiteSyncForValidators),
+		)
+		go m.ShutdownNodeFunc(1)
+		return "node is a primary network validator", errNoLiteSyncForValidators
+	})
+
+	if err := m.Health.RegisterHealthCheck("validation", liteSyncCheck, health.ApplicationTag); err != nil {
+		return fmt.Errorf("couldn't register validation health check: %w", err)
+	}
 	return nil
 }
 
diff --git a/config/config.go b/config/config.go
@@ -777,7 +777,7 @@ func getStakingConfig(v *viper.Viper, networkID uint32) (node.StakingConfig, err
 	config := node.StakingConfig{
 		SybilProtectionEnabled:        v.GetBool(SybilProtectionEnabledKey),
 		SybilProtectionDisabledWeight: v.GetUint64(SybilProtectionDisabledWeightKey),
-		ReducedMode:                   v.GetBool(ReducedModeKey),
+		LiteSyncPrimaryNetwork:        v.GetBool(LiteSyncPrimaryNetworkKey),
 		StakingKeyPath:                GetExpandedArg(v, StakingTLSKeyPathKey),
 		StakingCertPath:               GetExpandedArg(v, StakingCertPathKey),
 		StakingSignerPath:             GetExpandedArg(v, StakingSignerKeyPathKey),
diff --git a/config/flags.go b/config/flags.go
@@ -259,7 +259,7 @@ func addNodeFlags(fs *pflag.FlagSet) {
 	fs.String(StakingSignerKeyContentKey, "", "Specifies base64 encoded signer private key for staking")
 	fs.Bool(SybilProtectionEnabledKey, true, "Enables sybil protection. If enabled, Network TLS is required")
 	fs.Uint64(SybilProtectionDisabledWeightKey, 100, "Weight to provide to each peer when sybil protection is disabled")
-	fs.Bool(ReducedModeKey, false, "Disable C-chain and X-chain validation. Only allowed for non-validators. Node would shutdown otherwise")
+	fs.Bool(LiteSyncPrimaryNetworkKey, false, "Only sync the P-chain on the Primary Network. If the node is a Primary Network validator, it will shutdown")
 	// Uptime Requirement
 	fs.Float64(UptimeRequirementKey, genesis.LocalParams.UptimeRequirement, "Fraction of time a validator must be online to receive rewards")
 	// Minimum Stake required to validate the Primary Network
diff --git a/config/keys.go b/config/keys.go
@@ -74,7 +74,6 @@ const (
 	StakingSignerKeyContentKey                         = "staking-signer-key-file-content"
 	SybilProtectionEnabledKey                          = "sybil-protection-enabled"
 	SybilProtectionDisabledWeightKey                   = "sybil-protection-disabled-weight"
-	ReducedModeKey                                     = "reduced-mode"
 	NetworkInitialTimeoutKey                           = "network-initial-timeout"
 	NetworkMinimumTimeoutKey                           = "network-minimum-timeout"
 	NetworkMaximumTimeoutKey                           = "network-maximum-timeout"
@@ -130,6 +129,7 @@ const (
 	SnowOptimalProcessingKey                           = "snow-optimal-processing"
 	SnowMaxProcessingKey                               = "snow-max-processing"
 	SnowMaxTimeProcessingKey                           = "snow-max-time-processing"
+	LiteSyncPrimaryNetworkKey                          = "lite-sync-primary-network"
 	TrackSubnetsKey                                    = "track-subnets"
 	AdminAPIEnabledKey                                 = "api-admin-enabled"
 	InfoAPIEnabledKey                                  = "api-info-enabled"
diff --git a/node/config.go b/node/config.go
@@ -92,7 +92,7 @@ type IPConfig struct {
 type StakingConfig struct {
 	genesis.StakingConfig
 	SybilProtectionEnabled        bool            `json:"sybilProtectionEnabled"`
-	ReducedMode                   bool            `json:"reducedModeEnabled"`
+	LiteSyncPrimaryNetwork        bool            `json:"liteSyncPrimaryNetwork"`
 	StakingTLSCert                tls.Certificate `json:"-"`
 	StakingSigningKey             *bls.SecretKey  `json:"-"`
 	SybilProtectionDisabledWeight uint64          `json:"sybilProtectionDisabledWeight"`
diff --git a/node/node.go b/node/node.go
@@ -91,9 +91,8 @@ var (
 	genesisHashKey  = []byte("genesisID")
 	indexerDBPrefix = []byte{0x00}
 
-	errInvalidTLSKey              = errors.New("invalid TLS key")
-	errShuttingDown               = errors.New("server shutting down")
-	errNoReducedModeForValidators = errors.New("reduced mode cannot be configured for validators")
+	errInvalidTLSKey = errors.New("invalid TLS key")
+	errShuttingDown  = errors.New("server shutting down")
 )
 
 // Node is an instance of an Avalanche node.
@@ -818,7 +817,7 @@ func (n *Node) initChainManager(avaxAssetID ids.ID) error {
 		Router:                                  n.Config.ConsensusRouter,
 		Net:                                     n.Net,
 		Validators:                              n.vdrs,
-		ReducedMode:                             n.Config.ReducedMode,
+		LiteSyncPrimaryNetwork:                  n.Config.LiteSyncPrimaryNetwork,
 		NodeID:                                  n.ID,
 		NetworkID:                               n.Config.NetworkID,
 		Server:                                  n.APIServer,
@@ -887,7 +886,7 @@ func (n *Node) initVMs() error {
 				Validators:                    vdrs,
 				UptimeLockedCalculator:        n.uptimeCalculator,
 				SybilProtectionEnabled:        n.Config.SybilProtectionEnabled,
-				ReducedMode:                   n.Config.ReducedMode,
+				LiteSyncPrimaryNetwork:        n.Config.LiteSyncPrimaryNetwork,
 				TrackedSubnets:                n.Config.TrackedSubnets,
 				TxFee:                         n.Config.TxFee,
 				CreateAssetTxFee:              n.Config.CreateAssetTxFee,
@@ -1171,20 +1170,6 @@ func (n *Node) initHealthAPI() error {
 		return fmt.Errorf("couldn't register resource health check: %w", err)
 	}
 
-	reducedModeCheck := health.CheckerFunc(func(ctx context.Context) (interface{}, error) {
-		primaryValidators, _ := n.vdrs.Get(constants.PrimaryNetworkID)
-		if n.Config.ReducedMode && primaryValidators.GetWeight(n.ID) != 0 {
-			n.Log.Fatal(fmt.Sprintf("%s. Shutting down...", errNoReducedModeForValidators.Error()))
-			go n.Shutdown(1)
-			return nil, errNoReducedModeForValidators
-		}
-		return nil, nil
-	})
-	err = n.health.RegisterHealthCheck("reducedmode", reducedModeCheck, health.ApplicationTag)
-	if err != nil {
-		return fmt.Errorf("couldn't register reduced mode health check: %w", err)
-	}
-
 	handler, err := health.NewGetAndPostHandler(n.Log, healthChecker)
 	if err != nil {
 		return err
@@ -1445,7 +1430,6 @@ func (n *Node) Initialize(
 
 	// Start the Health API
 	// Has to be initialized before chain manager
-	// Has to be initialized after Validators Set
 	// [n.Net] must already be set
 	if err := n.initHealthAPI(); err != nil {
 		return fmt.Errorf("couldn't initialize health API: %w", err)
diff --git a/snow/engine/snowman/config.go b/snow/engine/snowman/config.go
@@ -16,11 +16,11 @@ import (
 type Config struct {
 	common.AllGetsServer
 
-	Ctx         *snow.ConsensusContext
-	VM          block.ChainVM
-	Sender      common.Sender
-	Validators  validators.Set
-	Params      snowball.Parameters
-	Consensus   snowman.Consensus
-	ReducedMode bool
+	Ctx        *snow.ConsensusContext
+	VM         block.ChainVM
+	Sender     common.Sender
+	Validators validators.Set
+	Params     snowball.Parameters
+	Consensus  snowman.Consensus
+	LiteSync   bool
 }
diff --git a/snow/engine/snowman/transitive.go b/snow/engine/snowman/transitive.go
@@ -455,7 +455,7 @@ func (t *Transitive) GetBlock(ctx context.Context, blkID ids.ID) (snowman.Block,
 
 func (t *Transitive) sendChits(ctx context.Context, nodeID ids.NodeID, requestID uint32) {
 	lastAccepted := t.Consensus.LastAccepted()
-	if t.Ctx.StateSyncing.Get() || t.Config.ReducedMode {
+	if t.Ctx.StateSyncing.Get() || t.Config.LiteSync {
 		t.Sender.SendChits(ctx, nodeID, requestID, lastAccepted, lastAccepted)
 	} else {
 		t.Sender.SendChits(ctx, nodeID, requestID, t.Consensus.Preference(), lastAccepted)
diff --git a/vms/platformvm/blocks/builder/builder.go b/vms/platformvm/blocks/builder/builder.go
@@ -147,8 +147,12 @@ func (b *builder) AddUnverifiedTx(tx *txs.Tx) error {
 		return err
 	}
 
-	if err := b.Mempool.Add(tx); err != nil {
-		return err
+	// If we are lite syncing the Primary Network, we should not be maintaining
+	// the transaction mempool locally.
+	if !b.txExecutorBackend.Config.LiteSyncPrimaryNetwork {
+		if err := b.Mempool.Add(tx); err != nil {
+			return err
+		}
 	}
 	return b.GossipTx(tx)
 }
diff --git a/vms/platformvm/blocks/executor/manager.go b/vms/platformvm/blocks/executor/manager.go
@@ -53,7 +53,10 @@ func NewManager(
 			validators:   validatorManager,
 			bootstrapped: txExecutorBackend.Bootstrapped,
 		},
-		rejector: &rejector{backend: backend},
+		rejector: &rejector{
+			backend:         backend,
+			addTxsToMempool: !txExecutorBackend.Config.LiteSyncPrimaryNetwork,
+		},
 	}
 }
 
diff --git a/vms/platformvm/blocks/executor/rejector.go b/vms/platformvm/blocks/executor/rejector.go
@@ -16,6 +16,7 @@ var _ blocks.Visitor = (*rejector)(nil)
 // being shutdown.
 type rejector struct {
 	*backend
+	addTxsToMempool bool
 }
 
 func (r *rejector) BanffAbortBlock(b *blocks.BanffAbortBlock) error {
@@ -66,6 +67,10 @@ func (r *rejector) rejectBlock(b blocks.Block, blockType string) error {
 		zap.Stringer("parentID", b.Parent()),
 	)
 
+	if !r.addTxsToMempool {
+		return nil
+	}
+
 	for _, tx := range b.Txs() {
 		if err := r.Mempool.Add(tx); err != nil {
 			r.ctx.Log.Debug(
diff --git a/vms/platformvm/blocks/executor/rejector_test.go b/vms/platformvm/blocks/executor/rejector_test.go
@@ -135,6 +135,7 @@ func TestRejectBlock(t *testing.T) {
 					Mempool:      mempool,
 					state:        state,
 				},
+				addTxsToMempool: true,
 			}
 
 			// Set expected calls on dependencies.
diff --git a/vms/platformvm/config/config.go b/vms/platformvm/config/config.go
@@ -35,8 +35,8 @@ type Config struct {
 	// True if the node is being run with staking enabled
 	SybilProtectionEnabled bool
 
-	// If true, C-chain and X-chain won't be instantiated. Allowed only while node is non-validator
-	ReducedMode bool
+	// If true, only the P-chain will be instantiated on the primary network.
+	LiteSyncPrimaryNetwork bool
 
 	// Set of subnets that this node is validating
 	TrackedSubnets set.Set[ids.ID]
diff --git a/vms/platformvm/txs/executor/standard_tx_executor.go b/vms/platformvm/txs/executor/standard_tx_executor.go
@@ -9,6 +9,8 @@ import (
 	"fmt"
 	"time"
 
+	"go.uber.org/zap"
+
 	"github.com/ava-labs/avalanchego/chains/atomic"
 	"github.com/ava-labs/avalanchego/ids"
 	"github.com/ava-labs/avalanchego/utils/constants"
@@ -136,7 +138,7 @@ func (e *StandardTxExecutor) ImportTx(tx *txs.ImportTx) error {
 		utxoIDs[i] = utxoID[:]
 	}
 
-	if e.Bootstrapped.Get() && !e.Config.ReducedMode {
+	if e.Bootstrapped.Get() && !e.Config.LiteSyncPrimaryNetwork {
 		if err := verify.SameSubnet(context.TODO(), e.Ctx, tx.SourceChain); err != nil {
 			return err
 		}
@@ -187,14 +189,14 @@ func (e *StandardTxExecutor) ImportTx(tx *txs.ImportTx) error {
 	// Produce the UTXOS
 	avax.Produce(e.State, txID, tx.Outs)
 
-	// we fill atomic requests even if reducedMode is enabled
-	// so to allow node to smoothly switch back from reducedMode
+	// Note: We apply atomic requests even if we are not verifying atomic
+	// requests to ensure the shared state will be correct if we later start
+	// verifying the requests.
 	e.AtomicRequests = map[ids.ID]*atomic.Requests{
 		tx.SourceChain: {
 			RemoveRequests: utxoIDs,
 		},
 	}
-
 	return nil
 }
 
@@ -234,8 +236,9 @@ func (e *StandardTxExecutor) ExportTx(tx *txs.ExportTx) error {
 	// Produce the UTXOS
 	avax.Produce(e.State, txID, tx.Outs)
 
-	// we fill atomic requests even if reducedMode is enabled
-	// so to allow node to smoothly switch back from reducedMode
+	// Note: We apply atomic requests even if we are not verifying atomic
+	// requests to ensure the shared state will be correct if we later start
+	// verifying the requests.
 	elems := make([]*atomic.Element, len(tx.ExportedOutputs))
 	for i, out := range tx.ExportedOutputs {
 		utxo := &avax.UTXO{
@@ -284,10 +287,6 @@ func (e *StandardTxExecutor) AddValidatorTx(tx *txs.AddValidatorTx) error {
 		return err
 	}
 
-	if e.Config.ReducedMode && (tx.NodeID() == e.Ctx.NodeID) {
-		e.Ctx.Log.Warn("Verified transaction that would promote node %v to validator. Reduced mode is active and node will shutdown when validation will start.")
-	}
-
 	txID := e.Tx.ID()
 	newStaker, err := state.NewPendingStaker(txID, tx)
 	if err != nil {
@@ -298,6 +297,14 @@ func (e *StandardTxExecutor) AddValidatorTx(tx *txs.AddValidatorTx) error {
 	avax.Consume(e.State, tx.Ins)
 	avax.Produce(e.State, txID, tx.Outs)
 
+	if e.Config.LiteSyncPrimaryNetwork && tx.Validator.NodeID == e.Ctx.NodeID {
+		e.Ctx.Log.Warn("verified transaction that would shutdown this node",
+			zap.String("reason", "primary network is not being fully synced"),
+			zap.Stringer("txID", txID),
+			zap.String("txType", "addValidator"),
+			zap.Stringer("nodeID", tx.Validator.NodeID),
+		)
+	}
 	return nil
 }
 
@@ -444,10 +451,15 @@ func (e *StandardTxExecutor) AddPermissionlessValidatorTx(tx *txs.AddPermissionl
 	avax.Consume(e.State, tx.Ins)
 	avax.Produce(e.State, txID, tx.Outs)
 
-	if e.Config.ReducedMode {
-		if tx.SubnetID() == constants.PrimaryNetworkID && tx.NodeID() == e.Ctx.NodeID {
-			e.Ctx.Log.Warn("Verified transaction that would promote node %v to validator. Reduced mode is active and node will shutdown when validation will start.")
-		}
+	if e.Config.LiteSyncPrimaryNetwork &&
+		tx.Subnet == constants.PrimaryNetworkID &&
+		tx.Validator.NodeID == e.Ctx.NodeID {
+		e.Ctx.Log.Warn("verified transaction that would shutdown this node",
+			zap.String("reason", "primary network is not being fully synced"),
+			zap.Stringer("txID", txID),
+			zap.String("txType", "addPermissionlessValidator"),
+			zap.Stringer("nodeID", tx.Validator.NodeID),
+		)
 	}
 
 	return nil
diff --git a/vms/platformvm/vm.go b/vms/platformvm/vm.go
@@ -147,9 +147,6 @@ func (vm *VM) Initialize(
 	if err != nil {
 		return err
 	}
-	// We don't bother checking whether reducedMode is allowed here
-	// (node must not be a primary network validator). Node-level health checks
-	// will take care of this and err
 
 	validatorManager := pvalidators.NewManager(chainCtx.Log, vm.Config, vm.state, vm.metrics, &vm.clock)
 	vm.State = validatorManager
@@ -219,7 +216,9 @@ func (vm *VM) Initialize(
 
 // Create all chains that exist that this node validates.
 func (vm *VM) initBlockchains() error {
-	if err := vm.createSubnet(constants.PrimaryNetworkID); err != nil {
+	if vm.Config.LiteSyncPrimaryNetwork {
+		vm.ctx.Log.Info("skipping primary network chain creation")
+	} else if err := vm.createSubnet(constants.PrimaryNetworkID); err != nil {
 		return err
 	}
 
@@ -254,10 +253,6 @@ func (vm *VM) createSubnet(subnetID ids.ID) error {
 		if !ok {
 			return fmt.Errorf("expected tx type *txs.CreateChainTx but got %T", chain.Unsigned)
 		}
-		if vm.Config.ReducedMode && tx.SubnetID == constants.PrimaryNetworkID {
-			vm.ctx.Log.Info("reduced mode enabled, skipping C-chain and X-chain validation.")
-			continue
-		}
 		vm.Config.CreateChain(chain.ID(), tx)
 	}
 	return nil

Original file line number	Diff line number	Diff line change
`@@ -147,8 +147,12 @@ func (b builder) AddUnverifiedTx(tx txs.Tx) error {`
`147`	`147`	`return err`
`148`	`148`	`}`
`149`	`149`
`150`		`- if err := b.Mempool.Add(tx); err != nil {`
`151`		`- return err`
	`150`	`+ // If we are lite syncing the Primary Network, we should not be maintaining`
	`151`	`+ // the transaction mempool locally.`
	`152`	`+ if !b.txExecutorBackend.Config.LiteSyncPrimaryNetwork {`
	`153`	`+ if err := b.Mempool.Add(tx); err != nil {`
	`154`	`+ return err`
	`155`	`+ }`
`152`	`156`	`}`
`153`	`157`	`return b.GossipTx(tx)`
`154`	`158`	`}`
Original file line number	Diff line number	Diff line change
`@@ -53,7 +53,10 @@ func NewManager(`
`53`	`53`	`validators: validatorManager,`
`54`	`54`	`bootstrapped: txExecutorBackend.Bootstrapped,`
`55`	`55`	`},`
`56`		`- rejector: &rejector{backend: backend},`
	`56`	`+ rejector: &rejector{`
	`57`	`+ backend: backend,`
	`58`	`+ addTxsToMempool: !txExecutorBackend.Config.LiteSyncPrimaryNetwork,`
	`59`	`+ },`
`57`	`60`	`}`
`58`	`61`	`}`
`59`	`62`