Support proposervm historical block deletion

chains/manager.go

@@ -751,14 +751,19 @@ func (m *manager) createAvalancheChain(
 	}
 
 	// Initialize the ProposerVM and the vm wrapped inside it
-	minBlockDelay := proposervm.DefaultMinBlockDelay
+	var (
+		minBlockDelay       = proposervm.DefaultMinBlockDelay
+		numHistoricalBlocks = proposervm.DefaultNumHistoricalBlocks
+	)
 	if subnetCfg, ok := m.SubnetConfigs[ctx.SubnetID]; ok {
 		minBlockDelay = subnetCfg.ProposerMinBlockDelay
+		numHistoricalBlocks = subnetCfg.ProposerNumHistoricalBlocks
 	}
 	m.Log.Info("creating proposervm wrapper",
 		zap.Time("activationTime", m.ApricotPhase4Time),
 		zap.Uint64("minPChainHeight", m.ApricotPhase4MinPChainHeight),
 		zap.Duration("minBlockDelay", minBlockDelay),
+		zap.Uint64("numHistoricalBlocks", numHistoricalBlocks),
 	)
 
 	chainAlias := m.PrimaryAliasOrDefault(ctx.ChainID)
@@ -778,6 +783,7 @@ func (m *manager) createAvalancheChain(
 		m.ApricotPhase4Time,
 		m.ApricotPhase4MinPChainHeight,
 		minBlockDelay,
+		numHistoricalBlocks,
 		m.stakingSigner,
 		m.stakingCert,
 	)
@@ -1100,14 +1106,19 @@ func (m *manager) createSnowmanChain(
 		return nil, fmt.Errorf("error while fetching chain config: %w", err)
 	}
 
-	minBlockDelay := proposervm.DefaultMinBlockDelay
+	var (
+		minBlockDelay       = proposervm.DefaultMinBlockDelay
+		numHistoricalBlocks = proposervm.DefaultNumHistoricalBlocks
+	)
 	if subnetCfg, ok := m.SubnetConfigs[ctx.SubnetID]; ok {
 		minBlockDelay = subnetCfg.ProposerMinBlockDelay
+		numHistoricalBlocks = subnetCfg.ProposerNumHistoricalBlocks
 	}
 	m.Log.Info("creating proposervm wrapper",
 		zap.Time("activationTime", m.ApricotPhase4Time),
 		zap.Uint64("minPChainHeight", m.ApricotPhase4MinPChainHeight),
 		zap.Duration("minBlockDelay", minBlockDelay),
+		zap.Uint64("numHistoricalBlocks", numHistoricalBlocks),
 	)
 
 	chainAlias := m.PrimaryAliasOrDefault(ctx.ChainID)
@@ -1120,6 +1131,7 @@ func (m *manager) createSnowmanChain(
 		m.ApricotPhase4Time,
 		m.ApricotPhase4MinPChainHeight,
 		minBlockDelay,
+		numHistoricalBlocks,
 		m.stakingSigner,
 		m.stakingCert,
 	)

config/config.go

@@ -1161,10 +1161,11 @@ func getSubnetConfigsFromDir(v *viper.Viper, subnetIDs []ids.ID) (map[ids.ID]sub
 
 func getDefaultSubnetConfig(v *viper.Viper) subnets.Config {
 	return subnets.Config{
-		ConsensusParameters:   getConsensusConfig(v),
-		ValidatorOnly:         false,
-		GossipConfig:          getGossipConfig(v),
-		ProposerMinBlockDelay: proposervm.DefaultMinBlockDelay,
+		ConsensusParameters:         getConsensusConfig(v),
+		ValidatorOnly:               false,
+		GossipConfig:                getGossipConfig(v),
+		ProposerMinBlockDelay:       proposervm.DefaultMinBlockDelay,
+		ProposerNumHistoricalBlocks: proposervm.DefaultNumHistoricalBlocks,
 	}
 }
 

subnets/config.go

@@ -42,8 +42,27 @@ type Config struct {
 
 	// ProposerMinBlockDelay is the minimum delay this node will enforce when
 	// building a snowman++ block.
+	//
 	// TODO: Remove this flag once all VMs throttle their own block production.
 	ProposerMinBlockDelay time.Duration `json:"proposerMinBlockDelay" yaml:"proposerMinBlockDelay"`
+	// ProposerNumHistoricalBlocks is the number of historical snowman++ blocks
+	// this node will index per chain. If set to 0, the node will index all
+	// snowman++ blocks.
+	//
+	// Note: The last accepted block is not considered a historical block. This
+	// prevents the user from only storing the last accepted block, which can
+	// never be safe due to the non-atomic commits between the proposervm
+	// database and the innerVM's database.
+	//
+	// Invariant: This value must be set such that the proposervm never needs to
+	// roll back more blocks than have been deleted. On startup, the proposervm
+	// rolls back its accepted chain to match the innerVM's accepted chain. If
+	// the innerVM is not persisting its last accepted block quickly enough, the
+	// database can become corrupted.
+	//
+	// TODO: Move this flag once the proposervm is configurable on a per-chain
+	// basis.
+	ProposerNumHistoricalBlocks uint64 `json:"proposerNumHistoricalBlocks" yaml:"proposerNumHistoricalBlocks"`
 }
 
 func (c *Config) Valid() error {

vms/proposervm/batched_vm_test.go

@@ -1022,6 +1022,7 @@ func initTestRemoteProposerVM(
 		proBlkStartTime,
 		0,
 		DefaultMinBlockDelay,
+		DefaultNumHistoricalBlocks,
 		pTestSigner,
 		pTestCert,
 	)

vms/proposervm/height_indexed_vm.go

@@ -109,7 +109,8 @@ func (vm *VM) updateHeightIndex(height uint64, blkID ids.ID) error {
 }
 
 func (vm *VM) storeHeightEntry(height uint64, blkID ids.ID) error {
-	switch _, err := vm.State.GetForkHeight(); err {
+	forkHeight, err := vm.State.GetForkHeight()
+	switch err {
 	case nil:
 		// The fork was already reached. Just update the index.
 
@@ -118,14 +119,100 @@ func (vm *VM) storeHeightEntry(height uint64, blkID ids.ID) error {
 		if err := vm.State.SetForkHeight(height); err != nil {
 			return fmt.Errorf("failed storing fork height: %w", err)
 		}
+		forkHeight = height
 
 	default:
 		return fmt.Errorf("failed to load fork height: %w", err)
 	}
 
+	if err := vm.State.SetBlockIDAtHeight(height, blkID); err != nil {
+		return err
+	}
+
 	vm.ctx.Log.Debug("indexed block",
 		zap.Stringer("blkID", blkID),
 		zap.Uint64("height", height),
 	)
-	return vm.State.SetBlockIDAtHeight(height, blkID)
+
+	if vm.numHistoricalBlocks == 0 {
+		return nil
+	}
+
+	blocksSinceFork := height - forkHeight
+	// Note: The last accepted block is not considered a historical block. Which
+	// is why <= is used rather than <. This prevents the user from only storing
+	// the last accepted block, which can never be safe due to the non-atomic
+	// commits between the proposervm database and the innerVM's database.
+	if blocksSinceFork <= vm.numHistoricalBlocks {
+		return nil
+	}
+
+	// Note: heightToDelete is >= forkHeight, so it is guaranteed not to
+	// underflow.
+	heightToDelete := height - vm.numHistoricalBlocks - 1
+	blockToDelete, err := vm.State.GetBlockIDAtHeight(heightToDelete)
+	if err == database.ErrNotFound {
+		// Block may have already been deleted. This can happen due to a
+		// proposervm rollback, the node having recently state-synced, or the
+		// user reconfiguring the node to store more historical blocks than a
+		// prior run.
+		return nil
+	}
+	if err != nil {
+		return err
+	}
+
+	if err := vm.State.DeleteBlockIDAtHeight(heightToDelete); err != nil {
+		return err
+	}
+	if err := vm.State.DeleteBlock(blockToDelete); err != nil {
+		return err
+	}
+
+	vm.ctx.Log.Debug("deleted block",
+		zap.Stringer("blkID", blockToDelete),
+		zap.Uint64("height", heightToDelete),
+	)
+	return nil
+}
+
+// TODO: Support async deletion of old blocks.
+func (vm *VM) pruneOldBlocks() error {
+	if vm.numHistoricalBlocks == 0 {
+		return nil
+	}
+
+	height, err := vm.State.GetMinimumHeight()
+	if err == database.ErrNotFound {
+		// Chain hasn't forked yet
+		return nil
+	}
+
+	// TODO: Refactor to use DB iterators and commit periodically.
+	//
+	// Note: vm.lastAcceptedHeight is guaranteed to be >= height, so the
+	// subtraction can never underflow.
+	for vm.lastAcceptedHeight-height > vm.numHistoricalBlocks {
+		blockToDelete, err := vm.State.GetBlockIDAtHeight(height)
+		if err != nil {
+			return err
+		}
+
+		if err := vm.State.DeleteBlockIDAtHeight(height); err != nil {
+			return err
+		}
+		if err := vm.State.DeleteBlock(blockToDelete); err != nil {
+			return err
+		}
+
+		vm.ctx.Log.Debug("deleted block",
+			zap.Stringer("blkID", blockToDelete),
+			zap.Uint64("height", height),
+		)
+
+		// Note: height is < vm.lastAcceptedHeight, so it is guaranteed not to
+		// overflow.
+		height++
+	}
+	return vm.db.Commit()
 }

vms/proposervm/post_fork_option_test.go

@@ -664,6 +664,7 @@ func TestOptionTimestampValidity(t *testing.T) {
 		time.Time{},
 		0,
 		DefaultMinBlockDelay,
+		DefaultNumHistoricalBlocks,
 		pTestSigner,
 		pTestCert,
 	)

vms/proposervm/state/block_height_index.go

@@ -24,6 +24,9 @@ var (
 )
 
 type HeightIndexGetter interface {
+	// GetMinimumHeight return the smallest height of an indexed blockID. If
+	// there are no indexed blockIDs, ErrNotFound will be returned.
+	GetMinimumHeight() (uint64, error)
 	GetBlockIDAtHeight(height uint64) (ids.ID, error)
 
 	// Fork height is stored when the first post-fork block/option is accepted.
@@ -32,8 +35,9 @@ type HeightIndexGetter interface {
 }
 
 type HeightIndexWriter interface {
-	SetBlockIDAtHeight(height uint64, blkID ids.ID) error
 	SetForkHeight(height uint64) error
+	SetBlockIDAtHeight(height uint64, blkID ids.ID) error
+	DeleteBlockIDAtHeight(height uint64) error
 }
 
 // A checkpoint is the blockID of the next block to be considered
@@ -75,6 +79,21 @@ func NewHeightIndex(db database.Database, commitable versiondb.Commitable) Heigh
 	}
 }
 
+func (hi *heightIndex) GetMinimumHeight() (uint64, error) {
+	it := hi.heightDB.NewIterator()
+	defer it.Release()
+
+	if !it.Next() {
+		return 0, database.ErrNotFound
+	}
+
+	height, err := database.ParseUInt64(it.Key())
+	if err != nil {
+		return 0, err
+	}
+	return height, it.Error()
+}
+
 func (hi *heightIndex) GetBlockIDAtHeight(height uint64) (ids.ID, error) {
 	if blkID, found := hi.heightsCache.Get(height); found {
 		return blkID, nil
@@ -95,6 +114,12 @@ func (hi *heightIndex) SetBlockIDAtHeight(height uint64, blkID ids.ID) error {
 	return database.PutID(hi.heightDB, key, blkID)
 }
 
+func (hi *heightIndex) DeleteBlockIDAtHeight(height uint64) error {
+	hi.heightsCache.Evict(height)
+	key := database.PackUInt64(height)
+	return hi.heightDB.Delete(key)
+}
+
 func (hi *heightIndex) GetForkHeight() (uint64, error) {
 	return database.GetUInt64(hi.metadataDB, forkKey)
 }

vms/proposervm/state/block_state.go

@@ -31,6 +31,7 @@ var (
 type BlockState interface {
 	GetBlock(blkID ids.ID) (block.Block, choices.Status, error)
 	PutBlock(blk block.Block, status choices.Status) error
+	DeleteBlock(blkID ids.ID) error
 }
 
 type blockState struct {
@@ -134,3 +135,8 @@ func (s *blockState) PutBlock(blk block.Block, status choices.Status) error {
 	s.blkCache.Put(blkID, &blkWrapper)
 	return s.db.Put(blkID[:], bytes)
 }
+
+func (s *blockState) DeleteBlock(blkID ids.ID) error {
+	s.blkCache.Evict(blkID)
+	return s.db.Delete(blkID[:])
+}

vms/proposervm/state_syncable_vm_test.go

@@ -76,6 +76,7 @@ func helperBuildStateSyncTestObjects(t *testing.T) (*fullVM, *VM) {
 		time.Time{},
 		0,
 		DefaultMinBlockDelay,
+		DefaultNumHistoricalBlocks,
 		pTestSigner,
 		pTestCert,
 	)