go/consensus: Enable periodic state checkpoints

.changelog/2880.breaking.md

@@ -0,0 +1,13 @@
+go/consensus: Enable periodic state checkpoints
+
+This adds the following consensus parameters which control how state
+checkpointing is to be performed (currently not enforced):
+
+- `state_checkpoint_interval` is the interval (in blocks) on which state
+  checkpoints should be taken.
+
+- `state_checkpoint_num_kept` is the number of past state checkpoints to
+  keep.
+
+- `state_checkpoint_chunk_size` is the chunk size that should be used when
+  creating state checkpoints.

go/consensus/genesis/genesis.go

@@ -27,6 +27,13 @@ type Parameters struct {
 	MaxEvidenceAgeBlocks uint64          `json:"max_evidence_age_blocks"`
 	MaxEvidenceAgeTime   time.Duration   `json:"max_evidence_age_time"`
 
+	// StateCheckpointInterval is the expected state checkpoint interval (in blocks).
+	StateCheckpointInterval uint64 `json:"state_checkpoint_interval"`
+	// StateCheckpointNumKept is the expected minimum number of state checkpoints to keep.
+	StateCheckpointNumKept uint64 `json:"state_checkpoint_num_kept,omitempty"`
+	// StateCheckpointChunkSize is the chunk size parameter for checkpoint creation.
+	StateCheckpointChunkSize uint64 `json:"state_checkpoint_chunk_size,omitempty"`
+
 	// GasCosts are the base transaction gas costs.
 	GasCosts transaction.Costs `json:"gas_costs,omitempty"`
 
@@ -41,13 +48,29 @@ const (
 
 // SanityCheck does basic sanity checking on the genesis state.
 func (g *Genesis) SanityCheck() error {
-	if g.Parameters.TimeoutCommit < 1*time.Millisecond && !g.Parameters.SkipTimeoutCommit {
+	params := g.Parameters
+
+	if params.TimeoutCommit < 1*time.Millisecond && !params.SkipTimeoutCommit {
 		return fmt.Errorf("consensus: sanity check failed: timeout commit must be >= 1ms")
 	}
 
+	if params.StateCheckpointInterval > 0 {
+		if params.StateCheckpointInterval < 1000 {
+			return fmt.Errorf("consensus: sanity check failed: state checkpoint interval must be >= 1000")
+		}
+
+		if params.StateCheckpointNumKept == 0 {
+			return fmt.Errorf("consensus: sanity check failed: number of kept state checkpoints must be > 0")
+		}
+
+		if params.StateCheckpointChunkSize < 1024*1024 {
+			return fmt.Errorf("consensus: sanity check failed: state checkpoint chunk size must be >= 1 MiB")
+		}
+	}
+
 	// Check for duplicate entries in the pk blacklist.
 	m := make(map[signature.PublicKey]bool)
-	for _, v := range g.Parameters.PublicKeyBlacklist {
+	for _, v := range params.PublicKeyBlacklist {
 		if m[v] {
 			return fmt.Errorf("consensus: sanity check failed: redundant blacklisted public key: '%s'", v)
 		}

go/consensus/tendermint/abci/state.go

@@ -26,6 +26,7 @@ import (
 	storage "github.com/oasisprotocol/oasis-core/go/storage/api"
 	storageDB "github.com/oasisprotocol/oasis-core/go/storage/database"
 	"github.com/oasisprotocol/oasis-core/go/storage/mkvs"
+	"github.com/oasisprotocol/oasis-core/go/storage/mkvs/checkpoint"
 )
 
 var _ api.ApplicationState = (*applicationState)(nil)
@@ -48,6 +49,8 @@ type applicationState struct { // nolint: maligned
 	prunerClosedCh chan struct{}
 	prunerNotifyCh *channels.RingChannel
 
+	checkpointer checkpoint.Checkpointer
+
 	blockLock   sync.RWMutex
 	blockTime   time.Time
 	blockCtx    *api.BlockContext
@@ -269,10 +272,12 @@ func (s *applicationState) doCommit(now time.Time) (uint64, error) {
 	s.checkTxTree.Close()
 	s.checkTxTree = mkvs.NewWithRoot(nil, s.storage.NodeDB(), s.stateRoot, mkvs.WithoutWriteLog())
 
-	// Notify pruner of a new block.
+	// Notify pruner and checkpointer of a new block.
 	s.prunerNotifyCh.In() <- s.stateRoot.Version
 	// Discover the version below which all versions can be discarded from block history.
 	lastRetainedVersion := s.statePruner.GetLastRetainedVersion()
+	// Notify the checkpointer of the new version.
+	s.checkpointer.NotifyNewVersion(s.stateRoot.Version)
 
 	return lastRetainedVersion, nil
 }
@@ -499,6 +504,25 @@ func newApplicationState(ctx context.Context, cfg *ApplicationConfig) (*applicat
 		}
 	}
 
+	// Initialize the checkpointer.
+	checkpointerCfg := checkpoint.CheckpointerConfig{
+		Name:            "consensus",
+		CheckInterval:   1 * time.Minute, // XXX: Make this configurable.
+		RootsPerVersion: 1,
+		GetParameters: func(ctx context.Context) (*checkpoint.CreationParameters, error) {
+			params := s.ConsensusParameters()
+			return &checkpoint.CreationParameters{
+				Interval:  params.StateCheckpointInterval,
+				NumKept:   params.StateCheckpointNumKept,
+				ChunkSize: params.StateCheckpointChunkSize,
+			}, nil
+		},
+	}
+	s.checkpointer, err = checkpoint.NewCheckpointer(s.ctx, ndb, ldb.Checkpointer(), checkpointerCfg)
+	if err != nil {
+		return nil, fmt.Errorf("state: failed to create checkpointer: %w", err)
+	}
+
 	go s.metricsWorker()
 	go s.pruneWorker()
 

go/genesis/genesis_test.go

@@ -128,7 +128,7 @@ func TestGenesisChainContext(t *testing.T) {
 	//       on each run.
 	stableDoc.Staking = staking.Genesis{}
 
-	require.Equal(t, "1024b5ca04a34e17cab59fdae43c32c05e1a51875841b99ea49321a4ec83adb3", stableDoc.ChainContext())
+	require.Equal(t, "935f38f4eba20a2391418c3c708f4a0359725e0c235821923edf5da43142e180", stableDoc.ChainContext())
 }
 
 func TestGenesisSanityCheck(t *testing.T) {

go/oasis-node/cmd/genesis/genesis.go

@@ -79,16 +79,19 @@ const (
 	cfgRoothashDebugBypassStake          = "roothash.debug.bypass_stake" // nolint: gosec
 
 	// Tendermint config flags.
-	cfgConsensusTimeoutCommit        = "consensus.tendermint.timeout_commit"
-	cfgConsensusSkipTimeoutCommit    = "consensus.tendermint.skip_timeout_commit"
-	cfgConsensusEmptyBlockInterval   = "consensus.tendermint.empty_block_interval"
-	cfgConsensusMaxTxSizeBytes       = "consensus.tendermint.max_tx_size"
-	cfgConsensusMaxBlockSizeBytes    = "consensus.tendermint.max_block_size"
-	cfgConsensusMaxBlockGas          = "consensus.tendermint.max_block_gas"
-	cfgConsensusMaxEvidenceAgeBlocks = "consensus.tendermint.max_evidence_age_blocks"
-	cfgConsensusMaxEvidenceAgeTime   = "consensus.tendermint.max_evidence_age_time"
-	CfgConsensusGasCostsTxByte       = "consensus.gas_costs.tx_byte"
-	cfgConsensusBlacklistPublicKey   = "consensus.blacklist_public_key"
+	cfgConsensusTimeoutCommit            = "consensus.tendermint.timeout_commit"
+	cfgConsensusSkipTimeoutCommit        = "consensus.tendermint.skip_timeout_commit"
+	cfgConsensusEmptyBlockInterval       = "consensus.tendermint.empty_block_interval"
+	cfgConsensusMaxTxSizeBytes           = "consensus.tendermint.max_tx_size"
+	cfgConsensusMaxBlockSizeBytes        = "consensus.tendermint.max_block_size"
+	cfgConsensusMaxBlockGas              = "consensus.tendermint.max_block_gas"
+	cfgConsensusMaxEvidenceAgeBlocks     = "consensus.tendermint.max_evidence_age_blocks"
+	cfgConsensusMaxEvidenceAgeTime       = "consensus.tendermint.max_evidence_age_time"
+	cfgConsensusStateCheckpointInterval  = "consensus.state_checkpoint.interval"
+	cfgConsensusStateCheckpointNumKept   = "consensus.state_checkpoint.num_kept"
+	cfgConsensusStateCheckpointChunkSize = "consensus.state_checkpoint.chunk_size"
+	CfgConsensusGasCostsTxByte           = "consensus.gas_costs.tx_byte"
+	cfgConsensusBlacklistPublicKey       = "consensus.blacklist_public_key"
 
 	// Consensus backend config flag.
 	cfgConsensusBackend = "consensus.backend"
@@ -231,14 +234,17 @@ func doInitGenesis(cmd *cobra.Command, args []string) {
 	doc.Consensus = consensusGenesis.Genesis{
 		Backend: viper.GetString(cfgConsensusBackend),
 		Parameters: consensusGenesis.Parameters{
-			TimeoutCommit:        viper.GetDuration(cfgConsensusTimeoutCommit),
-			SkipTimeoutCommit:    viper.GetBool(cfgConsensusSkipTimeoutCommit),
-			EmptyBlockInterval:   viper.GetDuration(cfgConsensusEmptyBlockInterval),
-			MaxTxSize:            uint64(viper.GetSizeInBytes(cfgConsensusMaxTxSizeBytes)),
-			MaxBlockSize:         uint64(viper.GetSizeInBytes(cfgConsensusMaxBlockSizeBytes)),
-			MaxBlockGas:          transaction.Gas(viper.GetUint64(cfgConsensusMaxBlockGas)),
-			MaxEvidenceAgeBlocks: viper.GetUint64(cfgConsensusMaxEvidenceAgeBlocks),
-			MaxEvidenceAgeTime:   viper.GetDuration(cfgConsensusMaxEvidenceAgeTime),
+			TimeoutCommit:            viper.GetDuration(cfgConsensusTimeoutCommit),
+			SkipTimeoutCommit:        viper.GetBool(cfgConsensusSkipTimeoutCommit),
+			EmptyBlockInterval:       viper.GetDuration(cfgConsensusEmptyBlockInterval),
+			MaxTxSize:                uint64(viper.GetSizeInBytes(cfgConsensusMaxTxSizeBytes)),
+			MaxBlockSize:             uint64(viper.GetSizeInBytes(cfgConsensusMaxBlockSizeBytes)),
+			MaxBlockGas:              transaction.Gas(viper.GetUint64(cfgConsensusMaxBlockGas)),
+			MaxEvidenceAgeBlocks:     viper.GetUint64(cfgConsensusMaxEvidenceAgeBlocks),
+			MaxEvidenceAgeTime:       viper.GetDuration(cfgConsensusMaxEvidenceAgeTime),
+			StateCheckpointInterval:  viper.GetUint64(cfgConsensusStateCheckpointInterval),
+			StateCheckpointNumKept:   viper.GetUint64(cfgConsensusStateCheckpointNumKept),
+			StateCheckpointChunkSize: uint64(viper.GetSizeInBytes(cfgConsensusStateCheckpointChunkSize)),
 			GasCosts: transaction.Costs{
 				consensusGenesis.GasOpTxByte: transaction.Gas(viper.GetUint64(CfgConsensusGasCostsTxByte)),
 			},
@@ -742,6 +748,9 @@ func init() {
 	initGenesisFlags.Uint64(cfgConsensusMaxBlockGas, 0, "tendermint max gas used per block")
 	initGenesisFlags.Uint64(cfgConsensusMaxEvidenceAgeBlocks, 100000, "tendermint max evidence age (in blocks)")
 	initGenesisFlags.Duration(cfgConsensusMaxEvidenceAgeTime, 48*time.Hour, "tendermint max evidence age (in time)")
+	initGenesisFlags.Uint64(cfgConsensusStateCheckpointInterval, 10000, "consensus state checkpoint interval (in blocks)")
+	initGenesisFlags.Uint64(cfgConsensusStateCheckpointNumKept, 2, "number of kept consensus state checkpoints")
+	initGenesisFlags.String(cfgConsensusStateCheckpointChunkSize, "8mb", "consensus state checkpoint chunk size (in bytes)")
 	initGenesisFlags.Uint64(CfgConsensusGasCostsTxByte, 1, "consensus gas costs: each transaction byte")
 	initGenesisFlags.StringSlice(cfgConsensusBlacklistPublicKey, nil, "blacklist public key")
 

go/storage/mkvs/checkpoint/checkpointer.go

@@ -17,6 +17,9 @@ import (
 
 // CheckpointerConfig is a checkpointer configuration.
 type CheckpointerConfig struct {
+	// Name identifying this checkpointer in logs.
+	Name string
+
 	// Namespace is the storage namespace this checkpointer is for.
 	Namespace common.Namespace
 
@@ -266,7 +269,7 @@ func NewCheckpointer(
 		ndb:      ndb,
 		creator:  creator,
 		notifyCh: channels.NewRingChannel(1),
-		logger:   logging.GetLogger("storage/mkvs/checkpointer").With("namespace", cfg.Namespace),
+		logger:   logging.GetLogger("storage/mkvs/checkpoint/"+cfg.Name).With("namespace", cfg.Namespace),
 	}
 	go c.worker(ctx)
 	return c, nil

go/worker/storage/committee/node.go

@@ -243,6 +243,7 @@ func NewNode(
 
 	// Create a new checkpointer.
 	checkpointerCfg = checkpoint.CheckpointerConfig{
+		Name:            "runtime",
 		Namespace:       commonNode.Runtime.ID(),
 		CheckInterval:   checkpointerCfg.CheckInterval,
 		RootsPerVersion: 2, // State root and I/O root.