Merge pull request #10282 from filecoin-project/feat/cache-migrated-s…

…tateroot

feat: stmgr: cache migrated stateroots

chain/gen/gen.go

@@ -256,7 +256,7 @@ func NewGeneratorWithSectorsAndUpgradeSchedule(numSectors int, us stmgr.UpgradeS
 	//return nil, xerrors.Errorf("creating drand beacon: %w", err)
 	//}
 
-	sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), sys, us, beac)
+	sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), sys, us, beac, ds)
 	if err != nil {
 		return nil, xerrors.Errorf("initing stmgr: %w", err)
 	}

chain/stmgr/forks.go

@@ -174,9 +174,16 @@ func (us UpgradeSchedule) GetNtwkVersion(e abi.ChainEpoch) (network.Version, err
 
 func (sm *StateManager) HandleStateForks(ctx context.Context, root cid.Cid, height abi.ChainEpoch, cb ExecMonitor, ts *types.TipSet) (cid.Cid, error) {
 	retCid := root
-	var err error
 	u := sm.stateMigrations[height]
 	if u != nil && u.upgrade != nil {
+		migCid, ok, err := u.migrationResultCache.Get(ctx, root)
+		if err == nil && ok {
+			log.Infow("CACHED migration", "height", height, "from", root, "to", migCid)
+			return migCid, nil
+		} else if err != nil {
+			log.Errorw("failed to lookup previous migration result", "err", err)
+		}
+
 		startTime := time.Now()
 		log.Warnw("STARTING migration", "height", height, "from", root)
 		// Yes, we clone the cache, even for the final upgrade epoch. Why? Reverts. We may
@@ -197,6 +204,11 @@ func (sm *StateManager) HandleStateForks(ctx context.Context, root cid.Cid, heig
 			"to", retCid,
 			"duration", time.Since(startTime),
 		)
+
+		// Only set if migration ran, we do not want a root => root mapping
+		if err := u.migrationResultCache.Store(ctx, root, retCid); err != nil {
+			log.Errorw("failed to store migration result", "err", err)
+		}
 	}
 
 	return retCid, nil

chain/stmgr/forks_test.go

@@ -10,6 +10,7 @@ import (
 	"testing"
 
 	"github.com/ipfs/go-cid"
+	"github.com/ipfs/go-datastore"
 	ipldcbor "github.com/ipfs/go-ipld-cbor"
 	logging "github.com/ipfs/go-log/v2"
 	"github.com/stretchr/testify/require"
@@ -35,6 +36,7 @@ import (
 	"github.com/filecoin-project/lotus/chain/consensus"
 	"github.com/filecoin-project/lotus/chain/consensus/filcns"
 	"github.com/filecoin-project/lotus/chain/gen"
+	"github.com/filecoin-project/lotus/chain/stmgr"
 	. "github.com/filecoin-project/lotus/chain/stmgr"
 	"github.com/filecoin-project/lotus/chain/types"
 	"github.com/filecoin-project/lotus/chain/vm"
@@ -166,7 +168,7 @@ func TestForkHeightTriggers(t *testing.T) {
 				}
 
 				return st.Flush(ctx)
-			}}}, cg.BeaconSchedule())
+			}}}, cg.BeaconSchedule(), datastore.NewMapDatastore())
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -284,7 +286,7 @@ func testForkRefuseCall(t *testing.T, nullsBefore, nullsAfter int) {
 				root cid.Cid, height abi.ChainEpoch, ts *types.TipSet) (cid.Cid, error) {
 				migrationCount++
 				return root, nil
-			}}}, cg.BeaconSchedule())
+			}}}, cg.BeaconSchedule(), datastore.NewMapDatastore())
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -502,7 +504,7 @@ func TestForkPreMigration(t *testing.T) {
 					return nil
 				},
 			}}},
-		}, cg.BeaconSchedule())
+		}, cg.BeaconSchedule(), datastore.NewMapDatastore())
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -576,6 +578,7 @@ func TestDisablePreMigration(t *testing.T) {
 			}}},
 		},
 		cg.BeaconSchedule(),
+		datastore.NewMapDatastore(),
 	)
 	require.NoError(t, err)
 	require.NoError(t, sm.Start(context.Background()))
@@ -603,3 +606,102 @@ func TestDisablePreMigration(t *testing.T) {
 
 	require.Equal(t, 1, len(counter))
 }
+
+func TestMigrtionCache(t *testing.T) {
+	logging.SetAllLoggers(logging.LevelInfo)
+
+	cg, err := gen.NewGenerator()
+	require.NoError(t, err)
+
+	counter := make(chan struct{}, 10)
+	metadataDs := datastore.NewMapDatastore()
+
+	sm, err := NewStateManager(
+		cg.ChainStore(),
+		consensus.NewTipSetExecutor(filcns.RewardFunc),
+		cg.StateManager().VMSys(),
+		UpgradeSchedule{{
+			Network: network.Version1,
+			Height:  testForkHeight,
+			Migration: func(_ context.Context, _ *StateManager, _ MigrationCache, _ ExecMonitor,
+				root cid.Cid, _ abi.ChainEpoch, _ *types.TipSet) (cid.Cid, error) {
+
+				counter <- struct{}{}
+
+				return root, nil
+			}},
+		},
+		cg.BeaconSchedule(),
+		metadataDs,
+	)
+	require.NoError(t, err)
+	require.NoError(t, sm.Start(context.Background()))
+	defer func() {
+		require.NoError(t, sm.Stop(context.Background()))
+	}()
+
+	inv := consensus.NewActorRegistry()
+	registry := builtin.MakeRegistryLegacy([]rtt.VMActor{testActor{}})
+	inv.Register(actorstypes.Version0, nil, registry)
+
+	sm.SetVMConstructor(func(ctx context.Context, vmopt *vm.VMOpts) (vm.Interface, error) {
+		nvm, err := vm.NewLegacyVM(ctx, vmopt)
+		require.NoError(t, err)
+		nvm.SetInvoker(inv)
+		return nvm, nil
+	})
+
+	cg.SetStateManager(sm)
+
+	for i := 0; i < 50; i++ {
+		_, err := cg.NextTipSet()
+		require.NoError(t, err)
+	}
+
+	ts, err := cg.ChainStore().GetTipsetByHeight(context.Background(), testForkHeight, nil, false)
+	require.NoError(t, err)
+
+	root, _, err := stmgr.ComputeState(context.Background(), sm, testForkHeight+1, []*types.Message{}, ts)
+	require.NoError(t, err)
+	t.Log(root)
+
+	require.Equal(t, 1, len(counter))
+
+	{
+		sm, err := NewStateManager(
+			cg.ChainStore(),
+			consensus.NewTipSetExecutor(filcns.RewardFunc),
+			cg.StateManager().VMSys(),
+			UpgradeSchedule{{
+				Network: network.Version1,
+				Height:  testForkHeight,
+				Migration: func(_ context.Context, _ *StateManager, _ MigrationCache, _ ExecMonitor,
+					root cid.Cid, _ abi.ChainEpoch, _ *types.TipSet) (cid.Cid, error) {
+
+					counter <- struct{}{}
+
+					return root, nil
+				}},
+			},
+			cg.BeaconSchedule(),
+			metadataDs,
+		)
+		require.NoError(t, err)
+		sm.SetVMConstructor(func(ctx context.Context, vmopt *vm.VMOpts) (vm.Interface, error) {
+			nvm, err := vm.NewLegacyVM(ctx, vmopt)
+			require.NoError(t, err)
+			nvm.SetInvoker(inv)
+			return nvm, nil
+		})
+
+		ctx := context.Background()
+
+		base, _, err := sm.ExecutionTrace(ctx, ts)
+		require.NoError(t, err)
+		_, err = sm.HandleStateForks(context.Background(), base, ts.Height(), nil, ts)
+		require.NoError(t, err)
+
+		// Should not have increased as we should be using the cached results in the metadataDs
+		require.Equal(t, 1, len(counter))
+	}
+}

chain/stmgr/stmgr.go

@@ -2,10 +2,13 @@ package stmgr
 
 import (
 	"context"
+	"fmt"
 	"sync"
 
 	"github.com/ipfs/go-cid"
+	dstore "github.com/ipfs/go-datastore"
 	cbor "github.com/ipfs/go-ipld-cbor"
+	ipld "github.com/ipfs/go-ipld-format"
 	logging "github.com/ipfs/go-log/v2"
 	"golang.org/x/xerrors"
 
@@ -51,9 +54,47 @@ type versionSpec struct {
 }
 
 type migration struct {
-	upgrade       MigrationFunc
-	preMigrations []PreMigration
-	cache         *nv16.MemMigrationCache
+	upgrade              MigrationFunc
+	preMigrations        []PreMigration
+	cache                *nv16.MemMigrationCache
+	migrationResultCache *migrationResultCache
+}
+
+type migrationResultCache struct {
+	ds        dstore.Batching
+	keyPrefix string
+}
+
+func (m *migrationResultCache) keyForMigration(root cid.Cid) dstore.Key {
+	kStr := fmt.Sprintf("%s/%s", m.keyPrefix, root)
+	return dstore.NewKey(kStr)
+}
+
+func (m *migrationResultCache) Get(ctx context.Context, root cid.Cid) (cid.Cid, bool, error) {
+	k := m.keyForMigration(root)
+
+	bs, err := m.ds.Get(ctx, k)
+	if ipld.IsNotFound(err) {
+		return cid.Undef, false, nil
+	} else if err != nil {
+		return cid.Undef, false, xerrors.Errorf("error loading migration result: %w", err)
+	}
+
+	c, err := cid.Parse(bs)
+	if err != nil {
+		return cid.Undef, false, xerrors.Errorf("error parsing migration result: %w", err)
+	}
+
+	return c, true, nil
+}
+
+func (m *migrationResultCache) Store(ctx context.Context, root cid.Cid, resultCid cid.Cid) error {
+	k := m.keyForMigration(root)
+	if err := m.ds.Put(ctx, k, resultCid.Bytes()); err != nil {
+		return err
+	}
+
+	return nil
 }
 
 type Executor interface {
@@ -103,7 +144,7 @@ type treeCache struct {
 	tree *state.StateTree
 }
 
-func NewStateManager(cs *store.ChainStore, exec Executor, sys vm.SyscallBuilder, us UpgradeSchedule, beacon beacon.Schedule) (*StateManager, error) {
+func NewStateManager(cs *store.ChainStore, exec Executor, sys vm.SyscallBuilder, us UpgradeSchedule, beacon beacon.Schedule, metadataDs dstore.Batching) (*StateManager, error) {
 	// If we have upgrades, make sure they're in-order and make sense.
 	if err := us.Validate(); err != nil {
 		return nil, err
@@ -122,12 +163,18 @@ func NewStateManager(cs *store.ChainStore, exec Executor, sys vm.SyscallBuilder,
 					upgrade:       upgrade.Migration,
 					preMigrations: upgrade.PreMigrations,
 					cache:         nv16.NewMemMigrationCache(),
+					migrationResultCache: &migrationResultCache{
+						keyPrefix: fmt.Sprintf("/migration-cache/nv%d", upgrade.Network),
+						ds:        metadataDs,
+					},
 				}
+
 				stateMigrations[upgrade.Height] = migration
 			}
 			if upgrade.Expensive {
 				expensiveUpgrades[upgrade.Height] = struct{}{}
 			}
+
 			networkVersions = append(networkVersions, versionSpec{
 				networkVersion: lastVersion,
 				atOrBelow:      upgrade.Height,
@@ -155,8 +202,8 @@ func NewStateManager(cs *store.ChainStore, exec Executor, sys vm.SyscallBuilder,
 	}, nil
 }
 
-func NewStateManagerWithUpgradeScheduleAndMonitor(cs *store.ChainStore, exec Executor, sys vm.SyscallBuilder, us UpgradeSchedule, b beacon.Schedule, em ExecMonitor) (*StateManager, error) {
-	sm, err := NewStateManager(cs, exec, sys, us, b)
+func NewStateManagerWithUpgradeScheduleAndMonitor(cs *store.ChainStore, exec Executor, sys vm.SyscallBuilder, us UpgradeSchedule, b beacon.Schedule, em ExecMonitor, metadataDs dstore.Batching) (*StateManager, error) {
+	sm, err := NewStateManager(cs, exec, sys, us, b, metadataDs)
 	if err != nil {
 		return nil, err
 	}

chain/store/store_test.go

@@ -196,7 +196,8 @@ func TestChainExportImportFull(t *testing.T) {
 	}
 
 	nbs := blockstore.NewMemorySync()
-	cs := store.NewChainStore(nbs, nbs, datastore.NewMapDatastore(), filcns.Weight, nil)
+	ds := datastore.NewMapDatastore()
+	cs := store.NewChainStore(nbs, nbs, ds, filcns.Weight, nil)
 	defer cs.Close() //nolint:errcheck
 
 	root, err := cs.Import(context.TODO(), buf)
@@ -213,7 +214,7 @@ func TestChainExportImportFull(t *testing.T) {
 		t.Fatal("imported chain differed from exported chain")
 	}
 
-	sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), nil, filcns.DefaultUpgradeSchedule(), cg.BeaconSchedule())
+	sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), nil, filcns.DefaultUpgradeSchedule(), cg.BeaconSchedule(), ds)
 	if err != nil {
 		t.Fatal(err)
 	}

cmd/lotus-bench/import.go

@@ -229,7 +229,7 @@ var importBenchCmd = &cli.Command{
 		defer cs.Close() //nolint:errcheck
 
 		// TODO: We need to supply the actual beacon after v14
-		stm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(verifier), filcns.DefaultUpgradeSchedule(), nil)
+		stm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(verifier), filcns.DefaultUpgradeSchedule(), nil, metadataDs)
 		if err != nil {
 			return err
 		}

cmd/lotus-shed/balances.go

@@ -513,7 +513,7 @@ var chainBalanceStateCmd = &cli.Command{
 		cst := cbor.NewCborStore(bs)
 		store := adt.WrapStore(ctx, cst)
 
-		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), nil)
+		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), nil, mds)
 		if err != nil {
 			return err
 		}
@@ -737,7 +737,7 @@ var chainPledgeCmd = &cli.Command{
 		cst := cbor.NewCborStore(bs)
 		store := adt.WrapStore(ctx, cst)
 
-		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), nil)
+		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), nil, mds)
 		if err != nil {
 			return err
 		}

cmd/lotus-shed/gas-estimation.go

@@ -111,7 +111,7 @@ var gasTraceCmd = &cli.Command{
 		cs := store.NewChainStore(bs, bs, mds, filcns.Weight, nil)
 		defer cs.Close() //nolint:errcheck
 
-		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), shd)
+		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), shd, mds)
 		if err != nil {
 			return err
 		}
@@ -212,7 +212,7 @@ var replayOfflineCmd = &cli.Command{
 		cs := store.NewChainStore(bs, bs, mds, filcns.Weight, nil)
 		defer cs.Close() //nolint:errcheck
 
-		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), shd)
+		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), shd, mds)
 		if err != nil {
 			return err
 		}

cmd/lotus-shed/invariants.go

@@ -90,7 +90,7 @@ var invariantsCmd = &cli.Command{
 		cs := store.NewChainStore(bs, bs, mds, filcns.Weight, nil)
 		defer cs.Close() //nolint:errcheck
 
-		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), nil)
+		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), nil, mds)
 		if err != nil {
 			return err
 		}

cmd/lotus-shed/migrations.go

@@ -8,6 +8,7 @@ import (
 	"time"
 
 	"github.com/ipfs/go-cid"
+	"github.com/ipfs/go-datastore"
 	"github.com/urfave/cli/v2"
 	cbg "github.com/whyrusleeping/cbor-gen"
 	"golang.org/x/xerrors"
@@ -121,7 +122,8 @@ var migrationsCmd = &cli.Command{
 		cs := store.NewChainStore(bs, bs, mds, filcns.Weight, nil)
 		defer cs.Close() //nolint:errcheck
 
-		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), nil)
+		// Note: we use a map datastore for the metadata to avoid writing / using cached migration results in the metadata store
+		sm, err := stmgr.NewStateManager(cs, consensus.NewTipSetExecutor(filcns.RewardFunc), vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), nil, datastore.NewMapDatastore())
 		if err != nil {
 			return err
 		}

cmd/lotus-shed/state-stats.go

@@ -308,7 +308,7 @@ to reduce the number of decode operations performed by caching the decoded objec
 		}
 
 		tsExec := consensus.NewTipSetExecutor(filcns.RewardFunc)
-		sm, err := stmgr.NewStateManager(cs, tsExec, vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), nil)
+		sm, err := stmgr.NewStateManager(cs, tsExec, vm.Syscalls(ffiwrapper.ProofVerifier), filcns.DefaultUpgradeSchedule(), nil, mds)
 		if err != nil {
 			return err
 		}