sealing: Handlef for PreCommitFailed

cmd/lotus-storage-miner/sectors.go

@@ -151,7 +151,7 @@ var sectorsListCmd = &cli.Command{
 			return list[i] < list[j]
 		})
 
-		w := tabwriter.NewWriter(os.Stdout, 8, 4, 0, ' ', 0)
+		w := tabwriter.NewWriter(os.Stdout, 8, 4, 1, ' ', 0)
 
 		for _, s := range list {
 			st, err := nodeApi.SectorsStatus(ctx, s)

storage/miner.go

@@ -50,7 +50,6 @@ type storageMinerApi interface {
 	StateGetActor(ctx context.Context, actor address.Address, ts *types.TipSet) (*types.Actor, error)
 	StateGetReceipt(context.Context, cid.Cid, *types.TipSet) (*types.MessageReceipt, error)
 	StateMarketStorageDeal(context.Context, uint64, *types.TipSet) (*actors.OnChainDeal, error)
-	StateReadState(ctx context.Context, act *types.Actor, ts *types.TipSet) (*api.ActorState, error)
 
 	MpoolPushMessage(context.Context, *types.Message) (*types.SignedMessage, error)
 
@@ -59,6 +58,7 @@ type storageMinerApi interface {
 	ChainGetRandomness(context.Context, types.TipSetKey, int64) ([]byte, error)
 	ChainGetTipSetByHeight(context.Context, uint64, *types.TipSet) (*types.TipSet, error)
 	ChainGetBlockMessages(context.Context, cid.Cid) (*api.BlockMessages, error)
+	ChainReadObj(context.Context, cid.Cid) ([]byte, error)
 
 	WalletSign(context.Context, address.Address, []byte) (*types.Signature, error)
 	WalletBalance(context.Context, address.Address) (types.BigInt, error)

storage/sealing/fsm.go

@@ -60,6 +60,10 @@ var fsmPlanners = []func(events []statemachine.Event, state *SectorInfo) error{
 	api.SealFailed: planOne(
 		on(SectorRetrySeal{}, api.Unsealed),
 	),
+	api.PreCommitFailed: planOne(
+		on(SectorRetryPreCommit{}, api.PreCommitting),
+		on(SectorRetryWaitSeed{}, api.WaitSeed),
+	),
 
 	api.Faulty: planOne(
 		on(SectorFaultReported{}, api.FaultReported),
@@ -153,7 +157,7 @@ func (m *Sealing) plan(events []statemachine.Event, state *SectorInfo) (func(sta
 	case api.SealFailed:
 		return m.handleSealFailed, nil
 	case api.PreCommitFailed:
-		log.Warnf("sector %d entered unimplemented state 'PreCommitFailed'", state.SectorID)
+		return m.handlePreCommitFailed, nil
 	case api.SealCommitFailed:
 		log.Warnf("sector %d entered unimplemented state 'SealCommitFailed'", state.SectorID)
 	case api.CommitFailed:

storage/sealing/fsm_events.go

@@ -126,6 +126,14 @@ type SectorRetrySeal struct{}
 
 func (evt SectorRetrySeal) apply(state *SectorInfo) {}
 
+type SectorRetryPreCommit struct{}
+
+func (evt SectorRetryPreCommit) apply(state *SectorInfo) {}
+
+type SectorRetryWaitSeed struct{}
+
+func (evt SectorRetryWaitSeed) apply(state *SectorInfo) {}
+
 // Faults
 
 type SectorFaulty struct{}

storage/sealing/sealing.go

@@ -39,7 +39,6 @@ type sealingApi interface { // TODO: trim down
 	StateGetActor(ctx context.Context, actor address.Address, ts *types.TipSet) (*types.Actor, error)
 	StateGetReceipt(context.Context, cid.Cid, *types.TipSet) (*types.MessageReceipt, error)
 	StateMarketStorageDeal(context.Context, uint64, *types.TipSet) (*actors.OnChainDeal, error)
-	StateReadState(ctx context.Context, act *types.Actor, ts *types.TipSet) (*api.ActorState, error)
 
 	MpoolPushMessage(context.Context, *types.Message) (*types.SignedMessage, error)
 
@@ -48,6 +47,7 @@ type sealingApi interface { // TODO: trim down
 	ChainGetRandomness(context.Context, types.TipSetKey, int64) ([]byte, error)
 	ChainGetTipSetByHeight(context.Context, uint64, *types.TipSet) (*types.TipSet, error)
 	ChainGetBlockMessages(context.Context, cid.Cid) (*api.BlockMessages, error)
+	ChainReadObj(context.Context, cid.Cid) ([]byte, error)
 
 	WalletSign(context.Context, address.Address, []byte) (*types.Signature, error)
 	WalletBalance(context.Context, address.Address) (types.BigInt, error)

storage/sealing/states_failed.go

@@ -1,10 +1,14 @@
 package sealing
 
 import (
+	"bytes"
 	"fmt"
 	"time"
 
+	"golang.org/x/xerrors"
+
 	"github.com/filecoin-project/lotus/api"
+	"github.com/filecoin-project/lotus/chain/actors"
 	"github.com/filecoin-project/lotus/lib/statemachine"
 )
 
@@ -24,28 +28,41 @@ func failedCooldown(ctx statemachine.Context, sector SectorInfo) error {
 	return nil
 }
 
-func (m *Sealing) handleSealFailed(ctx statemachine.Context, sector SectorInfo) error {
+func (m *Sealing) checkPreCommitted(ctx statemachine.Context, sector SectorInfo) (*actors.PreCommittedSector, bool) {
-	// TODO: <Remove this section after we can re-precommit>
-
 	act, err := m.api.StateGetActor(ctx.Context(), m.maddr, nil)
 	if err != nil {
 		log.Errorf("handleSealFailed(%d): temp error: %+v", sector.SectorID, err)
-		return nil
+		return nil, true
 	}
 
-	st, err := m.api.StateReadState(ctx.Context(), act, nil)
+	st, err := m.api.ChainReadObj(ctx.Context(), act.Head)
 	if err != nil {
 		log.Errorf("handleSealFailed(%d): temp error: %+v", sector.SectorID, err)
-		return nil
+		return nil, true
 	}
 
-	_, found := st.State.(map[string]interface{})["PreCommittedSectors"].(map[string]interface{})[fmt.Sprint(sector.SectorID)]
+	var state actors.StorageMinerActorState
+	if err := state.UnmarshalCBOR(bytes.NewReader(st)); err != nil {
+		log.Errorf("handleSealFailed(%d): temp error: unmarshaling miner state: %+v", sector.SectorID, err)
+		return nil, true
+	}
+
+	pci, found := state.PreCommittedSectors[fmt.Sprint(sector.SectorID)]
 	if found {
 		// TODO: If not expired yet, we can just try reusing sealticket
-		log.Errorf("sector found in miner preseal array: %+v", sector.SectorID, err)
+		log.Errorf("sector %d found in miner preseal array: %+v", sector.SectorID, err)
-		return nil
+		return pci, true
+	}
+
+	return nil, false
+}
+
+func (m *Sealing) handleSealFailed(ctx statemachine.Context, sector SectorInfo) error {
+
+	if _, is := m.checkPreCommitted(ctx, sector); is {
+		// TODO: Remove this after we can re-precommit
+		return nil // noop, for now
 	}
-	// </>
 
 	if err := failedCooldown(ctx, sector); err != nil {
 		return err
@@ -53,3 +70,50 @@ func (m *Sealing) handleSealFailed(ctx statemachine.Context, sector SectorInfo)
 
 	return ctx.Send(SectorRetrySeal{})
 }
+
+func (m *Sealing) handlePreCommitFailed(ctx statemachine.Context, sector SectorInfo) error {
+	if err := checkSeal(ctx.Context(), m.maddr, sector, m.api); err != nil {
+		switch err.(type) {
+		case *ErrApi:
+			log.Errorf("handlePreCommitFailed: api error, not proceeding: %+v", err)
+			return nil
+		case *ErrBadCommD: // TODO: Should this just back to packing? (not really needed since handleUnsealed will do that too)
+			return ctx.Send(SectorSealFailed{xerrors.Errorf("bad CommD error: %w", err)})
+		case *ErrExpiredTicket:
+			return ctx.Send(SectorSealFailed{xerrors.Errorf("bad CommD error: %w", err)})
+		default:
+			return xerrors.Errorf("checkSeal sanity check error: %w", err)
+		}
+	}
+
+	if pci, is := m.checkPreCommitted(ctx, sector); is && pci != nil {
+		if sector.PreCommitMessage != nil {
+			log.Warn("sector %d is precommitted on chain, but we don't have precommit message", sector.SectorID)
+			return nil // TODO: SeedWait needs this currently
+		}
+
+		if string(pci.Info.CommR) != string(sector.CommR) {
+			log.Warn("sector %d is precommitted on chain, with different CommR: %x != %x", sector.SectorID, pci.Info.CommR, sector.CommR)
+			return nil // TODO: remove when the actor allows re-precommit
+		}
+
+		// TODO: we could compare more things, but I don't think we really need to
+		//  CommR tells us that CommD (and CommPs), and the ticket are all matching
+
+		if err := failedCooldown(ctx, sector); err != nil {
+			return err
+		}
+
+		return ctx.Send(SectorRetryWaitSeed{})
+	}
+
+	if sector.PreCommitMessage != nil {
+		log.Warn("retrying precommit even though the message failed to apply")
+	}
+
+	if err := failedCooldown(ctx, sector); err != nil {
+		return err
+	}
+
+	return ctx.Send(SectorRetryPreCommit{})
+}