package storage import ( "bytes" "context" "errors" "time" "github.com/filecoin-project/go-bitfield" "github.com/filecoin-project/go-address" "github.com/filecoin-project/specs-actors/actors/abi" "github.com/filecoin-project/specs-actors/actors/abi/big" "github.com/filecoin-project/specs-actors/actors/builtin" "github.com/filecoin-project/specs-actors/actors/builtin/miner" "github.com/filecoin-project/specs-actors/actors/crypto" "github.com/ipfs/go-cid" "go.opencensus.io/trace" "golang.org/x/xerrors" "github.com/filecoin-project/lotus/api" "github.com/filecoin-project/lotus/build" "github.com/filecoin-project/lotus/chain/actors" "github.com/filecoin-project/lotus/chain/types" "github.com/filecoin-project/lotus/journal" ) var errNoPartitions = errors.New("no partitions") func (s *WindowPoStScheduler) failPost(err error, deadline *miner.DeadlineInfo) { journal.J.RecordEvent(s.evtTypes[evtTypeWdPoStScheduler], func() interface{} { return WdPoStSchedulerEvt{ evtCommon: s.getEvtCommon(err), State: SchedulerStateFaulted, } }) log.Errorf("TODO") /*s.failLk.Lock() if eps > s.failed { s.failed = eps } s.failLk.Unlock()*/ } func (s *WindowPoStScheduler) doPost(ctx context.Context, deadline *miner.DeadlineInfo, ts *types.TipSet) { ctx, abort := context.WithCancel(ctx) s.abort = abort s.activeDeadline = deadline journal.J.RecordEvent(s.evtTypes[evtTypeWdPoStScheduler], func() interface{} { return WdPoStSchedulerEvt{ evtCommon: s.getEvtCommon(nil), State: SchedulerStateStarted, } }) go func() { defer abort() ctx, span := trace.StartSpan(ctx, "WindowPoStScheduler.doPost") defer span.End() // recordProofsEvent records a successful proofs_processed event in the // journal, even if it was a noop (no partitions). recordProofsEvent := func(partitions []miner.PoStPartition, mcid cid.Cid) { journal.J.RecordEvent(s.evtTypes[evtTypeWdPoStProofs], func() interface{} { return &WdPoStProofsProcessedEvt{ evtCommon: s.getEvtCommon(nil), Partitions: partitions, MessageCID: mcid, } }) } proof, err := s.runPost(ctx, *deadline, ts) switch err { case errNoPartitions: recordProofsEvent(nil, cid.Undef) return case nil: sm, err := s.submitPost(ctx, proof) if err != nil { log.Errorf("submitPost failed: %+v", err) s.failPost(err, deadline) return } recordProofsEvent(proof.Partitions, sm.Cid()) default: log.Errorf("runPost failed: %+v", err) s.failPost(err, deadline) return } journal.J.RecordEvent(s.evtTypes[evtTypeWdPoStScheduler], func() interface{} { return WdPoStSchedulerEvt{ evtCommon: s.getEvtCommon(nil), State: SchedulerStateSucceeded, } }) }() } func (s *WindowPoStScheduler) checkSectors(ctx context.Context, check abi.BitField) (abi.BitField, error) { spt, err := s.proofType.RegisteredSealProof() if err != nil { return bitfield.BitField{}, xerrors.Errorf("getting seal proof type: %w", err) } mid, err := address.IDFromAddress(s.actor) if err != nil { return bitfield.BitField{}, err } sectors := make(map[abi.SectorID]struct{}) var tocheck []abi.SectorID err = check.ForEach(func(snum uint64) error { s := abi.SectorID{ Miner: abi.ActorID(mid), Number: abi.SectorNumber(snum), } tocheck = append(tocheck, s) sectors[s] = struct{}{} return nil }) if err != nil { return bitfield.BitField{}, xerrors.Errorf("iterating over bitfield: %w", err) } bad, err := s.faultTracker.CheckProvable(ctx, spt, tocheck) if err != nil { return bitfield.BitField{}, xerrors.Errorf("checking provable sectors: %w", err) } for _, id := range bad { delete(sectors, id) } log.Warnw("Checked sectors", "checked", len(tocheck), "good", len(sectors)) sbf := bitfield.New() for s := range sectors { sbf.Set(uint64(s.Number)) } return sbf, nil } func (s *WindowPoStScheduler) checkNextRecoveries(ctx context.Context, dlIdx uint64, partitions []*miner.Partition) ([]miner.RecoveryDeclaration, *types.SignedMessage, error) { ctx, span := trace.StartSpan(ctx, "storage.checkNextRecoveries") defer span.End() faulty := uint64(0) params := &miner.DeclareFaultsRecoveredParams{ Recoveries: []miner.RecoveryDeclaration{}, } for partIdx, partition := range partitions { unrecovered, err := bitfield.SubtractBitField(partition.Faults, partition.Recoveries) if err != nil { return nil, nil, xerrors.Errorf("subtracting recovered set from fault set: %w", err) } uc, err := unrecovered.Count() if err != nil { return nil, nil, xerrors.Errorf("counting unrecovered sectors: %w", err) } if uc == 0 { continue } faulty += uc recovered, err := s.checkSectors(ctx, unrecovered) if err != nil { return nil, nil, xerrors.Errorf("checking unrecovered sectors: %w", err) } // if all sectors failed to recover, don't declare recoveries recoveredCount, err := recovered.Count() if err != nil { return nil, nil, xerrors.Errorf("counting recovered sectors: %w", err) } if recoveredCount == 0 { continue } params.Recoveries = append(params.Recoveries, miner.RecoveryDeclaration{ Deadline: dlIdx, Partition: uint64(partIdx), Sectors: recovered, }) } recoveries := params.Recoveries if len(recoveries) == 0 { if faulty != 0 { log.Warnw("No recoveries to declare", "deadline", dlIdx, "faulty", faulty) } return recoveries, nil, nil } enc, aerr := actors.SerializeParams(params) if aerr != nil { return recoveries, nil, xerrors.Errorf("could not serialize declare recoveries parameters: %w", aerr) } msg := &types.Message{ To: s.actor, From: s.worker, Method: builtin.MethodsMiner.DeclareFaultsRecovered, Params: enc, Value: types.NewInt(0), } spec := &api.MessageSendSpec{MaxFee: abi.TokenAmount(s.feeCfg.MaxWindowPoStGasFee)} s.setSender(ctx, msg, spec) sm, err := s.api.MpoolPushMessage(ctx, msg, &api.MessageSendSpec{MaxFee: abi.TokenAmount(s.feeCfg.MaxWindowPoStGasFee)}) if err != nil { return recoveries, sm, xerrors.Errorf("pushing message to mpool: %w", err) } log.Warnw("declare faults recovered Message CID", "cid", sm.Cid()) rec, err := s.api.StateWaitMsg(context.TODO(), sm.Cid(), build.MessageConfidence) if err != nil { return recoveries, sm, xerrors.Errorf("declare faults recovered wait error: %w", err) } if rec.Receipt.ExitCode != 0 { return recoveries, sm, xerrors.Errorf("declare faults recovered wait non-0 exit code: %d", rec.Receipt.ExitCode) } return recoveries, sm, nil } func (s *WindowPoStScheduler) checkNextFaults(ctx context.Context, dlIdx uint64, partitions []*miner.Partition) ([]miner.FaultDeclaration, *types.SignedMessage, error) { ctx, span := trace.StartSpan(ctx, "storage.checkNextFaults") defer span.End() bad := uint64(0) params := &miner.DeclareFaultsParams{ Faults: []miner.FaultDeclaration{}, } for partIdx, partition := range partitions { toCheck, err := partition.ActiveSectors() if err != nil { return nil, nil, xerrors.Errorf("getting active sectors: %w", err) } good, err := s.checkSectors(ctx, toCheck) if err != nil { return nil, nil, xerrors.Errorf("checking sectors: %w", err) } faulty, err := bitfield.SubtractBitField(toCheck, good) if err != nil { return nil, nil, xerrors.Errorf("calculating faulty sector set: %w", err) } c, err := faulty.Count() if err != nil { return nil, nil, xerrors.Errorf("counting faulty sectors: %w", err) } if c == 0 { continue } bad += c params.Faults = append(params.Faults, miner.FaultDeclaration{ Deadline: dlIdx, Partition: uint64(partIdx), Sectors: faulty, }) } faults := params.Faults if len(faults) == 0 { return faults, nil, nil } log.Errorw("DETECTED FAULTY SECTORS, declaring faults", "count", bad) enc, aerr := actors.SerializeParams(params) if aerr != nil { return faults, nil, xerrors.Errorf("could not serialize declare faults parameters: %w", aerr) } msg := &types.Message{ To: s.actor, From: s.worker, Method: builtin.MethodsMiner.DeclareFaults, Params: enc, Value: types.NewInt(0), // TODO: Is there a fee? } spec := &api.MessageSendSpec{MaxFee: abi.TokenAmount(s.feeCfg.MaxWindowPoStGasFee)} s.setSender(ctx, msg, spec) sm, err := s.api.MpoolPushMessage(ctx, msg, spec) if err != nil { return faults, sm, xerrors.Errorf("pushing message to mpool: %w", err) } log.Warnw("declare faults Message CID", "cid", sm.Cid()) rec, err := s.api.StateWaitMsg(context.TODO(), sm.Cid(), build.MessageConfidence) if err != nil { return faults, sm, xerrors.Errorf("declare faults wait error: %w", err) } if rec.Receipt.ExitCode != 0 { return faults, sm, xerrors.Errorf("declare faults wait non-0 exit code: %d", rec.Receipt.ExitCode) } return faults, sm, nil } func (s *WindowPoStScheduler) runPost(ctx context.Context, di miner.DeadlineInfo, ts *types.TipSet) (*miner.SubmitWindowedPoStParams, error) { ctx, span := trace.StartSpan(ctx, "storage.runPost") defer span.End() go func() { // TODO: extract from runPost, run on fault cutoff boundaries // check faults / recoveries for the *next* deadline. It's already too // late to declare them for this deadline declDeadline := (di.Index + 2) % miner.WPoStPeriodDeadlines partitions, err := s.api.StateMinerPartitions(context.TODO(), s.actor, declDeadline, ts.Key()) if err != nil { log.Errorf("getting partitions: %v", err) return } var ( sigmsg *types.SignedMessage recoveries []miner.RecoveryDeclaration faults []miner.FaultDeclaration // optionalCid returns the CID of the message, or cid.Undef is the // message is nil. We don't need the argument (could capture the // pointer), but it's clearer and purer like that. optionalCid = func(sigmsg *types.SignedMessage) cid.Cid { if sigmsg == nil { return cid.Undef } return sigmsg.Cid() } ) if recoveries, sigmsg, err = s.checkNextRecoveries(context.TODO(), declDeadline, partitions); err != nil { // TODO: This is potentially quite bad, but not even trying to post when this fails is objectively worse log.Errorf("checking sector recoveries: %v", err) } journal.J.RecordEvent(s.evtTypes[evtTypeWdPoStRecoveries], func() interface{} { j := WdPoStRecoveriesProcessedEvt{ evtCommon: s.getEvtCommon(err), Declarations: recoveries, MessageCID: optionalCid(sigmsg), } j.Error = err return j }) if faults, sigmsg, err = s.checkNextFaults(context.TODO(), declDeadline, partitions); err != nil { // TODO: This is also potentially really bad, but we try to post anyways log.Errorf("checking sector faults: %v", err) } journal.J.RecordEvent(s.evtTypes[evtTypeWdPoStFaults], func() interface{} { return WdPoStFaultsProcessedEvt{ evtCommon: s.getEvtCommon(err), Declarations: faults, MessageCID: optionalCid(sigmsg), } }) }() buf := new(bytes.Buffer) if err := s.actor.MarshalCBOR(buf); err != nil { return nil, xerrors.Errorf("failed to marshal address to cbor: %w", err) } rand, err := s.api.ChainGetRandomnessFromBeacon(ctx, ts.Key(), crypto.DomainSeparationTag_WindowedPoStChallengeSeed, di.Challenge, buf.Bytes()) if err != nil { return nil, xerrors.Errorf("failed to get chain randomness for windowPost (ts=%d; deadline=%d): %w", ts.Height(), di, err) } partitions, err := s.api.StateMinerPartitions(ctx, s.actor, di.Index, ts.Key()) if err != nil { return nil, xerrors.Errorf("getting partitions: %w", err) } params := &miner.SubmitWindowedPoStParams{ Deadline: di.Index, Partitions: make([]miner.PoStPartition, 0, len(partitions)), Proofs: nil, } var sinfos []abi.SectorInfo sidToPart := map[abi.SectorNumber]uint64{} skipCount := uint64(0) for partIdx, partition := range partitions { // TODO: Can do this in parallel toProve, err := partition.ActiveSectors() if err != nil { return nil, xerrors.Errorf("getting active sectors: %w", err) } toProve, err = bitfield.MergeBitFields(toProve, partition.Recoveries) if err != nil { return nil, xerrors.Errorf("adding recoveries to set of sectors to prove: %w", err) } good, err := s.checkSectors(ctx, toProve) if err != nil { return nil, xerrors.Errorf("checking sectors to skip: %w", err) } skipped, err := bitfield.SubtractBitField(toProve, good) if err != nil { return nil, xerrors.Errorf("toProve - good: %w", err) } sc, err := skipped.Count() if err != nil { return nil, xerrors.Errorf("getting skipped sector count: %w", err) } skipCount += sc ssi, err := s.sectorInfo(ctx, good, ts) if err != nil { return nil, xerrors.Errorf("getting sorted sector info: %w", err) } if len(ssi) == 0 { continue } sinfos = append(sinfos, ssi...) for _, si := range ssi { sidToPart[si.SectorNumber] = uint64(partIdx) } params.Partitions = append(params.Partitions, miner.PoStPartition{ Index: uint64(partIdx), Skipped: skipped, }) } if len(sinfos) == 0 { // nothing to prove.. return nil, errNoPartitions } log.Infow("running windowPost", "chain-random", rand, "deadline", di, "height", ts.Height(), "skipped", skipCount) tsStart := build.Clock.Now() log.Infow("generating windowPost", "sectors", len(sinfos)) mid, err := address.IDFromAddress(s.actor) if err != nil { return nil, err } postOut, postSkipped, err := s.prover.GenerateWindowPoSt(ctx, abi.ActorID(mid), sinfos, abi.PoStRandomness(rand)) if err != nil { return nil, xerrors.Errorf("running post failed: %w", err) } if len(postOut) == 0 { return nil, xerrors.Errorf("received proofs back from generate window post") } params.Proofs = postOut for _, sector := range postSkipped { params.Partitions[sidToPart[sector.Number]].Skipped.Set(uint64(sector.Number)) } elapsed := time.Since(tsStart) commEpoch := di.Open commRand, err := s.api.ChainGetRandomnessFromTickets(ctx, ts.Key(), crypto.DomainSeparationTag_PoStChainCommit, commEpoch, nil) if err != nil { return nil, xerrors.Errorf("failed to get chain randomness for windowPost (ts=%d; deadline=%d): %w", ts.Height(), di, err) } params.ChainCommitEpoch = commEpoch params.ChainCommitRand = commRand log.Infow("submitting window PoSt", "elapsed", elapsed) return params, nil } func (s *WindowPoStScheduler) sectorInfo(ctx context.Context, deadlineSectors abi.BitField, ts *types.TipSet) ([]abi.SectorInfo, error) { sset, err := s.api.StateMinerSectors(ctx, s.actor, &deadlineSectors, false, ts.Key()) if err != nil { return nil, err } sbsi := make([]abi.SectorInfo, len(sset)) for k, sector := range sset { sbsi[k] = abi.SectorInfo{ SectorNumber: sector.ID, SealedCID: sector.Info.SealedCID, SealProof: sector.Info.SealProof, } } return sbsi, nil } func (s *WindowPoStScheduler) submitPost(ctx context.Context, proof *miner.SubmitWindowedPoStParams) (*types.SignedMessage, error) { ctx, span := trace.StartSpan(ctx, "storage.commitPost") defer span.End() var sm *types.SignedMessage enc, aerr := actors.SerializeParams(proof) if aerr != nil { return nil, xerrors.Errorf("could not serialize submit post parameters: %w", aerr) } msg := &types.Message{ To: s.actor, From: s.worker, Method: builtin.MethodsMiner.SubmitWindowedPoSt, Params: enc, Value: types.NewInt(1000), // currently hard-coded late fee in actor, returned if not late } spec := &api.MessageSendSpec{MaxFee: abi.TokenAmount(s.feeCfg.MaxWindowPoStGasFee)} s.setSender(ctx, msg, spec) // TODO: consider maybe caring about the output sm, err := s.api.MpoolPushMessage(ctx, msg, spec) if err != nil { return nil, xerrors.Errorf("pushing message to mpool: %w", err) } log.Infof("Submitted window post: %s", sm.Cid()) go func() { rec, err := s.api.StateWaitMsg(context.TODO(), sm.Cid(), build.MessageConfidence) if err != nil { log.Error(err) return } if rec.Receipt.ExitCode == 0 { return } log.Errorf("Submitting window post %s failed: exit %d", sm.Cid(), rec.Receipt.ExitCode) }() return sm, nil } func (s *WindowPoStScheduler) setSender(ctx context.Context, msg *types.Message, spec *api.MessageSendSpec) { mi, err := s.api.StateMinerInfo(ctx, s.actor, types.EmptyTSK) if err != nil { log.Errorw("error getting miner info", "error", err) // better than just failing msg.From = s.worker return } gm, err := s.api.GasEstimateMessageGas(ctx, msg, spec, types.EmptyTSK) if err != nil { log.Errorw("estimating gas", "error", err) msg.From = s.worker return } *msg = *gm minFunds := big.Add(msg.RequiredFunds(), msg.Value) pa, err := AddressFor(ctx, s.api, mi, PoStAddr, minFunds) if err != nil { log.Errorw("error selecting address for post", "error", err) msg.From = s.worker return } msg.From = pa }