Merge pull request #9680 from filecoin-project/feat/config-post-check…

…-timeout

feat: wdpost: Configurable pre-check timeouts

documentation/en/default-lotus-miner-config.toml

@@ -325,6 +325,29 @@
   # env var: LOTUS_PROVING_PARALLELCHECKLIMIT
   #ParallelCheckLimit = 128
 
+  # Maximum amount of time a proving pre-check can take for a sector. If the check times out the sector will be skipped
+  # 
+  # WARNING: Setting this value too low risks in sectors being skipped even though they are accessible, just reading the
+  # test challenge took longer than this timeout
+  # WARNING: Setting this value too high risks missing PoSt deadline in case IO operations related to this sector are
+  # blocked (e.g. in case of disconnected NFS mount)
+  #
+  # type: Duration
+  # env var: LOTUS_PROVING_SINGLECHECKTIMEOUT
+  #SingleCheckTimeout = "10m0s"
+
+  # Maximum amount of time a proving pre-check can take for an entire partition. If the check times out, sectors in
+  # the partition which didn't get checked on time will be skipped
+  # 
+  # WARNING: Setting this value too low risks in sectors being skipped even though they are accessible, just reading the
+  # test challenge took longer than this timeout
+  # WARNING: Setting this value too high risks missing PoSt deadline in case IO operations related to this partition are
+  # blocked or slow
+  #
+  # type: Duration
+  # env var: LOTUS_PROVING_PARTITIONCHECKTIMEOUT
+  #PartitionCheckTimeout = "20m0s"
+
   # Disable Window PoSt computation on the lotus-miner process even if no window PoSt workers are present.
   # 
   # WARNING: If no windowPoSt workers are connected, window PoSt WILL FAIL resulting in faulty sectors which will need

node/config/def.go

@@ -141,7 +141,9 @@ func DefaultStorageMiner() *StorageMiner {
 		},
 
 		Proving: ProvingConfig{
-			ParallelCheckLimit: 128,
+			ParallelCheckLimit:    128,
+			PartitionCheckTimeout: Duration(20 * time.Minute),
+			SingleCheckTimeout:    Duration(10 * time.Minute),
 		},
 
 		Storage: SealerConfig{

node/config/types.go

@@ -230,6 +230,23 @@ type ProvingConfig struct {
 	// 'lotus-miner proving compute window-post 0'
 	ParallelCheckLimit int
 
+	// Maximum amount of time a proving pre-check can take for a sector. If the check times out the sector will be skipped
+	//
+	// WARNING: Setting this value too low risks in sectors being skipped even though they are accessible, just reading the
+	// test challenge took longer than this timeout
+	// WARNING: Setting this value too high risks missing PoSt deadline in case IO operations related to this sector are
+	// blocked (e.g. in case of disconnected NFS mount)
+	SingleCheckTimeout Duration
+
+	// Maximum amount of time a proving pre-check can take for an entire partition. If the check times out, sectors in
+	// the partition which didn't get checked on time will be skipped
+	//
+	// WARNING: Setting this value too low risks in sectors being skipped even though they are accessible, just reading the
+	// test challenge took longer than this timeout
+	// WARNING: Setting this value too high risks missing PoSt deadline in case IO operations related to this partition are
+	// blocked or slow
+	PartitionCheckTimeout Duration
+
 	// Disable Window PoSt computation on the lotus-miner process even if no window PoSt workers are present.
 	//
 	// WARNING: If no windowPoSt workers are connected, window PoSt WILL FAIL resulting in faulty sectors which will need

storage/sealer/faults.go

@@ -5,7 +5,6 @@ import (
 	"crypto/rand"
 	"fmt"
 	"sync"
-	"time"
 
 	"golang.org/x/xerrors"
 
@@ -15,8 +14,6 @@ import (
 	"github.com/filecoin-project/lotus/storage/sealer/storiface"
 )
 
-var PostCheckTimeout = 160 * time.Second
-
 // FaultTracker TODO: Track things more actively
 type FaultTracker interface {
 	CheckProvable(ctx context.Context, pp abi.RegisteredPoStProof, sectors []storiface.SectorRef, rg storiface.RGetter) (map[abi.SectorID]string, error)
@@ -50,14 +47,22 @@ func (m *Manager) CheckProvable(ctx context.Context, pp abi.RegisteredPoStProof,
 		badLk.Unlock()
 	}
 
+	if m.partitionCheckTimeout > 0 {
+		var cancel2 context.CancelFunc
+		ctx, cancel2 = context.WithTimeout(ctx, m.partitionCheckTimeout)
+		defer cancel2()
+	}
+
 	var wg sync.WaitGroup
 	wg.Add(len(sectors))
 
 	for _, sector := range sectors {
 		select {
 		case throttle <- struct{}{}:
 		case <-ctx.Done():
-			return nil, ctx.Err()
+			addBad(sector.ID, fmt.Sprintf("waiting for check worker: %s", ctx.Err()))
+			wg.Done()
+			continue
 		}
 
 		go func(sector storiface.SectorRef) {
@@ -107,8 +112,13 @@ func (m *Manager) CheckProvable(ctx context.Context, pp abi.RegisteredPoStProof,
 				return
 			}
 
-			vctx, cancel2 := context.WithTimeout(ctx, PostCheckTimeout)
-			defer cancel2()
+			vctx := ctx
+
+			if m.singleCheckTimeout > 0 {
+				var cancel2 context.CancelFunc
+				vctx, cancel2 = context.WithTimeout(ctx, m.singleCheckTimeout)
+				defer cancel2()
+			}
 
 			_, err = m.storage.GenerateSingleVanillaProof(vctx, sector.ID.Miner, storiface.PostSectorChallenge{
 				SealProof:    sector.ProofType,

storage/sealer/manager.go

@@ -7,6 +7,7 @@ import (
 	"net/http"
 	"sort"
 	"sync"
+	"time"
 
 	"github.com/google/uuid"
 	"github.com/hashicorp/go-multierror"
@@ -72,6 +73,8 @@ type Manager struct {
 	work   *statestore.StateStore
 
 	parallelCheckLimit        int
+	singleCheckTimeout        time.Duration
+	partitionCheckTimeout     time.Duration
 	disableBuiltinWindowPoSt  bool
 	disableBuiltinWinningPoSt bool
 	disallowRemoteFinalize    bool
@@ -121,6 +124,8 @@ func New(ctx context.Context, lstor *paths.Local, stor paths.Store, ls paths.Loc
 		localProver: prover,
 
 		parallelCheckLimit:        pc.ParallelCheckLimit,
+		singleCheckTimeout:        time.Duration(pc.SingleCheckTimeout),
+		partitionCheckTimeout:     time.Duration(pc.PartitionCheckTimeout),
 		disableBuiltinWindowPoSt:  pc.DisableBuiltinWindowPoSt,
 		disableBuiltinWinningPoSt: pc.DisableBuiltinWinningPoSt,
 		disallowRemoteFinalize:    sc.DisallowRemoteFinalize,