Merge pull request #797 from filecoin-project/fix/fpost-restart

Restart fPoSt after fail

storage/fpost_run.go

@@ -14,6 +14,14 @@ import (
 	"github.com/filecoin-project/lotus/lib/sectorbuilder"
 )
 
+func (s *fpostScheduler) failPost(eps uint64) {
+	s.failLk.Lock()
+	if eps > s.failed {
+		s.failed = eps
+	}
+	s.failLk.Unlock()
+}
+
 func (s *fpostScheduler) doPost(ctx context.Context, eps uint64, ts *types.TipSet) {
 	ctx, abort := context.WithCancel(ctx)
 
@@ -29,13 +37,16 @@ func (s *fpostScheduler) doPost(ctx context.Context, eps uint64, ts *types.TipSe
 		proof, err := s.runPost(ctx, eps, ts)
 		if err != nil {
 			log.Errorf("runPost failed: %+v", err)
+			s.failPost(eps)
 			return
 		}
 
 		if err := s.submitPost(ctx, proof); err != nil {
 			log.Errorf("submitPost failed: %+v", err)
+			s.failPost(eps)
 			return
 		}
+
 	}()
 }
 

storage/fpost_sched.go

@@ -2,6 +2,7 @@ package storage
 
 import (
 	"context"
+	"sync"
 
 	"go.opencensus.io/trace"
 	"golang.org/x/xerrors"
@@ -29,6 +30,9 @@ type fpostScheduler struct {
 	// if a post is in progress, this indicates for which ElectionPeriodStart
 	activeEPS uint64
 	abort     context.CancelFunc
+
+	failed uint64 // eps
+	failLk sync.Mutex
 }
 
 func (s *fpostScheduler) run(ctx context.Context) {
@@ -111,6 +115,13 @@ func (s *fpostScheduler) update(ctx context.Context, new *types.TipSet) error {
 		return err
 	}
 
+	s.failLk.Lock()
+	if s.failed > 0 {
+		s.failed = 0
+		s.activeEPS = 0
+	}
+	s.failLk.Unlock()
+
 	if newEPS == s.activeEPS {
 		return nil
 	}