Merge branch 'develop' into add-index-for-first-in-epoch

.github/workflows/ci.yml

@@ -1,6 +1,7 @@
 name: CI
 
 env:
+  go-version: "1.21"
   GCLOUD_KEY: ${{ secrets.GCLOUD_KEY }}
   PROJECT_NAME: ${{ secrets.PROJECT_NAME }}
   CLUSTER_NAME: ${{ secrets.CLUSTER_NAME }}
@@ -65,7 +66,7 @@ jobs:
         uses: actions/setup-go@v4
         with:
           check-latest: true
-          go-version-file: "go.mod"
+          go-version: ${{ env.go-version }}
       - name: fmt, tidy, generate
         run: |
           make install
@@ -88,7 +89,7 @@ jobs:
         uses: actions/setup-go@v4
         with:
           check-latest: true
-          go-version-file: "go.mod"
+          go-version: ${{ env.go-version }}
       - name: setup env
         run: make install
       - name: staticcheck
@@ -126,7 +127,7 @@ jobs:
         uses: actions/setup-go@v4
         with:
           check-latest: true
-          go-version-file: "go.mod"
+          go-version: ${{ env.go-version }}
           cache: ${{ runner.arch != 'arm64' }}
       - name: setup env
         run: make install
@@ -174,7 +175,7 @@ jobs:
         uses: actions/setup-go@v4
         with:
           check-latest: true
-          go-version-file: "go.mod"
+          go-version: ${{ env.go-version }}
           cache: ${{ runner.arch != 'arm64' }}
       - name: Add OpenCL support - Ubuntu
         if: ${{ matrix.os == 'ubuntu-latest' }}
@@ -242,7 +243,7 @@ jobs:
         uses: actions/setup-go@v4
         with:
           check-latest: true
-          go-version-file: "go.mod"
+          go-version: ${{ env.go-version }}
           cache: ${{ runner.arch != 'arm64' }}
       - name: Add OpenCL support - Ubuntu
         if: ${{ matrix.os == 'ubuntu-latest' }}

CHANGELOG.md

@@ -19,10 +19,12 @@ Support for old certificate sync protocol is dropped. This update is incompatibl
 ### Features
 
 ### Improvements
+
 * [#4998](https://github.com/spacemeshos/go-spacemesh/pull/4998) First phase of state size reduction.
   Ephemeral data are deleted and state compacted at the time of upgrade. In steady-state, data is pruned periodically.
 * [#5021](https://github.com/spacemeshos/go-spacemesh/pull/5021) Drop support for old certificate sync protocol.
 * [#5024](https://github.com/spacemeshos/go-spacemesh/pull/5024) Active set will be saved in state separately from ballots.
+* [#5035](https://github.com/spacemeshos/go-spacemesh/pull/5035) Fix possible nil pointer panic when node fails to persist nipost builder state.
 
 ## v1.1.5
 

activation/nipost_state.go

@@ -22,30 +22,30 @@ const (
 )
 
 func write(path string, data []byte) error {
-	tmp, err := os.Create(fmt.Sprintf("%s.tmp", path))
+	tmpName := fmt.Sprintf("%s.tmp", path)
+	tmp, err := os.Create(tmpName)
 	if err != nil {
-		return fmt.Errorf("create temporary file %s: %w", tmp.Name(), err)
+		return fmt.Errorf("create temporary file %s: %w", tmpName, err)
 	}
+	defer tmp.Close()
 
 	checksum := crc64.New(crc64.MakeTable(crc64.ISO))
 	w := io.MultiWriter(tmp, checksum)
-	if _, err = w.Write(data); err != nil {
-		_ = tmp.Close()
+	if _, err := w.Write(data); err != nil {
 		return fmt.Errorf("write data %v: %w", tmp.Name(), err)
 	}
 
 	crc := make([]byte, crc64.Size)
 	binary.BigEndian.PutUint64(crc, checksum.Sum64())
-	if _, err = tmp.Write(crc); err != nil {
-		_ = tmp.Close()
+	if _, err := tmp.Write(crc); err != nil {
 		return fmt.Errorf("write checksum %s: %w", tmp.Name(), err)
 	}
 
-	if err = tmp.Close(); err != nil {
+	if err := tmp.Close(); err != nil {
 		return fmt.Errorf("failed to close tmp file %s: %w", tmp.Name(), err)
 	}
 
-	if err = atomic.ReplaceFile(tmp.Name(), path); err != nil {
+	if err := atomic.ReplaceFile(tmp.Name(), path); err != nil {
 		return fmt.Errorf("save file from %s, %s: %w", tmp.Name(), path, err)
 	}
 
@@ -57,7 +57,6 @@ func read(path string) ([]byte, error) {
 	if err != nil {
 		return nil, fmt.Errorf("open file %s: %w", path, err)
 	}
-
 	defer file.Close()
 
 	fInfo, err := file.Stat()
@@ -70,20 +69,19 @@ func read(path string) ([]byte, error) {
 
 	data := make([]byte, fInfo.Size()-crc64.Size)
 	checksum := crc64.New(crc64.MakeTable(crc64.ISO))
-	if _, err = io.TeeReader(file, checksum).Read(data); err != nil {
+	if _, err := io.TeeReader(file, checksum).Read(data); err != nil {
 		return nil, fmt.Errorf("read file %s: %w", path, err)
 	}
 
 	saved := make([]byte, crc64.Size)
-	if _, err = file.Read(saved); err != nil {
+	if _, err := file.Read(saved); err != nil {
 		return nil, fmt.Errorf("read checksum %s: %w", path, err)
 	}
 
 	savedChecksum := binary.BigEndian.Uint64(saved)
 
 	if savedChecksum != checksum.Sum64() {
-		return nil, fmt.Errorf(
-			"wrong checksum 0x%X, computed 0x%X", savedChecksum, checksum.Sum64())
+		return nil, fmt.Errorf("wrong checksum 0x%X, computed 0x%X", savedChecksum, checksum.Sum64())
 	}
 
 	return data, nil

go.mod

@@ -1,6 +1,6 @@
 module github.com/spacemeshos/go-spacemesh
 
-go 1.21
+go 1.21.1
 
 require (
 	cloud.google.com/go/storage v1.33.0

hare3/hare.go

@@ -381,24 +381,24 @@ func (h *Hare) run(layer types.LayerID, beacon types.Beacon, proto *protocol) er
 	if err := h.onOutput(layer, current, proto.Next(vrf != nil), vrf); err != nil {
 		return err
 	}
-	walltime = walltime.Add(h.config.RoundDuration)
 	result := false
 	for {
+		walltime = walltime.Add(h.config.RoundDuration)
+		current := proto.IterRound
+		var vrf *types.HareEligibility
+		if current.IsMessageRound() {
+			start := time.Now()
+			vrf = h.oracle.active(h.signer.NodeID(), layer, current)
+			activeLatency.Observe(time.Since(start).Seconds())
+		}
+		h.tracer.OnActive(vrf)
 		select {
 		case <-h.wallclock.After(walltime.Sub(h.wallclock.Now())):
 			h.log.Debug("execute round",
 				zap.Uint32("lid", layer.Uint32()),
 				zap.Uint8("iter", proto.Iter), zap.Stringer("round", proto.Round),
+				zap.Bool("active", vrf != nil),
 			)
-			current := proto.IterRound
-			var vrf *types.HareEligibility
-			if current.IsMessageRound() {
-				start := time.Now()
-				vrf = h.oracle.active(h.signer.NodeID(), layer, current)
-				activeLatency.Observe(time.Since(start).Seconds())
-			}
-			h.tracer.OnActive(vrf)
-
 			out := proto.Next(vrf != nil)
 			if out.result != nil {
 				result = true
@@ -416,7 +416,6 @@ func (h *Hare) run(layer types.LayerID, beacon types.Beacon, proto *protocol) er
 				return fmt.Errorf("hare failed to reach consensus in %d iterations",
 					h.config.IterationsLimit)
 			}
-			walltime = walltime.Add(h.config.RoundDuration)
 		case <-h.ctx.Done():
 			return nil
 		}
@@ -428,23 +427,17 @@ func (h *Hare) onOutput(layer types.LayerID, ir IterRound, out output, vrf *type
 		out.message.Layer = layer
 		out.message.Eligibility = *vrf
 		out.message.Sender = h.signer.NodeID()
+		out.message.Signature = h.signer.Sign(signing.HARE, out.message.ToMetadata().ToBytes())
+		if err := h.pubsub.Publish(h.ctx, h.config.ProtocolName, out.message.ToBytes()); err != nil {
+			h.log.Error("failed to publish", zap.Inline(out.message), zap.Error(err))
+		}
 	}
 	h.log.Debug("round output",
 		zap.Uint32("lid", layer.Uint32()),
 		zap.Uint8("iter", ir.Iter), zap.Stringer("round", ir.Round),
 		zap.Inline(&out),
-		zap.Bool("active", vrf != nil),
 	)
-	if out.message != nil {
-		h.eg.Go(func() error {
-			out.message.Signature = h.signer.Sign(signing.HARE, out.message.ToMetadata().ToBytes())
-			if err := h.pubsub.Publish(h.ctx, h.config.ProtocolName, out.message.ToBytes()); err != nil {
-				h.log.Error("failed to publish", zap.Inline(out.message), zap.Error(err))
-			}
-			h.tracer.OnMessageSent(out.message)
-			return nil
-		})
-	}
+	h.tracer.OnMessageSent(out.message)
 	if out.coin != nil {
 		select {
 		case <-h.ctx.Done():