sbft: fix restart bug and test

commit 24514a027377631cc4b3c53a56d7fbcad75b7e5b
Author: Simon Schubert <sis@zurich.ibm.com>
Date:   Wed Oct 26 18:59:33 2016 +0200

    sbft: resend in-flight messages on reconnect

    This allows a recently disconnected replica to catch up and cancel its
    request timer.

    Change-Id: I6e32176777fcf74e48bd2a6402793cf4712637d5
    Signed-off-by: Simon Schubert <sis@zurich.ibm.com>

commit 08232f825ae54569c8fd296717a8ad242d9ad84d
Author: Simon Schubert <sis@zurich.ibm.com>
Date:   Wed Oct 26 18:59:19 2016 +0200

    sbft: ignore duplicate preprepare messages

    Change-Id: Ibfb47ecf1e005bdd23ed40f3753b01f49dfffd94
    Signed-off-by: Simon Schubert <sis@zurich.ibm.com>

commit 8ba9a9548727c36b8e276003d9a0eb855b2a6afb
Author: Simon Schubert <sis@zurich.ibm.com>
Date:   Wed Oct 26 18:56:40 2016 +0200

    sbft: do not use non-deterministic map iteration in tests

    This change hides a bug that occasionally was exposed due to a specific
    map iteration order.  We add TestErroneousViewChange to explicitly test
    for this bug, which is fixed in a followup commit.

    Change-Id: I192903dd57d637ec000a628f9f61b44749caf602
    Signed-off-by: Simon Schubert <sis@zurich.ibm.com>

Change-Id: I32a3c424dd385704b77ed884b2ec0f6d04f6ac6c
Signed-off-by: Simon Schubert <sis@zurich.ibm.com>

Author: corecode

consensus/simplebft/checkpoint.go

@@ -21,14 +21,18 @@ import (
 	"reflect"
 )
 
-func (s *SBFT) sendCheckpoint() {
+func (s *SBFT) makeCheckpoint() *Checkpoint {
 	sig := s.sys.Sign(s.cur.subject.Digest)
 	c := &Checkpoint{
 		Seq:       s.cur.subject.Seq.Seq,
 		Digest:    s.cur.subject.Digest,
 		Signature: sig,
 	}
-	s.broadcast(&Msg{&Msg_Checkpoint{c}})
+	return c
+}
+
+func (s *SBFT) sendCheckpoint() {
+	s.broadcast(&Msg{&Msg_Checkpoint{s.makeCheckpoint()}})
 }
 
 func (s *SBFT) handleCheckpoint(c *Checkpoint, src uint64) {

consensus/simplebft/connection.go

@@ -16,6 +16,8 @@ limitations under the License.
 
 package simplebft
 
+import "github.com/golang/protobuf/proto"
+
 // Connection is an event from system to notify a new connection with
 // replica.
 // On connection, we send our latest (weak) checkpoint, and we expect
@@ -25,8 +27,6 @@ func (s *SBFT) Connection(replica uint64) {
 	batch.Payloads = nil // don't send the big payload
 	s.sys.Send(&Msg{&Msg_Hello{&batch}}, replica)
 
-	// TODO
-	//
 	// A reconnecting replica can play forward its blockchain to
 	// the batch listed in the hello message.  However, the
 	// currently in-flight batch will not be reflected in the
@@ -36,6 +36,32 @@ func (s *SBFT) Connection(replica uint64) {
 	// Therefore we also send the most recent (pre)prepare,
 	// commit, checkpoint so that the reconnecting replica can
 	// catch up on the in-flight batch.
+	//
+	// TODO We need to communicate the latest view to the
+	// connecting replica.  The new view message is not signed, so
+	// we cannot send that message.  The worst corner case is
+	// connecting right after a new-view message was received, and
+	// its xset batch is in-flight.
+
+	batchheader := &BatchHeader{}
+	err := proto.Unmarshal(batch.Header, batchheader)
+	if err != nil {
+		panic(err)
+	}
+
+	if s.cur.subject.Seq.Seq > batchheader.Seq {
+		if s.isPrimary() {
+			s.sys.Send(&Msg{&Msg_Preprepare{s.cur.preprep}}, replica)
+		} else {
+			s.sys.Send(&Msg{&Msg_Prepare{&s.cur.subject}}, replica)
+		}
+		if s.cur.sentCommit {
+			s.sys.Send(&Msg{&Msg_Commit{&s.cur.subject}}, replica)
+		}
+		if s.cur.executed {
+			s.sys.Send(&Msg{&Msg_Checkpoint{s.makeCheckpoint()}}, replica)
+		}
+	}
 }
 
 func (s *SBFT) handleHello(h *Batch, src uint64) {

consensus/simplebft/preprepare.go

@@ -50,6 +50,10 @@ func (s *SBFT) handlePreprepare(pp *Preprepare, src uint64) {
 		log.Infof("preprepare does not match expected %v, got %v", nextSeq, *pp.Seq)
 		return
 	}
+	if s.cur.subject.Seq.Seq == pp.Seq.Seq {
+		log.Infof("duplicate preprepare for %v", *pp.Seq)
+		return
+	}
 	var blockhash []byte
 	if pp.Batch != nil {
 		blockhash = hash(pp.Batch.Header)

consensus/simplebft/simplebft.go

@@ -129,17 +129,16 @@ func New(id uint64, config *Config, sys System) (*SBFT, error) {
 			s.seq = *pp.Seq
 			s.seq.Seq -= 1
 			s.handlePreprepare(pp, s.primaryIDView(pp.Seq.View))
+			// ideally we wouldn't send a prepare here
 		}
 	}
 	c := &Subject{}
 	if s.sys.Restore("commit", c) && reflect.DeepEqual(c, &s.cur.subject) {
 		s.cur.sentCommit = true
-		s.sendCommit()
 	}
 	ex := &Subject{}
 	if s.sys.Restore("execute", ex) && reflect.DeepEqual(c, &s.cur.subject) {
 		s.cur.executed = true
-		s.sendCheckpoint()
 	}
 
 	// XXX set active after checking with the network

consensus/simplebft/simplebft_test.go

@@ -34,8 +34,25 @@ func init() {
 }
 
 func connectAll(sys *testSystem) {
+	// map iteration is non-deterministic, so use linear iteration instead
+	max := uint64(0)
 	for _, a := range sys.adapters {
-		for _, b := range sys.adapters {
+		if a.id > max {
+			max = a.id
+		}
+	}
+
+	for i := uint64(0); i <= max; i++ {
+		a, ok := sys.adapters[i]
+		if !ok {
+			continue
+		}
+
+		for j := uint64(0); j <= max; j++ {
+			b, ok := sys.adapters[j]
+			if !ok {
+				continue
+			}
 			if a.id != b.id {
 				a.receiver.Connection(b.id)
 			}
@@ -342,7 +359,7 @@ func TestRestart(t *testing.T) {
 
 	testLog.Notice("restarting 0")
 	repls[0], _ = New(0, &Config{N: N, F: 1, BatchDurationNsec: 2000000000, BatchSizeBytes: 10, RequestTimeoutNsec: 20000000000}, adapters[0])
-	for _, a := range sys.adapters {
+	for _, a := range adapters {
 		if a.id != 0 {
 			a.receiver.Connection(0)
 			adapters[0].receiver.Connection(a.id)
@@ -394,7 +411,7 @@ func TestRestartAfterPrepare(t *testing.T) {
 			if p := msg.msg.GetPrepare(); p != nil && p.Seq.Seq == 3 && !restarted {
 				restarted = true
 				repls[0], _ = New(0, &Config{N: N, F: 1, BatchDurationNsec: 2000000000, BatchSizeBytes: 10, RequestTimeoutNsec: 20000000000}, adapters[0])
-				for _, a := range sys.adapters {
+				for _, a := range adapters {
 					if a.id != 0 {
 						a.receiver.Connection(0)
 						adapters[0].receiver.Connection(a.id)
@@ -462,7 +479,7 @@ func TestRestartAfterCommit(t *testing.T) {
 				restarted = true
 				testLog.Notice("restarting 0")
 				repls[0], _ = New(0, &Config{N: N, F: 1, BatchDurationNsec: 2000000000, BatchSizeBytes: 10, RequestTimeoutNsec: 20000000000}, adapters[0])
-				for _, a := range sys.adapters {
+				for _, a := range adapters {
 					if a.id != 0 {
 						a.receiver.Connection(0)
 						adapters[0].receiver.Connection(a.id)
@@ -503,8 +520,6 @@ func TestRestartAfterCommit(t *testing.T) {
 }
 
 func TestRestartAfterCheckpoint(t *testing.T) {
-	// TODO re-enable this test after https://jira.hyperledger.org/browse/FAB-624 has been resolved
-	t.Skip()
 	N := uint64(4)
 	sys := newTestSystem(N)
 	var repls []*SBFT
@@ -532,7 +547,7 @@ func TestRestartAfterCheckpoint(t *testing.T) {
 				restarted = true
 				testLog.Notice("restarting 0")
 				repls[0], _ = New(0, &Config{N: N, F: 1, BatchDurationNsec: 2000000000, BatchSizeBytes: 10, RequestTimeoutNsec: 20000000000}, adapters[0])
-				for _, a := range sys.adapters {
+				for _, a := range adapters {
 					if a.id != 0 {
 						a.receiver.Connection(0)
 						adapters[0].receiver.Connection(a.id)
@@ -571,3 +586,91 @@ func TestRestartAfterCheckpoint(t *testing.T) {
 		}
 	}
 }
+
+func TestErroneousViewChange(t *testing.T) {
+	N := uint64(4)
+	sys := newTestSystem(N)
+	var repls []*SBFT
+	var adapters []*testSystemAdapter
+	for i := uint64(0); i < N; i++ {
+		a := sys.NewAdapter(i)
+		s, err := New(i, &Config{N: N, F: 1, BatchDurationNsec: 2000000000, BatchSizeBytes: 10, RequestTimeoutNsec: 20000000000}, a)
+		if err != nil {
+			t.Fatal(err)
+		}
+		repls = append(repls, s)
+		adapters = append(adapters, a)
+	}
+
+	restarted := false
+
+	// network outage after prepares are received
+	sys.filterFn = func(e testElem) (testElem, bool) {
+		if msg, ok := e.ev.(*testMsgEvent); ok {
+			if msg.src == msg.dst || msg.src != 0 {
+				return e, true
+			}
+
+			if c := msg.msg.GetCheckpoint(); c != nil && c.Seq == 3 && !restarted {
+				restarted = true
+				testLog.Notice("restarting 0")
+				repls[0], _ = New(0, &Config{N: N, F: 1, BatchDurationNsec: 2000000000, BatchSizeBytes: 10, RequestTimeoutNsec: 20000000000}, adapters[0])
+				for _, a := range adapters {
+					if a.id != 0 {
+						a.receiver.Connection(0)
+						adapters[0].receiver.Connection(a.id)
+					}
+				}
+			}
+		}
+
+		return e, true
+	}
+
+	// iteration order here is essential to trigger the bug
+	outer := []uint64{2, 3, 0, 1}
+	inner := []uint64{0, 1, 2, 3}
+	for _, i := range outer {
+		a, ok := sys.adapters[i]
+		if !ok {
+			continue
+		}
+
+		for _, j := range inner {
+			b, ok := sys.adapters[j]
+			if !ok {
+				continue
+			}
+			if a.id != b.id {
+				a.receiver.Connection(b.id)
+			}
+		}
+	}
+	sys.Run()
+
+	// move to view 1
+	for _, r := range repls {
+		r.sendViewChange()
+	}
+
+	r1 := []byte{1, 2, 3}
+	repls[0].Request(r1)
+	sys.Run()
+
+	r2 := []byte{3, 1, 2}
+	r3 := []byte{3, 5, 2}
+	repls[1].Request(r2)
+	repls[1].Request(r3)
+	sys.Run()
+	for _, a := range adapters {
+		if len(a.batches) != 3 {
+			t.Fatal("expected execution of 3 batches")
+		}
+		if !reflect.DeepEqual([][]byte{r1}, a.batches[1].Payloads) {
+			t.Error("wrong request executed (1)")
+		}
+		if !reflect.DeepEqual([][]byte{r2, r3}, a.batches[2].Payloads) {
+			t.Error("wrong request executed (2)")
+		}
+	}
+}