Merge pull request #538 from FactomProject/FD-628_DBSig_extra_election

Fix handling of extra DBSig messages

engine/factomd_test.go

@@ -17,6 +17,7 @@ import (
 	"github.com/FactomProject/factomd/activations"
 	"github.com/FactomProject/factomd/common/globals"
 	"github.com/FactomProject/factomd/common/primitives"
+	"github.com/FactomProject/factomd/common/primitives/random"
 	. "github.com/FactomProject/factomd/engine"
 	"github.com/FactomProject/factomd/state"
 	"github.com/FactomProject/factomd/wsapi"
@@ -1191,3 +1192,122 @@ func CheckAuthoritySet(leaders int, audits int, t *testing.T) {
 		t.Fail()
 	}
 }
+
+func TestDBsigElectionEvery2Block(t *testing.T) {
+	if ranSimTest {
+		return
+	}
+
+	ranSimTest = true
+
+	iterations := 1
+	state0 := SetupSim("LLLLLLAF", "LOCAL", map[string]string{"--debuglog": "fault|badmsg|network|process|dbsig", "--faulttimeout": "10"}, t)
+
+	StatusEveryMinute(state0)
+	runCmd("S10") // Set Drop Rate to 1.0 on everyone
+
+	CheckAuthoritySet(6, 1, t)
+
+	for j := 0; j <= iterations; j++ {
+		// for leader 1 thu 7 kill each in turn
+		for i := 1; i < 7; i++ {
+			s := GetFnodes()[i].State
+			if !s.IsLeader() {
+				panic("Can't kill a audit and cause an election")
+			}
+			WaitForMinute(s, 9) // wait till the victim is at minute 9
+			// wait till minute flips
+			for s.CurrentMinute != 0 {
+				runtime.Gosched()
+			}
+			s.SetNetStateOff(true) // kill the victim
+			s.LogPrintf("faulting", "Stopped %s\n", s.FactomNodeName)
+			WaitForMinute(state0, 1) // Wait till FNode0 move ahead a minute (the election is over)
+			s.LogPrintf("faulting", "Start %s\n", s.FactomNodeName)
+			s.SetNetStateOff(false) // resurrect the victim
+
+			WaitBlocks(state0, 2)    // wait till the victim is back as the audit server
+			WaitForMinute(state0, 8) // Wait till ablock is loaded
+
+			CheckAuthoritySet(6, 1, t) // check the authority set is as expected
+		}
+	}
+	t.Log("Shutting down the network")
+	for _, fn := range GetFnodes() {
+		fn.State.ShutdownChan <- 1
+	}
+
+}
+
+func TestDBSigElection(t *testing.T) {
+	if ranSimTest {
+		return
+	}
+	ranSimTest = true
+
+	state0 := SetupSim("LLLAF", "LOCAL", map[string]string{"--debuglog": "fault|badmsg|network|process|dbsig", "--faulttimeout": "10"}, t)
+	StatusEveryMinute(state0)
+
+	CheckAuthoritySet(3, 1, t)
+
+	s := GetFnodes()[2].State
+	if !s.IsLeader() {
+		panic("Can't kill a audit and cause an election")
+	}
+	WaitForMinute(s, 9) // wait till the victim is at minute 9
+	// wait till minute flips
+	for s.CurrentMinute != 0 {
+		runtime.Gosched()
+	}
+	s.SetNetStateOff(true) // kill the victim
+	s.LogPrintf("faulting", "Stopped %s\n", s.FactomNodeName)
+	WaitForMinute(state0, 1) // Wait till FNode0 move ahead a minute (the election is over)
+	s.LogPrintf("faulting", "Start %s\n", s.FactomNodeName)
+	s.SetNetStateOff(false) // resurrect the victim
+
+	WaitBlocks(state0, 2)    // wait till the victim is back as the audit server
+	WaitForMinute(state0, 1) // Wait till ablock is loaded
+	WaitForMinute(state0, 2) // Wait another couple minute in case any nodes are behind
+
+	CheckAuthoritySet(3, 1, t) // check the authority set is as expected
+
+	t.Log("Shutting down the network")
+	for _, fn := range GetFnodes() {
+		fn.State.ShutdownChan <- 1
+	}
+
+}
+
+// Cheap tests for developing binary search commits algorithm
+
+func TestPass(t *testing.T) {
+	if ranSimTest {
+		return
+	}
+
+	ranSimTest = true
+
+}
+
+func TestFail(t *testing.T) {
+	if ranSimTest {
+		return
+	}
+
+	ranSimTest = true
+	t.Fatal("Failed")
+
+}
+
+func TestRandom(t *testing.T) {
+	if ranSimTest {
+		return
+	}
+
+	ranSimTest = true
+
+	if random.RandUInt8() > 200 {
+		t.Fatal("Failed")
+	}
+
+}

engine/search.sh

@@ -0,0 +1,49 @@
+#/bin/sh
+#set -x
+
+# git log v5.4.2 | head -n 1092 | grep -B5 " Merge " | grep -E "^commit" | awk ' {print $2;}' > x
+
+test="$1"
+shift
+################################
+# AWK scripts                  #
+################################
+read -d '' scriptVariable << 'EOF'
+
+func mysystem(cmd) {
+    print cmd;
+    print "\n",cmd,"\n" ;
+    rc = system(cmd " 2>&1");
+    if(rc!=0) {
+      print "Exit", rc;
+      exit(rc);
+    }
+    return rc;
+}
+
+
+/^#/ {print "Skipping", $1; next;}
+
+ {branches[x++] = $0;}
+
+
+END {
+ PROCINFO["@ind_num_asc"]
+    for(i in branches) {
+      branch = branches[i];
+      print "Testing Branch ", branch;
+      mysystem("git reset --hard");
+      mysystem("git checkout " branch);
+      mysystem("git checkout Rolling_DBSIG_Test_v5.2 factomd_test.go");
+      rc = system("go test --run " test);
+      print i, branch, "Exit Code", rc
+    }
+ }
+EOF
+################################
+# End of AWK Scripts           #
+################################
+rm -f runlog.txt
+awk -v test=$test "$scriptVariable" $@ 2>&1 | tee -i runlog.txt  | grep -E "Testing Branch|Exit Code"
+
+

state/stateConsensus.go

@@ -1222,8 +1222,21 @@ func (s *State) LeaderExecuteDBSig(m interfaces.IMsg) {
 		m.FollowerExecute(s)
 		return
 	}
+
 	if pl.VMs[dbs.VMIndex].Height > 0 {
-		s.LogMessage("executeMsg", "drop, slot 0 taken by", pl.VMs[dbs.VMIndex].List[0])
+		s.LogPrintf("executeMsg", "DBSig issue height = %d, length = %d", pl.VMs[dbs.VMIndex].Height, len(pl.VMs[dbs.VMIndex].List))
+		s.LogMessage("executeMsg", "drop, already processed ", pl.VMs[dbs.VMIndex].List[0])
+		return
+	}
+
+	if len(pl.VMs[dbs.VMIndex].List) > 0 && pl.VMs[dbs.VMIndex].List[0] != nil {
+		s.LogPrintf("executeMsg", "DBSig issue height = %d, length = %d", pl.VMs[dbs.VMIndex].Height, len(pl.VMs[dbs.VMIndex].List))
+		if pl.VMs[dbs.VMIndex].List[0] != m {
+			s.LogMessage("executeMsg", "drop, slot 0 taken by", pl.VMs[dbs.VMIndex].List[0])
+		} else {
+			s.LogMessage("executeMsg", "duplicate execute", pl.VMs[dbs.VMIndex].List[0])
+		}
+
 		return
 	}