func createRunningPbftWithManager(id uint64, config *viper.Viper, stack innerStack) (*pbftCore, events.Manager) {
manager := events.NewManagerImpl()
core := newPbftCore(id, loadConfig(), stack, events.NewTimerFactoryImpl(manager))
manager.SetReceiver(core)
manager.Start()
return core, manager
}
func makePBFTNetwork(N int, config *viper.Viper) *pbftNetwork {
if config == nil {
config = loadConfig()
}
config.Set("general.N", N)
config.Set("general.f", (N-1)/3)
endpointFunc := func(id uint64, net *testnet) endpoint {
tep := makeTestEndpoint(id, net)
pe := &pbftEndpoint{
testEndpoint: tep,
manager: events.NewManagerImpl(),
}
pe.sc = &simpleConsumer{
pe: pe,
}
pe.pbft = newPbftCore(id, config, pe.sc, events.NewTimerFactoryImpl(pe.manager))
pe.manager.SetReceiver(pe.pbft)
pe.manager.Start()
return pe
}
pn := &pbftNetwork{testnet: makeTestnet(N, endpointFunc)}
pn.pbftEndpoints = make([]*pbftEndpoint, len(pn.endpoints))
for i, ep := range pn.endpoints {
pn.pbftEndpoints[i] = ep.(*pbftEndpoint)
pn.pbftEndpoints[i].sc.pbftNet = pn
}
return pn
}
func (shim *legacyGenericShim) init(id uint64, config *viper.Viper, consumer legacyInnerStack) {
shim.pbft = &legacyPbftShim{
closed: make(chan struct{}),
manager: events.NewManagerImpl(),
consumer: consumer,
}
shim.pbft.pbftCore = newPbftCore(id, config, consumer, events.NewTimerFactoryImpl(shim.pbft.manager))
shim.pbft.manager.SetReceiver(shim.pbft)
logger.Debug("Replica %d Consumer is %p", shim.pbft.id, shim.pbft.consumer)
shim.pbft.manager.Start()
logger.Debug("Replica %d legacyGenericShim now initialized: %v", id, shim)
}
func newObcBatch(id uint64, config *viper.Viper, stack consensus.Stack) *obcBatch {
var err error
op := &obcBatch{
obcGeneric: obcGeneric{stack: stack},
}
op.persistForward.persistor = stack
logger.Debugf("Replica %d obtaining startup information", id)
op.manager = events.NewManagerImpl() // TODO, this is hacky, eventually rip it out
op.manager.SetReceiver(op)
etf := events.NewTimerFactoryImpl(op.manager)
op.pbft = newPbftCore(id, config, op, etf)
op.manager.Start()
op.externalEventReceiver.manager = op.manager
op.broadcaster = newBroadcaster(id, op.pbft.N, op.pbft.f, stack)
op.batchSize = config.GetInt("general.batchsize")
op.batchStore = nil
op.batchTimeout, err = time.ParseDuration(config.GetString("general.timeout.batch"))
if err != nil {
panic(fmt.Errorf("Cannot parse batch timeout: %s", err))
}
logger.Infof("PBFT Batch size = %d", op.batchSize)
logger.Infof("PBFT Batch timeout = %v", op.batchTimeout)
if op.batchTimeout >= op.pbft.requestTimeout {
op.pbft.requestTimeout = 3 * op.batchTimeout / 2
logger.Warningf("Configured request timeout must be greater than batch timeout, setting to %v", op.pbft.requestTimeout)
}
if op.pbft.requestTimeout >= op.pbft.nullRequestTimeout && op.pbft.nullRequestTimeout != 0 {
op.pbft.nullRequestTimeout = 3 * op.pbft.requestTimeout / 2
logger.Warningf("Configured null request timeout must be greater than request timeout, setting to %v", op.pbft.nullRequestTimeout)
}
op.incomingChan = make(chan *batchMessage)
op.batchTimer = etf.CreateTimer()
op.reqStore = newRequestStore()
op.deduplicator = newDeduplicator()
op.idleChan = make(chan struct{})
close(op.idleChan) // TODO remove eventually
return op
}
// TestReplicaCrash1 simulates the restart of replicas 0 and 1 after
// some state has been built (one request executed). At the time of
// the restart, replica 0 is also the primary. All three requests
// submitted should also be executed on all replicas.
func TestReplicaCrash1(t *testing.T) {
validatorCount := 4
config := loadConfig()
config.Set("general.K", 2)
config.Set("general.logmultiplier", 2)
net := makePBFTNetwork(validatorCount, config)
defer net.stop()
mkreq := func(n int64) *Request {
txTime := &gp.Timestamp{Seconds: n, Nanos: 0}
tx := &pb.Transaction{Type: pb.Transaction_CHAINCODE_DEPLOY, Timestamp: txTime}
txPacked, _ := proto.Marshal(tx)
return &Request{
Timestamp: &gp.Timestamp{Seconds: n, Nanos: 0},
Payload: txPacked,
ReplicaId: uint64(generateBroadcaster(validatorCount)),
}
}
net.pbftEndpoints[0].manager.Queue() <- mkreq(1)
net.process()
for id := 0; id < 2; id++ {
pe := net.pbftEndpoints[id]
pe.pbft = newPbftCore(uint64(id), loadConfig(), pe.sc, events.NewTimerFactoryImpl(pe.manager))
pe.manager.SetReceiver(pe.pbft)
pe.pbft.N = 4
pe.pbft.f = (4 - 1) / 3
pe.pbft.K = 2
pe.pbft.L = 2 * pe.pbft.K
}
net.pbftEndpoints[0].manager.Queue() <- mkreq(2)
net.pbftEndpoints[0].manager.Queue() <- (mkreq(3))
net.process()
for _, pep := range net.pbftEndpoints {
if pep.sc.executions != 3 {
t.Errorf("Expected 3 executions on replica %d, got %d", pep.id, pep.sc.executions)
continue
}
if pep.pbft.view != 0 {
t.Errorf("Replica %d should still be in view 0, is %v %d", pep.id, pep.pbft.activeView, pep.pbft.view)
}
}
}
// TestReplicaCrash3 simulates the restart requiring a view change
// to a checkpoint which was restored from the persistence state
// Replicas 0,1,2 participate up to a checkpoint, then all crash
// Then replicas 0,1,3 start back up, and a view change must be
// triggered to get vp3 up to speed
func TestReplicaCrash3(t *testing.T) {
validatorCount := 4
config := loadConfig()
config.Set("general.K", 2)
config.Set("general.logmultiplier", 2)
net := makePBFTNetwork(validatorCount, config)
defer net.stop()
twoOffline := false
threeOffline := true
net.filterFn = func(src int, dst int, msg []byte) []byte {
if twoOffline && dst == 2 { // 2 is 'offline'
return nil
}
if threeOffline && dst == 3 { // 3 is 'offline'
return nil
}
return msg
}
mkreq := func(n int64) *Request {
txTime := &gp.Timestamp{Seconds: n, Nanos: 0}
tx := &pb.Transaction{Type: pb.Transaction_CHAINCODE_DEPLOY, Timestamp: txTime}
txPacked, _ := proto.Marshal(tx)
return &Request{
Timestamp: &gp.Timestamp{Seconds: n, Nanos: 0},
Payload: txPacked,
ReplicaId: uint64(generateBroadcaster(validatorCount)),
}
}
for i := int64(1); i <= 8; i++ {
net.pbftEndpoints[0].manager.Queue() <- (mkreq(i))
}
net.process() // vp0,1,2 should have a stable checkpoint for seqNo 8
// Create new pbft instances to restore from persistence
for id := 0; id < 2; id++ {
pe := net.pbftEndpoints[id]
config := loadConfig()
config.Set("general.K", "2")
pe.pbft.close()
pe.pbft = newPbftCore(uint64(id), config, pe.sc, events.NewTimerFactoryImpl(pe.manager))
pe.manager.SetReceiver(pe.pbft)
pe.pbft.N = 4
pe.pbft.f = (4 - 1) / 3
pe.pbft.requestTimeout = 200 * time.Millisecond
}
threeOffline = false
twoOffline = true
// Because vp2 is 'offline', and vp3 is still at the genesis block, the network needs to make a view change
net.pbftEndpoints[0].manager.Queue() <- (mkreq(9))
net.process()
// Now vp0,1,3 should be in sync with 9 executions in view 1, and vp2 should be at 8 executions in view 0
for i, pep := range net.pbftEndpoints {
if i == 2 {
// 2 is 'offline'
if pep.pbft.view != 0 {
t.Errorf("Expected replica %d to be in view 0, got %d", pep.id, pep.pbft.view)
}
expectedExecutions := uint64(8)
if pep.sc.executions != expectedExecutions {
t.Errorf("Expected %d executions on replica %d, got %d", expectedExecutions, pep.id, pep.sc.executions)
}
continue
}
if pep.pbft.view != 1 {
t.Errorf("Expected replica %d to be in view 1, got %d", pep.id, pep.pbft.view)
}
expectedExecutions := uint64(9)
if pep.sc.executions != expectedExecutions {
t.Errorf("Expected %d executions on replica %d, got %d", expectedExecutions, pep.id, pep.sc.executions)
}
}
}