Merge pull request #438 from tendermint/replay-fix

consensus: timeout on replayLastBlock

blockchain/pool.go

@@ -63,7 +63,6 @@ func NewBlockPool(start int, requestsCh chan<- BlockRequest, timeoutsCh chan<- s
 }
 
 func (pool *BlockPool) OnStart() error {
-	pool.BaseService.OnStart()
 	go pool.makeRequestersRoutine()
 	pool.startTime = time.Now()
 	return nil
@@ -409,7 +408,6 @@ func newBPRequester(pool *BlockPool, height int) *bpRequester {
 }
 
 func (bpr *bpRequester) OnStart() error {
-	bpr.BaseService.OnStart()
 	go bpr.requestRoutine()
 	return nil
 }

config/tendermint/config.go

@@ -83,6 +83,9 @@ func GetConfig(rootDir string) cfg.Config {
 	mapConfig.SetDefault("block_size", 10000)      // max number of txs
 	mapConfig.SetDefault("block_part_size", 65536) // part size 64K
 	mapConfig.SetDefault("disable_data_hash", false)
+
+	// all timeouts are in ms
+	mapConfig.SetDefault("timeout_handshake", 10000)
 	mapConfig.SetDefault("timeout_propose", 3000)
 	mapConfig.SetDefault("timeout_propose_delta", 500)
 	mapConfig.SetDefault("timeout_prevote", 1000)
@@ -90,6 +93,7 @@ func GetConfig(rootDir string) cfg.Config {
 	mapConfig.SetDefault("timeout_precommit", 1000)
 	mapConfig.SetDefault("timeout_precommit_delta", 500)
 	mapConfig.SetDefault("timeout_commit", 1000)
+
 	// make progress asap (no `timeout_commit`) on full precommit votes
 	mapConfig.SetDefault("skip_timeout_commit", false)
 	mapConfig.SetDefault("mempool_recheck", true)

config/tendermint_test/config.go

@@ -93,6 +93,7 @@ func ResetConfig(localPath string) cfg.Config {
 	mapConfig.SetDefault("block_size", 10000)
 	mapConfig.SetDefault("block_part_size", 65536) // part size 64K
 	mapConfig.SetDefault("disable_data_hash", false)
+	mapConfig.SetDefault("timeout_handshake", 10000)
 	mapConfig.SetDefault("timeout_propose", 2000)
 	mapConfig.SetDefault("timeout_propose_delta", 1)
 	mapConfig.SetDefault("timeout_prevote", 10)

consensus/replay.go

@@ -190,6 +190,8 @@ func (h *Handshaker) NBlocks() int {
 	return h.nBlocks
 }
 
+var ErrReplayLastBlockTimeout = errors.New("Timed out waiting for last block to be replayed")
+
 // TODO: retry the handshake/replay if it fails ?
 func (h *Handshaker) Handshake(proxyApp proxy.AppConns) error {
 	// handshake is done via info request on the query conn
@@ -207,7 +209,11 @@ func (h *Handshaker) Handshake(proxyApp proxy.AppConns) error {
 
 	// replay blocks up to the latest in the blockstore
 	_, err = h.ReplayBlocks(appHash, blockHeight, proxyApp)
-	if err != nil {
+	if err == ErrReplayLastBlockTimeout {
+		log.Warn("Failed to sync via handshake. Trying other means. If they fail, please increase the timeout_handshake parameter")
+		return nil
+
+	} else if err != nil {
 		return errors.New(Fmt("Error on replay: %v", err))
 	}
 
@@ -286,10 +292,11 @@ func (h *Handshaker) ReplayBlocks(appHash []byte, appBlockHeight int, proxyApp p
 
 func (h *Handshaker) replayBlocks(proxyApp proxy.AppConns, appBlockHeight, storeBlockHeight int, useReplayFunc bool) ([]byte, error) {
 	// App is further behind than it should be, so we need to replay blocks.
-	// We replay all blocks from appBlockHeight+1 to storeBlockHeight-1,
+	// We replay all blocks from appBlockHeight+1.
-	// and let the final block be replayed through ReplayBlocks.
+	// If useReplayFunc == true, stop short of the last block
+	// so it can be replayed using the WAL in ReplayBlocks.
 	// Note that we don't have an old version of the state,
-	// so we by-pass state validation using applyBlock here.
+	// so we by-pass state validation using sm.ApplyBlock.
 
 	var appHash []byte
 	var err error
@@ -328,9 +335,20 @@ func (h *Handshaker) replayLastBlock(proxyApp proxy.AppConnConsensus) ([]byte, e
 	newBlockCh := subscribeToEvent(evsw, "consensus-replay", types.EventStringNewBlock(), 1)
 
 	// run through the WAL, commit new block, stop
-	cs.Start()
+	if _, err := cs.Start(); err != nil {
-	<-newBlockCh // TODO: use a timeout and return err?
+		return nil, err
-	cs.Stop()
+	}
+	defer cs.Stop()
+
+	timeout := h.config.GetInt("timeout_handshake")
+	timer := time.NewTimer(time.Duration(timeout) * time.Millisecond)
+	log.Notice("Attempting to replay last block", "height", h.store.Height(), "timeout", timeout)
+
+	select {
+	case <-newBlockCh:
+	case <-timer.C:
+		return nil, ErrReplayLastBlockTimeout
+	}
 
 	h.nBlocks += 1
 

consensus/replay_file.go

@@ -127,8 +127,6 @@ func (pb *playback) replayReset(count int, newStepCh chan interface{}) error {
 }
 
 func (cs *ConsensusState) startForReplay() {
-	// don't want to start full cs
-	cs.BaseService.OnStart()
 
 	log.Warn("Replay commands are disabled until someone updates them and writes tests")
 	/* TODO:!

consensus/state.go

@@ -340,16 +340,9 @@ func (cs *ConsensusState) LoadCommit(height int) *types.Commit {
 }
 
 func (cs *ConsensusState) OnStart() error {
-	cs.BaseService.OnStart()
 
 	walFile := cs.config.GetString("cs_wal_file")
-	err := EnsureDir(path.Dir(walFile), 0700)
+	if err := cs.OpenWAL(walFile); err != nil {
-	if err != nil {
-		log.Error("Error ensuring ConsensusState wal dir", "error", err.Error())
-		return err
-	}
-	err = cs.OpenWAL(walFile)
-	if err != nil {
 		log.Error("Error loading ConsensusState wal", "error", err.Error())
 		return err
 	}
@@ -364,8 +357,9 @@ func (cs *ConsensusState) OnStart() error {
 	// we may have lost some votes if the process crashed
 	// reload from consensus log to catchup
 	if err := cs.catchupReplay(cs.Height); err != nil {
-		log.Error("Error on catchup replay", "error", err.Error())
+		log.Error("Error on catchup replay. Proceeding to start ConsensusState anyway", "error", err.Error())
-		// let's go for it anyways, maybe we're fine
+		// NOTE: if we ever do return an error here,
+		// make sure to stop the timeoutTicker
 	}
 
 	// now start the receiveRoutine
@@ -404,6 +398,12 @@ func (cs *ConsensusState) Wait() {
 
 // Open file to log all consensus messages and timeouts for deterministic accountability
 func (cs *ConsensusState) OpenWAL(walFile string) (err error) {
+	err = EnsureDir(path.Dir(walFile), 0700)
+	if err != nil {
+		log.Error("Error ensuring ConsensusState wal dir", "error", err.Error())
+		return err
+	}
+
 	cs.mtx.Lock()
 	defer cs.mtx.Unlock()
 	wal, err := NewWAL(walFile, cs.config.GetBool("cs_wal_light"))

consensus/ticker.go

@@ -45,7 +45,6 @@ func NewTimeoutTicker() TimeoutTicker {
 }
 
 func (t *timeoutTicker) OnStart() error {
-	t.BaseService.OnStart()
 
 	go t.timeoutRoutine()
 

consensus/wal.go

@@ -55,7 +55,6 @@ func NewWAL(walFile string, light bool) (*WAL, error) {
 }
 
 func (wal *WAL) OnStart() error {
-	wal.BaseService.OnStart()
 	size, err := wal.group.Head.Size()
 	if err != nil {
 		return err

node/node.go

@@ -194,7 +194,6 @@ func NewNode(config cfg.Config, privValidator *types.PrivValidator, clientCreato
 }
 
 func (n *Node) OnStart() error {
-	n.BaseService.OnStart()
 
 	// Create & add listener
 	protocol, address := ProtocolAndAddress(n.config.GetString("node_laddr"))

proxy/multi_app_conn.go

@@ -72,7 +72,6 @@ func (app *multiAppConn) Query() AppConnQuery {
 }
 
 func (app *multiAppConn) OnStart() error {
-	app.BaseService.OnStart()
 
 	// query connection
 	querycli, err := app.clientCreator.NewABCIClient()