You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
723 lines
25 KiB
723 lines
25 KiB
package node
|
|
|
|
import (
|
|
"bytes"
|
|
"context"
|
|
"encoding/gob"
|
|
"encoding/hex"
|
|
"fmt"
|
|
"io/ioutil"
|
|
"math"
|
|
"os"
|
|
"os/exec"
|
|
"strconv"
|
|
"sync/atomic"
|
|
"syscall"
|
|
"time"
|
|
|
|
"github.com/ethereum/go-ethereum/crypto"
|
|
|
|
"github.com/harmony-one/harmony/core"
|
|
|
|
"github.com/ethereum/go-ethereum/rlp"
|
|
pb "github.com/golang/protobuf/proto"
|
|
"github.com/harmony-one/bls/ffi/go/bls"
|
|
|
|
"github.com/harmony-one/harmony/api/proto"
|
|
proto_discovery "github.com/harmony-one/harmony/api/proto/discovery"
|
|
"github.com/harmony-one/harmony/api/proto/message"
|
|
proto_node "github.com/harmony-one/harmony/api/proto/node"
|
|
"github.com/harmony-one/harmony/api/service"
|
|
"github.com/harmony-one/harmony/core/types"
|
|
"github.com/harmony-one/harmony/crypto/pki"
|
|
nodeconfig "github.com/harmony-one/harmony/internal/configs/node"
|
|
"github.com/harmony-one/harmony/internal/utils"
|
|
"github.com/harmony-one/harmony/p2p"
|
|
"github.com/harmony-one/harmony/p2p/host"
|
|
)
|
|
|
|
const (
|
|
// MaxNumberOfTransactionsPerBlock is the max number of transaction per a block.
|
|
MaxNumberOfTransactionsPerBlock = 8000
|
|
consensusTimeout = 10 * time.Second
|
|
)
|
|
|
|
// ReceiveGlobalMessage use libp2p pubsub mechanism to receive global broadcast messages
|
|
func (node *Node) ReceiveGlobalMessage() {
|
|
ctx := context.Background()
|
|
for {
|
|
if node.globalGroupReceiver == nil {
|
|
time.Sleep(100 * time.Millisecond)
|
|
continue
|
|
}
|
|
msg, sender, err := node.globalGroupReceiver.Receive(ctx)
|
|
if sender != node.host.GetID() {
|
|
//utils.GetLogInstance().Info("[PUBSUB]", "received global msg", len(msg), "sender", sender)
|
|
if err == nil {
|
|
// skip the first 5 bytes, 1 byte is p2p type, 4 bytes are message size
|
|
go node.messageHandler(msg[5:], string(sender))
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// ReceiveGroupMessage use libp2p pubsub mechanism to receive broadcast messages
|
|
func (node *Node) ReceiveGroupMessage() {
|
|
ctx := context.Background()
|
|
for {
|
|
if node.shardGroupReceiver == nil {
|
|
time.Sleep(100 * time.Millisecond)
|
|
continue
|
|
}
|
|
msg, sender, err := node.shardGroupReceiver.Receive(ctx)
|
|
if sender != node.host.GetID() {
|
|
//utils.GetLogInstance().Info("[PUBSUB]", "received group msg", len(msg), "sender", sender)
|
|
if err == nil {
|
|
// skip the first 5 bytes, 1 byte is p2p type, 4 bytes are message size
|
|
go node.messageHandler(msg[5:], string(sender))
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// ReceiveClientGroupMessage use libp2p pubsub mechanism to receive broadcast messages for client
|
|
func (node *Node) ReceiveClientGroupMessage() {
|
|
ctx := context.Background()
|
|
for {
|
|
if node.clientReceiver == nil {
|
|
// check less frequent on client messages
|
|
time.Sleep(100 * time.Millisecond)
|
|
continue
|
|
}
|
|
msg, sender, err := node.clientReceiver.Receive(ctx)
|
|
if sender != node.host.GetID() {
|
|
// utils.GetLogInstance().Info("[CLIENT]", "received group msg", len(msg), "sender", sender, "error", err)
|
|
if err == nil {
|
|
// skip the first 5 bytes, 1 byte is p2p type, 4 bytes are message size
|
|
go node.messageHandler(msg[5:], string(sender))
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// messageHandler parses the message and dispatch the actions
|
|
func (node *Node) messageHandler(content []byte, sender string) {
|
|
msgCategory, err := proto.GetMessageCategory(content)
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("Read node type failed", "err", err, "node", node)
|
|
return
|
|
}
|
|
|
|
msgType, err := proto.GetMessageType(content)
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("Read action type failed", "err", err, "node", node)
|
|
return
|
|
}
|
|
|
|
msgPayload, err := proto.GetMessagePayload(content)
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("Read message payload failed", "err", err, "node", node)
|
|
return
|
|
}
|
|
|
|
switch msgCategory {
|
|
case proto.Consensus:
|
|
msgPayload, _ := proto.GetConsensusMessagePayload(content)
|
|
node.ConsensusMessageHandler(msgPayload)
|
|
case proto.DRand:
|
|
msgPayload, _ := proto.GetDRandMessagePayload(content)
|
|
if node.DRand != nil {
|
|
if node.DRand.IsLeader {
|
|
node.DRand.ProcessMessageLeader(msgPayload)
|
|
} else {
|
|
node.DRand.ProcessMessageValidator(msgPayload)
|
|
}
|
|
}
|
|
case proto.Staking:
|
|
utils.GetLogInstance().Info("NET: Received staking message")
|
|
msgPayload, _ := proto.GetStakingMessagePayload(content)
|
|
// Only beacon leader processes staking txn
|
|
if node.NodeConfig.Role() != nodeconfig.BeaconLeader {
|
|
return
|
|
}
|
|
node.processStakingMessage(msgPayload)
|
|
case proto.Node:
|
|
actionType := proto_node.MessageType(msgType)
|
|
switch actionType {
|
|
case proto_node.Transaction:
|
|
utils.GetLogInstance().Info("NET: received message: Node/Transaction")
|
|
node.transactionMessageHandler(msgPayload)
|
|
case proto_node.Block:
|
|
utils.GetLogInstance().Info("NET: received message: Node/Block")
|
|
blockMsgType := proto_node.BlockMessageType(msgPayload[0])
|
|
switch blockMsgType {
|
|
case proto_node.Sync:
|
|
utils.GetLogInstance().Info("NET: received message: Node/Sync")
|
|
var blocks []*types.Block
|
|
err := rlp.DecodeBytes(msgPayload[1:], &blocks)
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("block sync", "error", err)
|
|
} else {
|
|
// for non-beaconchain node, subscribe to beacon block broadcast
|
|
role := node.NodeConfig.Role()
|
|
if proto_node.BlockMessageType(msgPayload[0]) == proto_node.Sync && (role == nodeconfig.ShardValidator || role == nodeconfig.ShardLeader || role == nodeconfig.NewNode) {
|
|
utils.GetLogInstance().Info("Block being handled by block channel", "self peer", node.SelfPeer, "block", blocks[0].NumberU64())
|
|
for _, block := range blocks {
|
|
node.BeaconBlockChannel <- block
|
|
}
|
|
}
|
|
if node.Client != nil && node.Client.UpdateBlocks != nil && blocks != nil {
|
|
utils.GetLogInstance().Info("Block being handled by client by", "self peer", node.SelfPeer)
|
|
node.Client.UpdateBlocks(blocks)
|
|
}
|
|
}
|
|
}
|
|
case proto_node.PING:
|
|
node.pingMessageHandler(msgPayload, sender)
|
|
case proto_node.PONG:
|
|
node.pongMessageHandler(msgPayload)
|
|
case proto_node.ShardState:
|
|
node.epochShardStateMessageHandler(msgPayload)
|
|
}
|
|
default:
|
|
utils.GetLogInstance().Error("Unknown", "MsgCategory", msgCategory)
|
|
}
|
|
}
|
|
|
|
func (node *Node) processStakingMessage(msgPayload []byte) {
|
|
msg := &message.Message{}
|
|
err := pb.Unmarshal(msgPayload, msg)
|
|
if err == nil {
|
|
stakingRequest := msg.GetStaking()
|
|
txs := types.Transactions{}
|
|
if err = rlp.DecodeBytes(stakingRequest.Transaction, &txs); err == nil {
|
|
utils.GetLogInstance().Info("Successfully added staking transaction to pending list.")
|
|
node.addPendingTransactions(txs)
|
|
} else {
|
|
utils.GetLogInstance().Error("Failed to unmarshal staking transaction list", "error", err)
|
|
}
|
|
} else {
|
|
utils.GetLogInstance().Error("Failed to unmarshal staking msg payload", "error", err)
|
|
}
|
|
}
|
|
|
|
func (node *Node) transactionMessageHandler(msgPayload []byte) {
|
|
txMessageType := proto_node.TransactionMessageType(msgPayload[0])
|
|
|
|
switch txMessageType {
|
|
case proto_node.Send:
|
|
txs := types.Transactions{}
|
|
err := rlp.Decode(bytes.NewReader(msgPayload[1:]), &txs) // skip the Send messge type
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("Failed to deserialize transaction list", "error", err)
|
|
}
|
|
node.addPendingTransactions(txs)
|
|
|
|
case proto_node.Request:
|
|
reader := bytes.NewBuffer(msgPayload[1:])
|
|
txIDs := make(map[[32]byte]bool)
|
|
buf := make([]byte, 32) // 32 byte hash Id
|
|
for {
|
|
_, err := reader.Read(buf)
|
|
if err != nil {
|
|
break
|
|
}
|
|
|
|
var txID [32]byte
|
|
copy(txID[:], buf)
|
|
txIDs[txID] = true
|
|
}
|
|
|
|
var txToReturn []*types.Transaction
|
|
for _, tx := range node.pendingTransactions {
|
|
if txIDs[tx.Hash()] {
|
|
txToReturn = append(txToReturn, tx)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// BroadcastNewBlock is called by consensus leader to sync new blocks with other clients/nodes.
|
|
// NOTE: For now, just send to the client (basically not broadcasting)
|
|
// TODO (lc): broadcast the new blocks to new nodes doing state sync
|
|
func (node *Node) BroadcastNewBlock(newBlock *types.Block) {
|
|
if node.ClientPeer != nil {
|
|
utils.GetLogInstance().Debug("Sending new block to client", "client", node.ClientPeer)
|
|
node.host.SendMessageToGroups([]p2p.GroupID{node.NodeConfig.GetClientGroupID()}, host.ConstructP2pMessage(byte(0), proto_node.ConstructBlocksSyncMessage([]*types.Block{newBlock})))
|
|
}
|
|
}
|
|
|
|
// VerifyNewBlock is called by consensus participants to verify the block (account model) they are running consensus on
|
|
func (node *Node) VerifyNewBlock(newBlock *types.Block) bool {
|
|
err := node.blockchain.ValidateNewBlock(newBlock, pki.GetAddressFromPublicKey(node.SelfPeer.ConsensusPubKey))
|
|
if err != nil {
|
|
utils.GetLogInstance().Debug("Failed verifying new block", "Error", err, "tx", newBlock.Transactions()[0])
|
|
return false
|
|
}
|
|
|
|
// TODO: verify the vrf randomness
|
|
_ = newBlock.Header().RandPreimage
|
|
|
|
err = node.blockchain.ValidateNewShardState(newBlock, &node.CurrentStakes)
|
|
if err != nil {
|
|
utils.GetLogInstance().Debug("Failed to verify new sharding state", "err", err)
|
|
}
|
|
return true
|
|
}
|
|
|
|
// PostConsensusProcessing is called by consensus participants, after consensus is done, to:
|
|
// 1. add the new block to blockchain
|
|
// 2. [leader] send new block to the client
|
|
func (node *Node) PostConsensusProcessing(newBlock *types.Block) {
|
|
if node.Consensus.PubKey.IsEqual(node.Consensus.LeaderPubKey) {
|
|
node.BroadcastNewBlock(newBlock)
|
|
} else {
|
|
utils.GetLogInstance().Info("BINGO !!! Reached Consensus", "ViewID", node.Consensus.GetViewID())
|
|
}
|
|
|
|
node.AddNewBlock(newBlock)
|
|
|
|
// Update contract deployer's nonce so default contract like faucet can issue transaction with current nonce
|
|
nonce := node.GetNonceOfAddress(crypto.PubkeyToAddress(node.ContractDeployerKey.PublicKey))
|
|
atomic.StoreUint64(&node.ContractDeployerCurrentNonce, nonce)
|
|
|
|
for _, tx := range newBlock.Transactions() {
|
|
msg, err := tx.AsMessage(types.HomesteadSigner{})
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("Error when parsing tx into message")
|
|
}
|
|
if _, ok := node.AddressNonce.Load(msg.From()); ok {
|
|
nonce := node.GetNonceOfAddress(msg.From())
|
|
node.AddressNonce.Store(msg.From(), nonce)
|
|
}
|
|
}
|
|
|
|
if node.Consensus.ShardID == 0 {
|
|
// Update contract deployer's nonce so default contract like faucet can issue transaction with current nonce
|
|
nonce := node.GetNonceOfAddress(crypto.PubkeyToAddress(node.ContractDeployerKey.PublicKey))
|
|
atomic.StoreUint64(&node.ContractDeployerCurrentNonce, nonce)
|
|
|
|
// TODO: enable drand only for beacon chain
|
|
// ConfirmedBlockChannel which is listened by drand leader who will initiate DRG if its a epoch block (first block of a epoch)
|
|
if node.DRand != nil {
|
|
go func() {
|
|
node.ConfirmedBlockChannel <- newBlock
|
|
}()
|
|
}
|
|
|
|
// ConfirmedBlockChannel which is listened by drand leader who will initiate DRG if its a epoch block (first block of a epoch)
|
|
if node.DRand != nil {
|
|
go func() {
|
|
node.ConfirmedBlockChannel <- newBlock
|
|
}()
|
|
}
|
|
|
|
// TODO: update staking information once per epoch.
|
|
node.UpdateStakingList(node.QueryStakeInfo())
|
|
node.printStakingList()
|
|
if core.IsEpochBlock(newBlock) {
|
|
shardState := node.blockchain.StoreNewShardState(newBlock, &node.CurrentStakes)
|
|
if shardState != nil {
|
|
if nodeconfig.GetDefaultConfig().IsLeader() {
|
|
epochShardState := types.EpochShardState{Epoch: core.GetEpochFromBlockNumber(newBlock.NumberU64()), ShardState: shardState}
|
|
epochShardStateMessage := proto_node.ConstructEpochShardStateMessage(epochShardState)
|
|
// Broadcast new shard state
|
|
err := node.host.SendMessageToGroups([]p2p.GroupID{node.NodeConfig.GetClientGroupID()}, host.ConstructP2pMessage(byte(0), epochShardStateMessage))
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("[Resharding] failed to broadcast shard state message", "group", node.NodeConfig.GetClientGroupID())
|
|
} else {
|
|
utils.GetLogInstance().Info("[Resharding] broadcasted shard state message to", "group", node.NodeConfig.GetClientGroupID())
|
|
}
|
|
}
|
|
node.processEpochShardState(&types.EpochShardState{Epoch: core.GetEpochFromBlockNumber(newBlock.NumberU64()), ShardState: shardState})
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// AddNewBlock is usedd to add new block into the blockchain.
|
|
func (node *Node) AddNewBlock(newBlock *types.Block) {
|
|
blockNum, err := node.blockchain.InsertChain([]*types.Block{newBlock})
|
|
if err != nil {
|
|
utils.GetLogInstance().Debug("Error adding new block to blockchain", "blockNum", blockNum, "hash", newBlock.Header().Hash(), "Error", err)
|
|
} else {
|
|
utils.GetLogInstance().Info("adding new block to blockchain", "blockNum", blockNum, "hash", newBlock.Header().Hash(), "by node", node.SelfPeer)
|
|
}
|
|
}
|
|
|
|
func (node *Node) pingMessageHandler(msgPayload []byte, sender string) int {
|
|
if sender != "" {
|
|
_, ok := node.duplicatedPing.LoadOrStore(sender, true)
|
|
if ok {
|
|
// duplicated ping message return
|
|
return 0
|
|
}
|
|
}
|
|
|
|
ping, err := proto_discovery.GetPingMessage(msgPayload)
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("Can't get Ping Message", "error", err)
|
|
return -1
|
|
}
|
|
|
|
peer := new(p2p.Peer)
|
|
peer.IP = ping.Node.IP
|
|
peer.Port = ping.Node.Port
|
|
peer.PeerID = ping.Node.PeerID
|
|
peer.ConsensusPubKey = nil
|
|
|
|
if ping.Node.PubKey != nil {
|
|
peer.ConsensusPubKey = &bls.PublicKey{}
|
|
if err := peer.ConsensusPubKey.Deserialize(ping.Node.PubKey[:]); err != nil {
|
|
utils.GetLogInstance().Error("UnmarshalBinary Failed", "error", err)
|
|
return -1
|
|
}
|
|
}
|
|
|
|
// utils.GetLogInstance().Debug("[pingMessageHandler]", "incoming peer", peer)
|
|
|
|
// add to incoming peer list
|
|
//node.host.AddIncomingPeer(*peer)
|
|
node.host.ConnectHostPeer(*peer)
|
|
|
|
if ping.Node.Role == proto_node.ClientRole {
|
|
utils.GetLogInstance().Info("Add Client Peer to Node", "Address", node.Consensus.GetSelfAddress(), "Client", peer)
|
|
node.ClientPeer = peer
|
|
} else {
|
|
node.AddPeers([]*p2p.Peer{peer})
|
|
utils.GetLogInstance().Info("Add Peer to Node", "Address", node.Consensus.GetSelfAddress(), "Peer", peer, "# Peers", len(node.Consensus.PublicKeys))
|
|
}
|
|
|
|
return 1
|
|
}
|
|
|
|
// SendPongMessage is the a goroutine to periodcally send pong message to all peers
|
|
func (node *Node) SendPongMessage() {
|
|
tick := time.NewTicker(2 * time.Second)
|
|
tick2 := time.NewTicker(120 * time.Second)
|
|
|
|
numPeers := len(node.Consensus.GetValidatorPeers())
|
|
numPubKeys := len(node.Consensus.PublicKeys)
|
|
sentMessage := false
|
|
firstTime := true
|
|
|
|
// Send Pong Message only when there is change on the number of peers
|
|
for {
|
|
select {
|
|
case <-tick.C:
|
|
peers := node.Consensus.GetValidatorPeers()
|
|
numPeersNow := len(peers)
|
|
numPubKeysNow := len(node.Consensus.PublicKeys)
|
|
|
|
// no peers, wait for another tick
|
|
if numPubKeysNow == 0 {
|
|
utils.GetLogInstance().Info("[PONG] no peers, continue", "numPeers", numPeers, "numPeersNow", numPeersNow)
|
|
continue
|
|
}
|
|
// new peers added
|
|
if numPubKeysNow != numPubKeys || numPeersNow != numPeers {
|
|
utils.GetLogInstance().Info("[PONG] different number of peers", "numPeers", numPeers, "numPeersNow", numPeersNow)
|
|
sentMessage = false
|
|
} else {
|
|
// stable number of peers/pubkeys, sent the pong message
|
|
// also make sure number of peers is greater than the minimal required number
|
|
if !sentMessage && numPubKeysNow >= node.Consensus.MinPeers {
|
|
pong := proto_discovery.NewPongMessage(peers, node.Consensus.PublicKeys, node.Consensus.GetLeaderPubKey(), node.Consensus.ShardID)
|
|
buffer := pong.ConstructPongMessage()
|
|
err := node.host.SendMessageToGroups([]p2p.GroupID{node.NodeConfig.GetShardGroupID()}, host.ConstructP2pMessage(byte(0), buffer))
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("[PONG] failed to send pong message", "group", node.NodeConfig.GetShardGroupID())
|
|
continue
|
|
} else {
|
|
utils.GetLogInstance().Info("[PONG] sent pong message to", "group", node.NodeConfig.GetShardGroupID(), "# nodes", numPeersNow)
|
|
}
|
|
sentMessage = true
|
|
// wait a bit until all validators received pong message
|
|
time.Sleep(200 * time.Millisecond)
|
|
|
|
// only need to notify consensus leader once to start the consensus
|
|
if firstTime {
|
|
// Leader stops sending ping message
|
|
time.Sleep(5 * time.Second)
|
|
node.serviceManager.TakeAction(&service.Action{Action: service.Stop, ServiceType: service.PeerDiscovery})
|
|
node.startConsensus <- struct{}{}
|
|
firstTime = false
|
|
}
|
|
}
|
|
}
|
|
numPeers = numPeersNow
|
|
numPubKeys = numPubKeysNow
|
|
case <-tick2.C:
|
|
// send pong message regularly to make sure new node received all the public keys
|
|
// also nodes offline/online will receive the public keys
|
|
peers := node.Consensus.GetValidatorPeers()
|
|
pong := proto_discovery.NewPongMessage(peers, node.Consensus.PublicKeys, node.Consensus.GetLeaderPubKey(), node.Consensus.ShardID)
|
|
buffer := pong.ConstructPongMessage()
|
|
err := node.host.SendMessageToGroups([]p2p.GroupID{node.NodeConfig.GetShardGroupID()}, host.ConstructP2pMessage(byte(0), buffer))
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("[PONG] failed to send regular pong message", "group", node.NodeConfig.GetShardGroupID())
|
|
continue
|
|
} else {
|
|
utils.GetLogInstance().Info("[PONG] sent regular pong message to", "group", node.NodeConfig.GetShardGroupID(), "# nodes", len(peers))
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
func (node *Node) pongMessageHandler(msgPayload []byte) int {
|
|
utils.GetLogInstance().Error("Got Pong Message")
|
|
pong, err := proto_discovery.GetPongMessage(msgPayload)
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("Can't get Pong Message", "error", err)
|
|
return -1
|
|
}
|
|
|
|
if pong.ShardID != node.Consensus.ShardID {
|
|
utils.GetLogInstance().Error("Received Pong message for the wrong shard", "receivedShardID", pong.ShardID)
|
|
return 0
|
|
}
|
|
|
|
// set the leader pub key is the first thing to do
|
|
// otherwise, we may not be able to validate the consensus messages received
|
|
// which will result in first consensus timeout
|
|
// TODO: remove this after fully migrating to beacon chain-based committee membership
|
|
//err = node.Consensus.SetLeaderPubKey(pong.LeaderPubKey)
|
|
//if err != nil {
|
|
// utils.GetLogInstance().Error("Unmarshal Consensus Leader PubKey Failed", "error", err)
|
|
//} else {
|
|
// utils.GetLogInstance().Info("Set Consensus Leader PubKey", "key", node.Consensus.GetLeaderPubKey())
|
|
//}
|
|
//err = node.DRand.SetLeaderPubKey(pong.LeaderPubKey)
|
|
//if err != nil {
|
|
// utils.GetLogInstance().Error("Unmarshal DRand Leader PubKey Failed", "error", err)
|
|
//} else {
|
|
// utils.GetLogInstance().Info("Set DRand Leader PubKey", "key", node.Consensus.GetLeaderPubKey())
|
|
//}
|
|
|
|
peers := make([]*p2p.Peer, 0)
|
|
|
|
for _, p := range pong.Peers {
|
|
peer := new(p2p.Peer)
|
|
peer.IP = p.IP
|
|
peer.Port = p.Port
|
|
peer.PeerID = p.PeerID
|
|
|
|
peer.ConsensusPubKey = &bls.PublicKey{}
|
|
err = peer.ConsensusPubKey.Deserialize(p.PubKey[:])
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("UnmarshalBinary Failed", "error", err)
|
|
continue
|
|
}
|
|
peers = append(peers, peer)
|
|
}
|
|
|
|
if len(peers) > 0 {
|
|
node.AddPeers(peers)
|
|
}
|
|
|
|
// Reset Validator PublicKeys every time we receive PONG message from Leader
|
|
// The PublicKeys has to be idential across the shard on every node
|
|
// TODO (lc): we need to handle RemovePeer situation
|
|
publicKeys := make([]*bls.PublicKey, 0)
|
|
|
|
// Create the the PubKey from the []byte sent from leader
|
|
for _, k := range pong.PubKeys {
|
|
key := bls.PublicKey{}
|
|
err = key.Deserialize(k[:])
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("UnmarshalBinary Failed PubKeys", "error", err)
|
|
continue
|
|
}
|
|
publicKeys = append(publicKeys, &key)
|
|
}
|
|
|
|
utils.GetLogInstance().Debug("[pongMessageHandler]", "#keys", len(publicKeys), "#peers", len(peers))
|
|
|
|
if node.State == NodeWaitToJoin {
|
|
node.State = NodeReadyForConsensus
|
|
}
|
|
|
|
// Stop discovery service after received pong message
|
|
data := make(map[string]interface{})
|
|
data["peer"] = p2p.GroupAction{Name: node.NodeConfig.GetShardGroupID(), Action: p2p.ActionPause}
|
|
|
|
node.serviceManager.TakeAction(&service.Action{Action: service.Notify, ServiceType: service.PeerDiscovery, Params: data})
|
|
|
|
// TODO: remove this after fully migrating to beacon chain-based committee membership
|
|
return 0
|
|
}
|
|
|
|
func (node *Node) epochShardStateMessageHandler(msgPayload []byte) int {
|
|
utils.GetLogInstance().Error("[Received new shard state]")
|
|
epochShardState, err := proto_node.DeserializeEpochShardStateFromMessage(msgPayload)
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("Can't get shard state Message", "error", err)
|
|
return -1
|
|
}
|
|
if (node.Consensus != nil && node.Consensus.ShardID != 0) || node.NodeConfig.Role() == nodeconfig.NewNode {
|
|
node.processEpochShardState(epochShardState)
|
|
}
|
|
return 0
|
|
}
|
|
|
|
func (node *Node) processEpochShardState(epochShardState *types.EpochShardState) {
|
|
shardState := epochShardState.ShardState
|
|
epoch := epochShardState.Epoch
|
|
|
|
for _, c := range shardState {
|
|
utils.GetLogInstance().Debug("new shard information", "shardID", c.ShardID, "NodeList", c.NodeList)
|
|
}
|
|
|
|
myShardID := uint32(math.MaxUint32)
|
|
isNextLeader := false
|
|
myBlsPubKey := node.Consensus.PubKey.Serialize()
|
|
myShardState := types.Committee{}
|
|
for _, shard := range shardState {
|
|
for _, nodeID := range shard.NodeList {
|
|
if bytes.Compare(nodeID.BlsPublicKey[:], myBlsPubKey) == 0 {
|
|
myShardID = shard.ShardID
|
|
isNextLeader = shard.Leader == nodeID
|
|
myShardState = shard
|
|
}
|
|
}
|
|
}
|
|
|
|
if myShardID != uint32(math.MaxUint32) {
|
|
// Update public keys
|
|
ss := myShardState
|
|
publicKeys := []*bls.PublicKey{}
|
|
for _, nodeID := range ss.NodeList {
|
|
key := &bls.PublicKey{}
|
|
err := key.Deserialize(nodeID.BlsPublicKey[:])
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("Failed to deserialize BLS public key in shard state", "error", err)
|
|
}
|
|
publicKeys = append(publicKeys, key)
|
|
}
|
|
node.Consensus.UpdatePublicKeys(publicKeys)
|
|
node.DRand.UpdatePublicKeys(publicKeys)
|
|
|
|
aboutLeader := ""
|
|
if nodeconfig.GetDefaultConfig().IsLeader() {
|
|
aboutLeader = "I am not leader anymore"
|
|
if isNextLeader {
|
|
aboutLeader = "I am still leader"
|
|
}
|
|
} else {
|
|
aboutLeader = "I am still validator"
|
|
if isNextLeader {
|
|
aboutLeader = "I become the leader"
|
|
}
|
|
}
|
|
if node.blockchain.ShardID() == myShardID {
|
|
utils.GetLogInstance().Info(fmt.Sprintf("[Resharded][epoch:%d] I stay at shard %d, %s", epoch, myShardID, aboutLeader), "BlsPubKey", hex.EncodeToString(myBlsPubKey))
|
|
} else {
|
|
utils.GetLogInstance().Info(fmt.Sprintf("[Resharded][epoch:%d] I got resharded to shard %d from shard %d, %s", epoch, myShardID, node.blockchain.ShardID(), aboutLeader), "BlsPubKey", hex.EncodeToString(myBlsPubKey))
|
|
node.storeEpochShardState(epochShardState)
|
|
|
|
execFile, err := getBinaryPath()
|
|
if err != nil {
|
|
utils.GetLogInstance().Crit("Failed to get program path when restarting program", "error", err, "file", execFile)
|
|
}
|
|
args := getRestartArguments(myShardID)
|
|
utils.GetLogInstance().Info("Restarting program", "args", args, "env", os.Environ())
|
|
err = syscall.Exec(execFile, args, os.Environ())
|
|
if err != nil {
|
|
utils.GetLogInstance().Crit("Failed to restart program after resharding", "error", err)
|
|
}
|
|
}
|
|
} else {
|
|
utils.GetLogInstance().Info(fmt.Sprintf("[Resharded][epoch:%d] Somehow I got kicked out. Exiting", epoch), "BlsPubKey", hex.EncodeToString(myBlsPubKey))
|
|
os.Exit(8) // 8 represents it's a loop and the program restart itself
|
|
}
|
|
}
|
|
|
|
func getRestartArguments(myShardID uint32) []string {
|
|
args := os.Args
|
|
hasShardID := false
|
|
shardIDFlag := "-shard_id"
|
|
// newNodeFlag := "-is_newnode"
|
|
for i, arg := range args {
|
|
if arg == shardIDFlag {
|
|
if i+1 < len(args) {
|
|
args[i+1] = strconv.Itoa(int(myShardID))
|
|
} else {
|
|
args = append(args, strconv.Itoa(int(myShardID)))
|
|
}
|
|
hasShardID = true
|
|
}
|
|
// TODO: enable this
|
|
//if arg == newNodeFlag {
|
|
// args[i] = "" // remove new node flag
|
|
//}
|
|
}
|
|
if !hasShardID {
|
|
args = append(args, shardIDFlag)
|
|
args = append(args, strconv.Itoa(int(myShardID)))
|
|
}
|
|
return args
|
|
}
|
|
|
|
// Gets the path of this currently running binary program.
|
|
func getBinaryPath() (argv0 string, err error) {
|
|
argv0, err = exec.LookPath(os.Args[0])
|
|
if nil != err {
|
|
return
|
|
}
|
|
if _, err = os.Stat(argv0); nil != err {
|
|
return
|
|
}
|
|
return
|
|
}
|
|
|
|
// Stores the epoch shard state into local file
|
|
// TODO: think about storing it into level db.
|
|
func (node *Node) storeEpochShardState(epochShardState *types.EpochShardState) {
|
|
byteBuffer := bytes.NewBuffer([]byte{})
|
|
encoder := gob.NewEncoder(byteBuffer)
|
|
err := encoder.Encode(epochShardState)
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("[Resharded] Failed to encode epoch shard state", "error", err)
|
|
}
|
|
err = ioutil.WriteFile("./epoch_shard_state"+node.SelfPeer.IP+node.SelfPeer.Port, byteBuffer.Bytes(), 0644)
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("[Resharded] Failed to store epoch shard state in local file", "error", err)
|
|
}
|
|
}
|
|
|
|
func (node *Node) retrieveEpochShardState() (*types.EpochShardState, error) {
|
|
b, err := ioutil.ReadFile("./epoch_shard_state" + node.SelfPeer.IP + node.SelfPeer.Port)
|
|
if err != nil {
|
|
utils.GetLogInstance().Error("[Resharded] Failed to retrieve epoch shard state", "error", err)
|
|
}
|
|
epochShardState := new(types.EpochShardState)
|
|
|
|
r := bytes.NewBuffer(b)
|
|
decoder := gob.NewDecoder(r)
|
|
err = decoder.Decode(epochShardState)
|
|
|
|
if err != nil {
|
|
return nil, fmt.Errorf("Decode local epoch shard state error")
|
|
}
|
|
return epochShardState, nil
|
|
}
|
|
|
|
// ConsensusMessageHandler passes received message in node_handler to consensus
|
|
func (node *Node) ConsensusMessageHandler(msgPayload []byte) {
|
|
if node.Consensus.ConsensusVersion == "v1" {
|
|
if nodeconfig.GetDefaultConfig().IsLeader() {
|
|
node.Consensus.ProcessMessageLeader(msgPayload)
|
|
} else {
|
|
node.Consensus.ProcessMessageValidator(msgPayload)
|
|
}
|
|
return
|
|
}
|
|
if node.Consensus.ConsensusVersion == "v2" {
|
|
select {
|
|
case node.Consensus.MsgChan <- msgPayload:
|
|
case <-time.After(consensusTimeout):
|
|
utils.GetLogInstance().Debug("[Consensus] ConsensusMessageHandler timeout", "duration", consensusTimeout, "msgPayload", len(msgPayload))
|
|
}
|
|
return
|
|
}
|
|
}
|
|
|