refactor: improve errors & fix blame panic (#53)

* Fix error messages

* Fix error messages [2]

* Fix error messages [3]

* Move back to github.com/pkg/errors

* Fix error messages [4]

* Minor StreamMgr improvements

* Minor improvement

* Fix typo

* Fix panic in blame logic

* Fix code style (address PR comments)

* Format Keygen code (pr comments)

* Rename file

Author: swift1337

blame/blame.go

@@ -3,8 +3,6 @@ package blame
 import (
 	"bytes"
 	"fmt"
-	"strconv"
-	"strings"
 	"sync"
 )
 
@@ -32,13 +30,14 @@ func NewBlame(reason string, blameNodes []Node) Blame {
 	}
 }
 
-// String implement fmt.Stringer
 func (b Blame) String() string {
-	sb := strings.Builder{}
-	sb.WriteString("reason:" + b.FailReason + " is_unicast:" + strconv.FormatBool(b.IsUnicast) + "\n")
-	sb.WriteString(fmt.Sprintf("round:%s\n", b.Round))
-	sb.WriteString(fmt.Sprintf("nodes:%+v\n", b.BlameNodes))
-	return sb.String()
+	return fmt.Sprintf(
+		"{reason=%q, is_unicast=%t, round:%s nodes:%+v}",
+		b.FailReason,
+		b.IsUnicast,
+		b.Round,
+		b.BlameNodes,
+	)
 }
 
 // SetBlame update the field values of Blame

blame/policy.go

@@ -1,33 +1,32 @@
 package blame
 
 import (
-	"errors"
-	"fmt"
-
 	btss "github.com/bnb-chain/tss-lib/tss"
 	mapset "github.com/deckarep/golang-set"
 	"github.com/libp2p/go-libp2p/core/peer"
+	"github.com/pkg/errors"
 
 	"github.com/zeta-chain/go-tss/conversion"
 	"github.com/zeta-chain/go-tss/messages"
 )
 
 func (m *Manager) tssTimeoutBlame(lastMessageType string, partyIDMap map[string]*btss.PartyID) ([]string, error) {
+	standbyNodes := m.roundMgr.GetByRound(lastMessageType)
+	if len(standbyNodes) == 0 {
+		return nil, nil
+	}
+
+	standbySet := mapset.NewSet()
+	for _, v := range standbyNodes {
+		standbySet.Add(v)
+	}
+
 	peersSet := mapset.NewSet()
 	for _, el := range partyIDMap {
 		if el.Id != m.localPartyID {
 			peersSet.Add(el.Id)
 		}
 	}
-	standbyNodes := m.roundMgr.GetByRound(lastMessageType)
-	if len(standbyNodes) == 0 {
-		return nil, nil
-	}
-	s := make([]interface{}, len(standbyNodes))
-	for i, v := range standbyNodes {
-		s[i] = v
-	}
-	standbySet := mapset.NewSetFromSlice(s)
 
 	var blames []string
 	diff := peersSet.Difference(standbySet).ToSlice()
@@ -37,8 +36,7 @@ func (m *Manager) tssTimeoutBlame(lastMessageType string, partyIDMap map[string]
 
 	blamePubKeys, err := conversion.AccPubKeysFromPartyIDs(blames, m.partyInfo.PartyIDMap)
 	if err != nil {
-		m.logger.Error().Err(err).Msg("fail to get the public keys of the blame node")
-		return nil, err
+		return nil, errors.Wrap(err, "unable to derive blame public keys")
 	}
 
 	return blamePubKeys, nil
@@ -47,39 +45,47 @@ func (m *Manager) tssTimeoutBlame(lastMessageType string, partyIDMap map[string]
 // this blame blames the node who cause the timeout in node sync
 func (m *Manager) NodeSyncBlame(keys []string, onlinePeers []peer.ID) (Blame, error) {
 	blame := NewBlame(TssSyncFail, nil)
+
 	for _, item := range keys {
 		found := false
 		peerID, err := conversion.GetPeerIDFromPubKey(item)
 		if err != nil {
-			return blame, fmt.Errorf("fail to get peer id from pub key")
+			return blame, errors.Wrap(err, "unable to get peer id from pub key")
 		}
+
 		for _, p := range onlinePeers {
 			if p == peerID {
 				found = true
 				break
 			}
 		}
+
 		if !found {
 			blame.BlameNodes = append(blame.BlameNodes, NewNode(item, nil, nil))
 		}
 	}
+
 	return blame, nil
 }
 
 // this blame blames the node who cause the timeout in unicast message
 func (m *Manager) GetUnicastBlame(lastMsgType string) ([]Node, error) {
 	m.lastMsgLocker.RLock()
+
 	if len(m.lastUnicastPeer) == 0 {
 		m.lastMsgLocker.RUnlock()
 		m.logger.Debug().Msg("we do not have any unicast message received yet")
 		return nil, nil
 	}
+
 	peersMap := make(map[string]bool)
 	peersID, ok := m.lastUnicastPeer[lastMsgType]
 	m.lastMsgLocker.RUnlock()
+
 	if !ok {
-		return nil, fmt.Errorf("fail to find peers of the given msg type %w", ErrTimeoutTSS)
+		return nil, errors.Wrap(ErrTimeoutTSS, "fail to find peers of the given msg type")
 	}
+
 	for _, el := range peersID {
 		peersMap[el.String()] = true
 	}
@@ -88,55 +94,62 @@ func (m *Manager) GetUnicastBlame(lastMsgType string) ([]Node, error) {
 	for key := range peersMap {
 		onlinePeers = append(onlinePeers, key)
 	}
+
 	_, blamePeers, err := m.GetBlamePubKeysLists(onlinePeers)
 	if err != nil {
-		m.logger.Error().Err(err).Msg("fail to get the blamed peers")
-		return nil, fmt.Errorf("fail to get the blamed peers %w", ErrTimeoutTSS)
+		return nil, errors.Wrap(err, "unable to get the blamed peers")
 	}
+
 	var blameNodes []Node
 	for _, el := range blamePeers {
 		blameNodes = append(blameNodes, NewNode(el, nil, nil))
 	}
+
 	return blameNodes, nil
 }
 
 // this blame blames the node who cause the timeout in broadcast message
 func (m *Manager) GetBroadcastBlame(lastMessageType string) ([]Node, error) {
 	blamePeers, err := m.tssTimeoutBlame(lastMessageType, m.partyInfo.PartyIDMap)
 	if err != nil {
-		m.logger.Error().Err(err).Msg("fail to get the blamed peers")
-		return nil, fmt.Errorf("fail to get the blamed peers %w", ErrTimeoutTSS)
+		return nil, errors.Wrap(err, "tssTimeoutBlame")
 	}
+
 	var blameNodes []Node
 	for _, el := range blamePeers {
 		blameNodes = append(blameNodes, NewNode(el, nil, nil))
 	}
+
 	return blameNodes, nil
 }
 
-// this blame blames the node who provide the wrong share
-func (m *Manager) TssWrongShareBlame(wiredMsg *messages.WireMessage) (string, error) {
+// TSSWrongShareBlame blames the node who provide the wrong share
+func (m *Manager) TSSWrongShareBlame(wiredMsg *messages.WireMessage) (string, error) {
 	shareOwner := wiredMsg.Routing.From
 	owner, ok := m.partyInfo.PartyIDMap[shareOwner.Id]
 	if !ok {
-		m.logger.Error().Msg("cannot find the blame node public key")
-		return "", errors.New("fail to find the share Owner")
+		return "", errors.New("unable to find the share owner")
 	}
+
 	pk, err := conversion.PartyIDtoPubKey(owner)
 	if err != nil {
-		return "", err
+		return "", errors.Wrap(err, "unable to convert party id to pub key")
 	}
+
 	return pk, nil
 }
 
-// this blame blames the node fail to send the shares to the node
+// TSSMissingShareBlame blames the node fail to send the shares to the node
 // with batch signing, we need to put the accepted shares into different message group
-// then search the missing share for each keysign message
-func (m *Manager) TssMissingShareBlame(rounds int, algo messages.Algo) ([]Node, bool, error) {
-	acceptedShareForMsg := make(map[string][][]string)
-	var blameNodes []Node
-	var peers []string
-	isUnicast := false
+// then search the missing share for each keysign message.
+func (m *Manager) TSSMissingShareBlame(rounds int, algo messages.Algo) ([]Node, bool, error) {
+	var (
+		acceptedShareForMsg = make(map[string][][]string)
+		blameNodes          []Node
+		peers               []string
+		isUnicast           bool
+	)
+
 	m.acceptShareLocker.Lock()
 	for roundInfo, value := range m.acceptedShares {
 		cachedShares, ok := acceptedShareForMsg[roundInfo.MsgIdentifier]
@@ -146,6 +159,19 @@ func (m *Manager) TssMissingShareBlame(rounds int, algo messages.Algo) ([]Node,
 			acceptedShareForMsg[roundInfo.MsgIdentifier] = cachedShares
 			continue
 		}
+
+		// should not happen
+		if roundInfo.Index >= len(cachedShares) {
+			m.logger.Error().
+				Int("round_index", roundInfo.Index).
+				Int("cached_shares_len", len(cachedShares)).
+				Int("rounds", rounds).
+				Int("algo", int(algo)).
+				Msg("Unexpected round index")
+
+			continue
+		}
+
 		cachedShares[roundInfo.Index] = value
 	}
 	m.acceptShareLocker.Unlock()
@@ -198,8 +224,9 @@ func (m *Manager) TssMissingShareBlame(rounds int, algo messages.Algo) ([]Node,
 		}
 		blamePubKeys, err := m.getBlamePubKeysNotInList(peers)
 		if err != nil {
-			return nil, isUnicast, err
+			return nil, isUnicast, errors.Wrap(err, "getBlamePubKeysNotInList")
 		}
+
 		for _, el := range blamePubKeys {
 			node := Node{
 				el,

blame/policy_test.go

@@ -121,7 +121,7 @@ func (p *policyTestSuite) TestTssWrongShareBlame(c *C) {
 		RoundInfo: "key2",
 		Message:   nil,
 	}
-	target, err := p.blameMgr.TssWrongShareBlame(&msg)
+	target, err := p.blameMgr.TSSWrongShareBlame(&msg)
 	c.Assert(err, IsNil)
 	c.Assert(target, Equals, "thorpub1addwnpepqfjcw5l4ay5t00c32mmlky7qrppepxzdlkcwfs2fd5u73qrwna0vzag3y4j")
 }
@@ -136,14 +136,14 @@ func (p *policyTestSuite) TestTssMissingShareBlame(c *C) {
 	acceptedShares[RoundInfo{0, "testRound", "123:0"}] = []string{"1", "2"}
 	acceptedShares[RoundInfo{1, "testRound", "123:0"}] = []string{"1"}
 	blameMgr.acceptShareLocker.Unlock()
-	nodes, _, err := blameMgr.TssMissingShareBlame(2, messages.ECDSAKEYGEN)
+	nodes, _, err := blameMgr.TSSMissingShareBlame(2, messages.ECDSAKEYGEN)
 	c.Assert(err, IsNil)
 	c.Assert(nodes[0].Pubkey, Equals, localTestPubKeys[3])
 	// we test if the missing share happens in round2
 	blameMgr.acceptShareLocker.Lock()
 	acceptedShares[RoundInfo{0, "testRound", "123:0"}] = []string{"1", "2", "3"}
 	blameMgr.acceptShareLocker.Unlock()
-	nodes, _, err = blameMgr.TssMissingShareBlame(2, messages.ECDSAKEYGEN)
+	nodes, _, err = blameMgr.TSSMissingShareBlame(2, messages.ECDSAKEYGEN)
 	c.Assert(err, IsNil)
 	results := []string{nodes[0].Pubkey, nodes[1].Pubkey}
 	sort.Strings(results)

blame/types.go

@@ -1,10 +1,10 @@
 package blame
 
 import (
-	"errors"
 	"sync"
 
 	btss "github.com/bnb-chain/tss-lib/tss"
+	"github.com/pkg/errors"
 )
 
 const (

cmd/tss-benchgen/main.go

@@ -165,7 +165,7 @@ outer:
 				}
 			} else { // point-to-point!
 				if dest[0].Index == msg.GetFrom().Index {
-					panic(fmt.Errorf("party %d tried to send a message to itself (%d)", dest[0].Index, msg.GetFrom().Index))
+					panic(errors.Errorf("party %d tried to send a message to itself (%d)", dest[0].Index, msg.GetFrom().Index))
 				}
 				go updater(parties[dest[0].Index], msg, errCh)
 			}

cmd/tss-benchsign/main.go

@@ -118,8 +118,9 @@ func runSign(dir string, t int) {
 	if err != nil {
 		panic(err)
 	}
+
 	if len(keys) != q || len(signPIDs) != q {
-		panic(fmt.Errorf("wanted %d keys but got %d keys and %d signPIDs", q, len(keys), len(signPIDs)))
+		panic(errors.Errorf("wanted %d keys but got %d keys and %d signPIDs", q, len(keys), len(signPIDs)))
 	}
 
 	msg := common.GetRandomPrimeInt(256)
@@ -163,7 +164,7 @@ outer:
 				}
 			} else {
 				if dest[0].Index == msg.GetFrom().Index {
-					panic(fmt.Errorf("party %d tried to send a message to itself (%d)", dest[0].Index, msg.GetFrom().Index))
+					panic(errors.Errorf("party %d tried to send a message to itself (%d)", dest[0].Index, msg.GetFrom().Index))
 				}
 				go updater(parties[dest[0].Index], msg, errCh)
 			}

cmd/tss-recovery/tool.go

@@ -5,9 +5,6 @@ import (
 	"crypto/cipher"
 	"crypto/rand"
 	"encoding/json"
-	"errors"
-	"fmt"
-	"io/ioutil"
 	"math/big"
 	"os"
 
@@ -16,13 +13,14 @@ import (
 	coskey "github.com/cosmos/cosmos-sdk/crypto/keys/secp256k1"
 	sdk "github.com/cosmos/cosmos-sdk/types"
 	bech32 "github.com/cosmos/cosmos-sdk/types/bech32/legacybech32"
+	"github.com/pkg/errors"
 )
 
 type (
 	KeygenLocalState struct {
 		PubKey          string                    `json:"pub_key"`
 		LocalData       keygen.LocalPartySaveData `json:"local_data"`
-		ParticipantKeys []string                  `json:"participant_keys"` // the paticipant of last key gen
+		ParticipantKeys []string                  `json:"participant_keys"` // the participant of last key gen
 		LocalPartyKey   string                    `json:"local_party_key"`
 	}
 )
@@ -32,14 +30,17 @@ func getTssSecretFile(file string) (KeygenLocalState, error) {
 	if err != nil {
 		return KeygenLocalState{}, err
 	}
-	buf, err := ioutil.ReadFile(file)
+
+	buf, err := os.ReadFile(file)
 	if err != nil {
-		return KeygenLocalState{}, fmt.Errorf("file to read from file(%s): %w", file, err)
+		return KeygenLocalState{}, errors.Wrapf(err, "file to read from file %q", file)
 	}
+
 	var localState KeygenLocalState
 	if err := json.Unmarshal(buf, &localState); nil != err {
-		return KeygenLocalState{}, fmt.Errorf("fail to unmarshal KeygenLocalState: %w", err)
+		return KeygenLocalState{}, errors.Wrapf(err, "fail to unmarshal KeygenLocalState")
 	}
+
 	return localState, nil
 }
 

cmd/tss/mock_tss_server.go

@@ -1,9 +1,8 @@
 package main
 
 import (
-	"errors"
-
 	"github.com/libp2p/go-libp2p/core/peer"
+	"github.com/pkg/errors"
 
 	"github.com/zeta-chain/go-tss/blame"
 	"github.com/zeta-chain/go-tss/common"