update more errors

Signed-off-by: Ryan Leung <rleungx@gmail.com>

pkg/dashboard/adapter/manager.go

@@ -21,7 +21,6 @@ import (
 
 	"github.com/pingcap-incubator/tidb-dashboard/pkg/apiserver"
 	"github.com/pingcap/kvproto/pkg/pdpb"
-	"go.uber.org/zap"
 
 	"github.com/pingcap/log"
 	"github.com/tikv/pd/pkg/errs"
@@ -106,7 +105,7 @@ func (m *Manager) updateInfo() {
 
 	var err error
 	if m.members, err = cluster.GetMembers(m.srv.GetClient()); err != nil {
-		log.Warn("failed to get members", zap.Error(err))
+		log.Warn("failed to get members", errs.ZapError(err))
 		m.members = nil
 		return
 	}

pkg/errs/errno.go

@@ -126,5 +126,15 @@ var (
 
 // ioutil error
 var (
-	ErrIORead = errors.Normalize("IO read err", errors.RFCCodeText("PD:ioutil:ErrIORead"))
+	ErrIORead = errors.Normalize("IO read error", errors.RFCCodeText("PD:ioutil:ErrIORead"))
+)
+
+// netstat error
+var (
+	ErrNetstatTCPSocks = errors.Normalize("TCP socks error", errors.RFCCodeText("PD:netstat:ErrNetstatTCPSocks"))
+)
+
+// hex error
+var (
+	ErrHexDecodingString = errors.Normalize("decode string %s error", errors.RFCCodeText("PD:hex:ErrHexDecodingString"))
 )

pkg/errs/errs.go

@@ -21,6 +21,9 @@ import (
 
 // ZapError is used to make the log output eaiser.
 func ZapError(err error, causeError ...error) zap.Field {
+	if err == nil {
+		return zap.Skip()
+	}
 	if e, ok := err.(*errors.Error); ok {
 		if len(causeError) >= 1 {
 			err = e.Wrap(causeError[0]).FastGenWithCause()

pkg/tempurl/check_env_linux.go

@@ -17,13 +17,13 @@ package tempurl
 import (
 	"github.com/cakturk/go-netstat/netstat"
 	"github.com/pingcap/log"
-	"go.uber.org/zap"
+	"github.com/tikv/pd/pkg/errs"
 )
 
 func environmentCheck(addr string) bool {
 	valid, err := checkAddr(addr[len("http://"):])
 	if err != nil {
-		log.Error("check port status failed", zap.Error(err))
+		log.Error("check port status failed", errs.ZapError(err))
 		return false
 	}
 	return valid
@@ -34,7 +34,7 @@ func checkAddr(addr string) (bool, error) {
 		return s.RemoteAddr.String() == addr || s.LocalAddr.String() == addr
 	})
 	if err != nil {
-		return false, err
+		return false, errs.ErrNetstatTCPSocks.Wrap(err).FastGenWithCause()
 	}
 	return len(tabs) < 1, nil
 }

plugin/scheduler_example/evict_leader.go

@@ -23,14 +23,14 @@ import (
 	"github.com/pingcap/errors"
 	"github.com/pingcap/log"
 	"github.com/tikv/pd/pkg/apiutil"
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/server/core"
 	"github.com/tikv/pd/server/schedule"
 	"github.com/tikv/pd/server/schedule/filter"
 	"github.com/tikv/pd/server/schedule/operator"
 	"github.com/tikv/pd/server/schedule/opt"
 	"github.com/tikv/pd/server/schedulers"
 	"github.com/unrolled/render"
-	"go.uber.org/zap"
 )
 
 const (
@@ -226,7 +226,7 @@ func (s *evictLeaderScheduler) Schedule(cluster opt.Cluster) []*operator.Operato
 		}
 		op, err := operator.CreateTransferLeaderOperator(EvictLeaderType, cluster, region, region.GetLeader().GetStoreId(), target.GetID(), operator.OpLeader)
 		if err != nil {
-			log.Debug("fail to create evict leader operator", zap.Error(err))
+			log.Debug("fail to create evict leader operator", errs.ZapError(err))
 			continue
 
 		}

server/cluster/cluster_worker.go

@@ -192,7 +192,7 @@ func (c *RaftCluster) HandleReportSplit(request *pdpb.ReportSplitRequest) (*pdpb
 		log.Warn("report split region is invalid",
 			zap.Stringer("left-region", core.RegionToHexMeta(left)),
 			zap.Stringer("right-region", core.RegionToHexMeta(right)),
-			zap.Error(err))
+			errs.ZapError(err))
 		return nil, err
 	}
 
@@ -215,7 +215,7 @@ func (c *RaftCluster) HandleBatchReportSplit(request *pdpb.ReportBatchSplitReque
 	if err != nil {
 		log.Warn("report batch split region is invalid",
 			zap.Stringer("region-meta", hrm),
-			zap.Error(err))
+			errs.ZapError(err))
 		return nil, err
 	}
 	last := len(regions) - 1

server/cluster/coordinator.go

@@ -645,7 +645,7 @@ func (c *coordinator) runScheduler(s *scheduleController) {
 		case <-s.Ctx().Done():
 			log.Info("scheduler has been stopped",
 				zap.String("scheduler-name", s.GetName()),
-				zap.Error(s.Ctx().Err()))
+				errs.ZapError(s.Ctx().Err()))
 			return
 		}
 	}

server/core/basic_cluster.go

@@ -19,6 +19,7 @@ import (
 
 	"github.com/pingcap/kvproto/pkg/metapb"
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/pkg/slice"
 	"github.com/tikv/pd/server/schedule/storelimit"
 	"go.uber.org/zap"
@@ -329,7 +330,7 @@ func (bc *BasicCluster) PutRegion(region *RegionInfo) []*RegionInfo {
 func (bc *BasicCluster) CheckAndPutRegion(region *RegionInfo) []*RegionInfo {
 	origin, err := bc.PreCheckPutRegion(region)
 	if err != nil {
-		log.Debug("region is stale", zap.Error(err), zap.Stringer("origin", origin.GetMeta()))
+		log.Debug("region is stale", zap.Stringer("origin", origin.GetMeta()), errs.ZapError(err))
 		// return the state region to delete.
 		return []*RegionInfo{region}
 	}

server/election/leadership.go

@@ -101,7 +101,7 @@ func (ls *Leadership) Campaign(leaseTimeout int64, leaderData string) error {
 		If(clientv3.Compare(clientv3.CreateRevision(ls.leaderKey), "=", 0)).
 		Then(clientv3.OpPut(ls.leaderKey, leaderData, clientv3.WithLease(ls.getLease().ID))).
 		Commit()
-	log.Info("check campaign resp", zap.Any("resp", resp), zap.Error(err))
+	log.Info("check campaign resp", zap.Any("resp", resp))
 	if err != nil {
 		return errors.WithStack(err)
 	}

server/election/lease.go

@@ -20,6 +20,7 @@ import (
 
 	"github.com/pingcap/errors"
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/pkg/etcdutil"
 	"go.etcd.io/etcd/clientv3"
 	"go.uber.org/zap"
@@ -125,7 +126,7 @@ func (l *lease) keepAliveWorker(ctx context.Context, interval time.Duration) <-c
 				defer cancel()
 				res, err := l.lease.KeepAliveOnce(ctx1, l.ID)
 				if err != nil {
-					log.Warn("lease keep alive failed", zap.Error(err), zap.String("purpose", l.Purpose))
+					log.Warn("lease keep alive failed", zap.String("purpose", l.Purpose), errs.ZapError(err))
 					return
 				}
 				if res.TTL > 0 {

server/handler.go

@@ -28,6 +28,7 @@ import (
 	"github.com/pingcap/kvproto/pkg/metapb"
 	"github.com/pingcap/kvproto/pkg/pdpb"
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/server/cluster"
 	"github.com/tikv/pd/server/config"
 	"github.com/tikv/pd/server/core"
@@ -457,7 +458,7 @@ func (h *Handler) AddTransferLeaderOperator(regionID uint64, storeID uint64) err
 
 	op, err := operator.CreateTransferLeaderOperator("admin-transfer-leader", c, region, region.GetLeader().GetStoreId(), newLeader.GetStoreId(), operator.OpAdmin)
 	if err != nil {
-		log.Debug("fail to create transfer leader operator", zap.Error(err))
+		log.Debug("fail to create transfer leader operator", errs.ZapError(err))
 		return err
 	}
 	if ok := c.GetOperatorController().AddOperator(op); !ok {
@@ -505,7 +506,7 @@ func (h *Handler) AddTransferRegionOperator(regionID uint64, storeIDs map[uint64
 
 	op, err := operator.CreateMoveRegionOperator("admin-move-region", c, region, operator.OpAdmin, peers)
 	if err != nil {
-		log.Debug("fail to create move region operator", zap.Error(err))
+		log.Debug("fail to create move region operator", errs.ZapError(err))
 		return err
 	}
 	if ok := c.GetOperatorController().AddOperator(op); !ok {
@@ -542,7 +543,7 @@ func (h *Handler) AddTransferPeerOperator(regionID uint64, fromStoreID, toStoreI
 	newPeer := &metapb.Peer{StoreId: toStoreID, Role: oldPeer.GetRole()}
 	op, err := operator.CreateMovePeerOperator("admin-move-peer", c, region, operator.OpAdmin, fromStoreID, newPeer)
 	if err != nil {
-		log.Debug("fail to create move peer operator", zap.Error(err))
+		log.Debug("fail to create move peer operator", errs.ZapError(err))
 		return err
 	}
 	if ok := c.GetOperatorController().AddOperator(op); !ok {
@@ -588,7 +589,7 @@ func (h *Handler) AddAddPeerOperator(regionID uint64, toStoreID uint64) error {
 	newPeer := &metapb.Peer{StoreId: toStoreID}
 	op, err := operator.CreateAddPeerOperator("admin-add-peer", c, region, newPeer, operator.OpAdmin)
 	if err != nil {
-		log.Debug("fail to create add peer operator", zap.Error(err))
+		log.Debug("fail to create add peer operator", errs.ZapError(err))
 		return err
 	}
 	if ok := c.GetOperatorController().AddOperator(op); !ok {
@@ -611,7 +612,7 @@ func (h *Handler) AddAddLearnerOperator(regionID uint64, toStoreID uint64) error
 
 	op, err := operator.CreateAddPeerOperator("admin-add-learner", c, region, newPeer, operator.OpAdmin)
 	if err != nil {
-		log.Debug("fail to create add learner operator", zap.Error(err))
+		log.Debug("fail to create add learner operator", errs.ZapError(err))
 		return err
 	}
 	if ok := c.GetOperatorController().AddOperator(op); !ok {
@@ -638,7 +639,7 @@ func (h *Handler) AddRemovePeerOperator(regionID uint64, fromStoreID uint64) err
 
 	op, err := operator.CreateRemovePeerOperator("admin-remove-peer", c, operator.OpAdmin, region, fromStoreID)
 	if err != nil {
-		log.Debug("fail to create move peer operator", zap.Error(err))
+		log.Debug("fail to create move peer operator", errs.ZapError(err))
 		return err
 	}
 	if ok := c.GetOperatorController().AddOperator(op); !ok {
@@ -680,7 +681,7 @@ func (h *Handler) AddMergeRegionOperator(regionID uint64, targetID uint64) error
 
 	ops, err := operator.CreateMergeRegionOperator("admin-merge-region", c, region, target, operator.OpAdmin)
 	if err != nil {
-		log.Debug("fail to create merge region operator", zap.Error(err))
+		log.Debug("fail to create merge region operator", errs.ZapError(err))
 		return err
 	}
 	if ok := c.GetOperatorController().AddOperator(ops...); !ok {

server/heartbeat_streams.go

@@ -22,6 +22,7 @@ import (
 	"github.com/pingcap/kvproto/pkg/metapb"
 	"github.com/pingcap/kvproto/pkg/pdpb"
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/pkg/logutil"
 	"github.com/tikv/pd/server/cluster"
 	"github.com/tikv/pd/server/core"
@@ -120,7 +121,7 @@ func (s *heartbeatStreams) run() {
 				if err := stream.Send(keepAlive); err != nil {
 					log.Warn("send keepalive message fail, store maybe disconnected",
 						zap.Uint64("target-store-id", storeID),
-						zap.Error(err))
+						errs.ZapError(err))
 					delete(s.streams, storeID)
 					regionHeartbeatCounter.WithLabelValues(storeAddress, storeLabel, "keepalive", "err").Inc()
 				} else {

server/kv/etcd_kv.go

@@ -89,9 +89,9 @@ func (kv *etcdKVBase) Save(key, value string) error {
 	txn := NewSlowLogTxn(kv.client)
 	resp, err := txn.Then(clientv3.OpPut(key, value)).Commit()
 	if err != nil {
-		err = errs.ErrEtcdKVPut.Wrap(err).GenWithStackByCause()
-		log.Error("save to etcd meet error", zap.String("key", key), zap.String("value", value), errs.ZapError(err))
-		return err
+		e := errs.ErrEtcdKVPut.Wrap(err).GenWithStackByCause()
+		log.Error("save to etcd meet error", zap.String("key", key), zap.String("value", value), errs.ZapError(e))
+		return e
 	}
 	if !resp.Succeeded {
 		return errs.ErrEtcdTxn.FastGenByArgs()
@@ -158,9 +158,9 @@ func (t *SlowLogTxn) Commit() (*clientv3.TxnResponse, error) {
 	cost := time.Since(start)
 	if cost > slowRequestTime {
 		log.Warn("txn runs too slow",
-			zap.Error(err),
 			zap.Reflect("response", resp),
-			zap.Duration("cost", cost))
+			zap.Duration("cost", cost),
+			errs.ZapError(err))
 	}
 	label := "success"
 	if err != nil {

server/region_syncer/history_buffer.go

@@ -18,6 +18,7 @@ import (
 	"sync"
 
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/server/core"
 	"github.com/tikv/pd/server/kv"
 	"go.uber.org/zap"
@@ -151,6 +152,6 @@ func (h *historyBuffer) persist() {
 	regionSyncerStatus.WithLabelValues("last_index").Set(float64(h.nextIndex()))
 	err := h.kv.Save(historyKey, strconv.FormatUint(h.nextIndex(), 10))
 	if err != nil {
-		log.Warn("persist history index failed", zap.Uint64("persist-index", h.nextIndex()), zap.Error(err))
+		log.Warn("persist history index failed", zap.Uint64("persist-index", h.nextIndex()), errs.ZapError(err))
 	}
 }

server/replication/replication_mode.go

@@ -23,6 +23,7 @@ import (
 
 	pb "github.com/pingcap/kvproto/pkg/replication_modepb"
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/server/config"
 	"github.com/tikv/pd/server/core"
 	"github.com/tikv/pd/server/schedule/opt"
@@ -217,15 +218,15 @@ func (m *ModeManager) drSwitchToAsync() error {
 func (m *ModeManager) drSwitchToAsyncWithLock() error {
 	id, err := m.cluster.AllocID()
 	if err != nil {
-		log.Warn("failed to switch to async state", zap.String("replicate-mode", modeDRAutoSync), zap.Error(err))
+		log.Warn("failed to switch to async state", zap.String("replicate-mode", modeDRAutoSync), errs.ZapError(err))
 		return err
 	}
 	dr := drAutoSyncStatus{State: drStateAsync, StateID: id}
 	if err := m.drPersistStatus(dr); err != nil {
 		return err
 	}
 	if err := m.storage.SaveReplicationStatus(modeDRAutoSync, dr); err != nil {
-		log.Warn("failed to switch to async state", zap.String("replicate-mode", modeDRAutoSync), zap.Error(err))
+		log.Warn("failed to switch to async state", zap.String("replicate-mode", modeDRAutoSync), errs.ZapError(err))
 		return err
 	}
 	m.drAutoSync = dr
@@ -242,15 +243,15 @@ func (m *ModeManager) drSwitchToSyncRecover() error {
 func (m *ModeManager) drSwitchToSyncRecoverWithLock() error {
 	id, err := m.cluster.AllocID()
 	if err != nil {
-		log.Warn("failed to switch to sync_recover state", zap.String("replicate-mode", modeDRAutoSync), zap.Error(err))
+		log.Warn("failed to switch to sync_recover state", zap.String("replicate-mode", modeDRAutoSync), errs.ZapError(err))
 		return err
 	}
 	dr := drAutoSyncStatus{State: drStateSyncRecover, StateID: id, RecoverStartTime: time.Now()}
 	if err := m.drPersistStatus(dr); err != nil {
 		return err
 	}
 	if err = m.storage.SaveReplicationStatus(modeDRAutoSync, dr); err != nil {
-		log.Warn("failed to switch to sync_recover state", zap.String("replicate-mode", modeDRAutoSync), zap.Error(err))
+		log.Warn("failed to switch to sync_recover state", zap.String("replicate-mode", modeDRAutoSync), errs.ZapError(err))
 		return err
 	}
 	m.drAutoSync = dr
@@ -264,15 +265,15 @@ func (m *ModeManager) drSwitchToSync() error {
 	defer m.Unlock()
 	id, err := m.cluster.AllocID()
 	if err != nil {
-		log.Warn("failed to switch to sync state", zap.String("replicate-mode", modeDRAutoSync), zap.Error(err))
+		log.Warn("failed to switch to sync state", zap.String("replicate-mode", modeDRAutoSync), errs.ZapError(err))
 		return err
 	}
 	dr := drAutoSyncStatus{State: drStateSync, StateID: id}
 	if err := m.drPersistStatus(dr); err != nil {
 		return err
 	}
 	if err := m.storage.SaveReplicationStatus(modeDRAutoSync, dr); err != nil {
-		log.Warn("failed to switch to sync state", zap.String("replicate-mode", modeDRAutoSync), zap.Error(err))
+		log.Warn("failed to switch to sync state", zap.String("replicate-mode", modeDRAutoSync), errs.ZapError(err))
 		return err
 	}
 	m.drAutoSync = dr
@@ -286,7 +287,7 @@ func (m *ModeManager) drPersistStatus(status drAutoSyncStatus) error {
 		defer cancel()
 		data, _ := json.Marshal(status)
 		if err := m.fileReplicater.ReplicateFileToAllMembers(ctx, drStatusFile, data); err != nil {
-			log.Warn("failed to switch state", zap.String("replicate-mode", modeDRAutoSync), zap.String("new-state", status.State), zap.Error(err))
+			log.Warn("failed to switch state", zap.String("replicate-mode", modeDRAutoSync), zap.String("new-state", status.State), errs.ZapError(err))
 			// Throw away the error to make it possible to switch to async when
 			// primary and dr DC are disconnected. This will result in the
 			// inability to accurately determine whether data is fully

server/schedule/checker/learner_checker.go

@@ -15,10 +15,10 @@ package checker
 
 import (
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/server/core"
 	"github.com/tikv/pd/server/schedule/operator"
 	"github.com/tikv/pd/server/schedule/opt"
-	"go.uber.org/zap"
 )
 
 // LearnerChecker ensures region has a learner will be promoted.
@@ -41,7 +41,7 @@ func (l *LearnerChecker) Check(region *core.RegionInfo) *operator.Operator {
 		}
 		op, err := operator.CreatePromoteLearnerOperator("promote-learner", l.cluster, region, p)
 		if err != nil {
-			log.Debug("fail to create promote learner operator", zap.Error(err))
+			log.Debug("fail to create promote learner operator", errs.ZapError(err))
 			return nil
 		}
 		return op