cherry pick tikv#2896 to release-4.0

Signed-off-by: ti-srebot <ti-srebot@pingcap.com>

cmd/pd-server/main.go

@@ -24,6 +24,7 @@ import (
 	"github.com/pingcap/errors"
 	"github.com/pingcap/log"
 	"github.com/tikv/pd/pkg/dashboard"
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/pkg/logutil"
 	"github.com/tikv/pd/pkg/metricutil"
 	"github.com/tikv/pd/pkg/swaggerserver"
@@ -53,7 +54,7 @@ func main() {
 	case flag.ErrHelp:
 		exit(0)
 	default:
-		log.Fatal("parse cmd flags error", zap.Error(err))
+		log.Fatal("parse cmd flags error", errs.ZapError(errs.ErrParseFlags))
 	}
 
 	if cfg.ConfigCheck {
@@ -66,15 +67,15 @@ func main() {
 	if err == nil {
 		log.ReplaceGlobals(cfg.GetZapLogger(), cfg.GetZapLogProperties())
 	} else {
-		log.Fatal("initialize logger error", zap.Error(err))
+		log.Fatal("initialize logger error", errs.ZapError(err))
 	}
 	// Flushing any buffered log entries
 	defer log.Sync()
 
 	// The old logger
 	err = logutil.InitLogger(&cfg.Log)
 	if err != nil {
-		log.Fatal("initialize logger error", zap.Error(err))
+		log.Fatal("initialize logger error", errs.ZapError(err))
 	}
 
 	server.LogPDInfo()
@@ -99,7 +100,7 @@ func main() {
 	serviceBuilders = append(serviceBuilders, dashboard.GetServiceBuilders()...)
 	svr, err := server.CreateServer(ctx, cfg, serviceBuilders...)
 	if err != nil {
-		log.Fatal("create server failed", zap.Error(err))
+		log.Fatal("create server failed", errs.ZapError(err))
 	}
 
 	sc := make(chan os.Signal, 1)

pkg/errs/errno.go

@@ -16,6 +16,11 @@ package errs
 import "github.com/pingcap/errors"
 
 // The internal error which is generated in PD project.
+// main errors
+var (
+	ErrParseFlags = errors.Normalize("parse flags error", errors.RFCCodeText("PD:main:ErrParseFlags"))
+)
+
 // tso errors
 var (
 	ErrInvalidTimestamp    = errors.Normalize("invalid timestamp", errors.RFCCodeText("PD:tso:ErrInvalidTimestamp"))
@@ -41,13 +46,24 @@ var (
 
 // scheduler errors
 var (
+<<<<<<< HEAD
 	ErrGetSourceStore         = errors.Normalize("failed to get the source store", errors.RFCCodeText("PD:scheduler:ErrGetSourceStore"))
 	ErrSchedulerExisted       = errors.Normalize("scheduler existed", errors.RFCCodeText("PD:scheduler:ErrSchedulerExisted"))
 	ErrSchedulerNotFound      = errors.Normalize("scheduler not found", errors.RFCCodeText("PD:scheduler:ErrSchedulerNotFound"))
 	ErrScheduleConfigNotExist = errors.Normalize("the config does not exist", errors.RFCCodeText("PD:scheduler:ErrScheduleConfigNotExist"))
 	ErrSchedulerConfig        = errors.Normalize("wrong scheduler config %s", errors.RFCCodeText("PD:scheduler:ErrSchedulerConfig"))
 	ErrCacheOverflow          = errors.Normalize("cache overflow", errors.RFCCodeText("PD:scheduler:ErrCacheOverflow"))
 	ErrInternalGrowth         = errors.Normalize("unknown interval growth type error", errors.RFCCodeText("PD:scheduler:ErrInternalGrowth"))
+=======
+	ErrSchedulerExisted                 = errors.Normalize("scheduler existed", errors.RFCCodeText("PD:scheduler:ErrSchedulerExisted"))
+	ErrSchedulerDuplicated              = errors.Normalize("scheduler duplicated", errors.RFCCodeText("PD:scheduler:ErrSchedulerDuplicated"))
+	ErrSchedulerNotFound                = errors.Normalize("scheduler not found", errors.RFCCodeText("PD:scheduler:ErrSchedulerNotFound"))
+	ErrScheduleConfigNotExist           = errors.Normalize("the config does not exist", errors.RFCCodeText("PD:scheduler:ErrScheduleConfigNotExist"))
+	ErrSchedulerConfig                  = errors.Normalize("wrong scheduler config %s", errors.RFCCodeText("PD:scheduler:ErrSchedulerConfig"))
+	ErrCacheOverflow                    = errors.Normalize("cache overflow", errors.RFCCodeText("PD:scheduler:ErrCacheOverflow"))
+	ErrInternalGrowth                   = errors.Normalize("unknown interval growth type error", errors.RFCCodeText("PD:scheduler:ErrInternalGrowth"))
+	ErrSchedulerCreateFuncNotRegistered = errors.Normalize("create func of %v is not registered", errors.RFCCodeText("PD:scheduler:ErrSchedulerCreateFuncNotRegistered"))
+>>>>>>> 67cc378... *: refine the fatal log (#2896)
 )
 
 // placement errors
@@ -60,20 +76,67 @@ var (
 
 // cluster errors
 var (
+<<<<<<< HEAD
 	ErrPersistStore          = errors.Normalize("failed to persist store", errors.RFCCodeText("PD:cluster:ErrPersistStore"))
 	ErrDeleteRegion          = errors.Normalize("failed to delete region from storage", errors.RFCCodeText("PD:cluster:ErrDeleteRegion"))
 	ErrSaveRegion            = errors.Normalize("failed to save region from storage", errors.RFCCodeText("PD:cluster:ErrSaveRegion"))
 	ErrBuryStore             = errors.Normalize("failed to bury store", errors.RFCCodeText("PD:cluster:ErrBuryStore"))
 	ErrDeleteStore           = errors.Normalize("failed to delete store", errors.RFCCodeText("PD:cluster:ErrDeleteStore"))
 	ErrPersistClusterVersion = errors.Normalize("persist cluster version meet error", errors.RFCCodeText("PD:cluster:ErrPersistClusterVersion"))
 	ErrGetMembers            = errors.Normalize("get members failed", errors.RFCCodeText("PD:cluster:ErrGetMembers"))
+=======
+	ErrNotBootstrapped = errors.Normalize("TiKV cluster not bootstrapped, please start TiKV first", errors.RFCCodeText("PD:cluster:ErrNotBootstrapped"))
+	ErrStoreIsUp       = errors.Normalize("store is still up, please remove store gracefully", errors.RFCCodeText("PD:cluster:ErrStoreIsUp"))
+)
+
+// versioninfo errors
+var (
+	ErrFeatureNotExisted = errors.Normalize("feature not existed", errors.RFCCodeText("PD:versioninfo:ErrFeatureNotExisted"))
+)
+
+// autoscaling errors
+var (
+	ErrUnsupportedMetricsType   = errors.Normalize("unsupported metrics type %v", errors.RFCCodeText("PD:autoscaling:ErrUnsupportedMetricsType"))
+	ErrUnsupportedComponentType = errors.Normalize("unsupported component type %v", errors.RFCCodeText("PD:autoscaling:ErrUnsupportedComponentType"))
+	ErrUnexpectedType           = errors.Normalize("unexpected type %v", errors.RFCCodeText("PD:autoscaling:ErrUnexpectedType"))
+	ErrTypeConversion           = errors.Normalize(("type conversion error"), errors.RFCCodeText("PD:autoscaling:ErrTypeConversion"))
+	ErrEmptyMetricsResponse     = errors.Normalize("metrics response from Prometheus is empty", errors.RFCCodeText("PD:autoscaling:ErrEmptyMetricsResponse"))
+	ErrEmptyMetricsResult       = errors.Normalize("result from Prometheus is empty, %s", errors.RFCCodeText("PD:autoscaling:ErrEmptyMetricsResult"))
+)
+
+// apiutil errors
+var (
+	ErrRedirect = errors.Normalize("redirect failed", errors.RFCCodeText("PD:apiutil:ErrRedirect"))
+>>>>>>> 67cc378... *: refine the fatal log (#2896)
 )
 
 // grpcutil errors
 var (
 	ErrSecurityConfig = errors.Normalize("security config error: %s", errors.RFCCodeText("PD:grpcutil:ErrSecurityConfig"))
 )
 
+<<<<<<< HEAD
+=======
+// server errors
+var (
+	ErrServiceRegistered     = errors.Normalize("service with path [%s] already registered", errors.RFCCodeText("PD:server:ErrServiceRegistered"))
+	ErrAPIInformationInvalid = errors.Normalize("invalid api information, group %s version %s", errors.RFCCodeText("PD:server:ErrAPIInformationInvalid"))
+	ErrClientURLEmpty        = errors.Normalize("client url empty", errors.RFCCodeText("PD:server:ErrClientEmpty"))
+	ErrLeaderNil             = errors.Normalize("leader is nil", errors.RFCCodeText("PD:server:ErrLeaderNil"))
+	ErrCancelStartEtcd       = errors.Normalize("etcd start canceled", errors.RFCCodeText("PD:server:ErrCancelStartEtcd"))
+)
+
+// logutil errors
+var (
+	ErrInitFileLog = errors.Normalize("init file log error, %s", errors.RFCCodeText("PD:logutil:ErrInitFileLog"))
+)
+
+// typeutil errors
+var (
+	ErrBytesToUint64 = errors.Normalize("invalid data, must 8 bytes, but %d", errors.RFCCodeText("PD:typeutil:ErrBytesToUint64"))
+)
+
+>>>>>>> 67cc378... *: refine the fatal log (#2896)
 // The third-party project error.
 // url errors
 var (
@@ -149,3 +212,40 @@ var (
 var (
 	ErrHexDecodingString = errors.Normalize("decode string %s error", errors.RFCCodeText("PD:hex:ErrHexDecodingString"))
 )
+<<<<<<< HEAD
+=======
+
+// filepath errors
+var (
+	ErrFilePathAbs = errors.Normalize("failed to convert a path to absolute path", errors.RFCCodeText("PD:filepath:ErrFilePathAbs"))
+)
+
+// plugin errors
+var (
+	ErrLoadPlugin       = errors.Normalize("failed to load plugin", errors.RFCCodeText("PD:plugin:ErrLoadPlugin"))
+	ErrLookupPluginFunc = errors.Normalize("failed to lookup plugin function", errors.RFCCodeText("PD:plugin:ErrLookupPluginFunc"))
+)
+
+// json errors
+var (
+	ErrJSONMarshal   = errors.Normalize("failed to marshal json", errors.RFCCodeText("PD:json:ErrJSONMarshal"))
+	ErrJSONUnmarshal = errors.Normalize("failed to unmarshal json", errors.RFCCodeText("PD:json:ErrJSONUnmarshal"))
+)
+
+// leveldb errors
+var (
+	ErrLevelDBClose = errors.Normalize("close leveldb error", errors.RFCCodeText("PD:leveldb:ErrLevelDBClose"))
+	ErrLevelDBWrite = errors.Normalize("leveldb write error", errors.RFCCodeText("PD:leveldb:ErrLevelDBWrite"))
+	ErrLevelDBOpen  = errors.Normalize("leveldb open file error", errors.RFCCodeText("PD:leveldb:ErrLevelDBOpen"))
+)
+
+// semver
+var (
+	ErrSemverNewVersion = errors.Normalize("new version error", errors.RFCCodeText("PD:semver:ErrSemverNewVersion"))
+)
+
+// log
+var (
+	ErrInitLogger = errors.Normalize("init logger error", errors.RFCCodeText("PD:log:ErrInitLogger"))
+)
+>>>>>>> 67cc378... *: refine the fatal log (#2896)

pkg/logutil/log.go

@@ -23,9 +23,9 @@ import (
 	"sync"
 
 	"github.com/coreos/pkg/capnslog"
-	"github.com/pingcap/errors"
 	zaplog "github.com/pingcap/log"
 	log "github.com/sirupsen/logrus"
+	"github.com/tikv/pd/pkg/errs"
 	"go.etcd.io/etcd/raft"
 	"go.uber.org/zap"
 	"go.uber.org/zap/zapcore"
@@ -217,7 +217,7 @@ func StringToLogFormatter(format string, disableTimestamp bool) log.Formatter {
 func InitFileLog(cfg *zaplog.FileLogConfig) error {
 	if st, err := os.Stat(cfg.Filename); err == nil {
 		if st.IsDir() {
-			return errors.New("can't use directory as log file name")
+			return errs.ErrInitFileLog.FastGenByArgs("can't use directory as log file name")
 		}
 	}
 	if cfg.MaxSize == 0 {

pkg/typeutil/convension.go

@@ -16,13 +16,13 @@ package typeutil
 import (
 	"encoding/binary"
 
-	"github.com/pingcap/errors"
+	"github.com/tikv/pd/pkg/errs"
 )
 
 // BytesToUint64 converts a byte slice to uint64.
 func BytesToUint64(b []byte) (uint64, error) {
 	if len(b) != 8 {
-		return 0, errors.Errorf("invalid data, must 8 bytes, but %d", len(b))
+		return 0, errs.ErrBytesToUint64.FastGenByArgs(len(b))
 	}
 
 	return binary.BigEndian.Uint64(b), nil

server/cluster/version.go

@@ -17,6 +17,7 @@ import (
 	"github.com/coreos/go-semver/semver"
 	"github.com/pingcap/errors"
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	"go.uber.org/zap"
 )
 
@@ -50,7 +51,7 @@ var featuresDict = map[Feature]string{
 func MinSupportedVersion(v Feature) *semver.Version {
 	target, ok := featuresDict[v]
 	if !ok {
-		log.Fatal("the corresponding version of the feature doesn't exist", zap.Int("feature-number", int(v)))
+		log.Fatal("the corresponding version of the feature doesn't exist", zap.Int("feature-number", int(v)), errs.ZapError(errs.ErrFeatureNotExisted))
 	}
 	version := MustParseVersion(target)
 	return version

server/config/config.go

@@ -26,6 +26,7 @@ import (
 	"sync"
 	"time"
 
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/pkg/grpcutil"
 	"github.com/tikv/pd/pkg/metricutil"
 	"github.com/tikv/pd/pkg/typeutil"
@@ -1088,7 +1089,7 @@ func ParseUrls(s string) ([]url.URL, error) {
 	for _, item := range items {
 		u, err := url.Parse(item)
 		if err != nil {
-			return nil, errors.WithStack(err)
+			return nil, errs.ErrURLParse.Wrap(err).GenWithStackByCause()
 		}
 
 		urls = append(urls, *u)
@@ -1101,7 +1102,7 @@ func ParseUrls(s string) ([]url.URL, error) {
 func (c *Config) SetupLogger() error {
 	lg, p, err := log.InitLogger(&c.Log, zap.AddStacktrace(zapcore.FatalLevel))
 	if err != nil {
-		return err
+		return errs.ErrInitLogger.Wrap(err).FastGenWithCause()
 	}
 	c.logger = lg
 	c.logProps = p

server/core/store.go

@@ -579,8 +579,13 @@ func (s *StoresInfo) BlockStore(storeID uint64) errcode.ErrorCode {
 func (s *StoresInfo) UnblockStore(storeID uint64) {
 	store, ok := s.stores[storeID]
 	if !ok {
+<<<<<<< HEAD
 		log.Fatal("store is unblocked, but it is not found",
 			zap.Uint64("store-id", storeID))
+=======
+		log.Fatal("try to clean a store's pause state, but it is not found",
+			zap.Uint64("store-id", storeID), errs.ZapError(errs.ErrStoreNotFound.FastGenByArgs(storeID)))
+>>>>>>> 67cc378... *: refine the fatal log (#2896)
 	}
 	s.stores[storeID] = store.Clone(SetStoreUnBlock())
 }

server/join/join.go

@@ -28,7 +28,6 @@ import (
 	"github.com/tikv/pd/server/config"
 	"go.etcd.io/etcd/clientv3"
 	"go.etcd.io/etcd/embed"
-	"go.uber.org/zap"
 )
 
 const (
@@ -93,7 +92,7 @@ func PrepareJoinCluster(cfg *config.Config) error {
 	if _, err := os.Stat(filePath); !os.IsNotExist(err) {
 		s, err := ioutil.ReadFile(filePath)
 		if err != nil {
-			log.Fatal("read the join config meet error", zap.Error(err))
+			log.Fatal("read the join config meet error", errs.ZapError(errs.ErrIORead, err))
 		}
 		cfg.InitialCluster = strings.TrimSpace(string(s))
 		cfg.InitialClusterState = embed.ClusterStateFlagExisting

server/kv/levedb_kv.go

@@ -30,7 +30,7 @@ type LeveldbKV struct {
 func NewLeveldbKV(path string) (*LeveldbKV, error) {
 	db, err := leveldb.OpenFile(path, nil)
 	if err != nil {
-		return nil, errors.WithStack(err)
+		return nil, errs.ErrLevelDBOpen.Wrap(err).GenWithStackByCause()
 	}
 	return &LeveldbKV{db}, nil
 }

server/region_syncer/history_buffer.go

@@ -141,7 +141,7 @@ func (h *historyBuffer) reload() {
 	if v != "" {
 		h.index, err = strconv.ParseUint(v, 10, 64)
 		if err != nil {
-			log.Fatal("load history index failed", zap.Error(err))
+			log.Fatal("load history index failed", errs.ZapError(errs.ErrStrconvParseUint, err))
 		}
 	}
 	log.Info("start from history index", zap.Uint64("start-index", h.firstIndex()))

server/schedule/scheduler.go

@@ -86,7 +86,7 @@ var schedulerArgsToDecoder = make(map[string]ConfigSliceDecoderBuilder)
 // func of a package.
 func RegisterScheduler(typ string, createFn CreateSchedulerFunc) {
 	if _, ok := schedulerMap[typ]; ok {
-		log.Fatal("duplicated scheduler", zap.String("type", typ))
+		log.Fatal("duplicated scheduler", zap.String("type", typ), errs.ZapError(errs.ErrSchedulerDuplicated))
 	}
 	schedulerMap[typ] = createFn
 }
@@ -95,7 +95,7 @@ func RegisterScheduler(typ string, createFn CreateSchedulerFunc) {
 // func of package.
 func RegisterSliceDecoderBuilder(typ string, builder ConfigSliceDecoderBuilder) {
 	if _, ok := schedulerArgsToDecoder[typ]; ok {
-		log.Fatal("duplicated scheduler", zap.String("type", typ))
+		log.Fatal("duplicated scheduler", zap.String("type", typ), errs.ZapError(errs.ErrSchedulerDuplicated))
 	}
 	schedulerArgsToDecoder[typ] = builder
 }

server/util.go

@@ -23,6 +23,7 @@ import (
 	"github.com/pingcap/errors"
 	"github.com/pingcap/kvproto/pkg/pdpb"
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	"github.com/tikv/pd/pkg/etcdutil"
 	"github.com/tikv/pd/pkg/typeutil"
 	"github.com/tikv/pd/server/cluster"
@@ -101,7 +102,7 @@ func initOrGetClusterID(c *clientv3.Client, key string) (uint64, error) {
 		Else(clientv3.OpGet(key)).
 		Commit()
 	if err != nil {
-		return 0, errors.WithStack(err)
+		return 0, errs.ErrEtcdTxn.Wrap(err).GenWithStackByCause()
 	}
 
 	// Txn commits ok, return the generated cluster ID.
@@ -111,12 +112,12 @@ func initOrGetClusterID(c *clientv3.Client, key string) (uint64, error) {
 
 	// Otherwise, parse the committed cluster ID.
 	if len(resp.Responses) == 0 {
-		return 0, errors.Errorf("txn returns empty response: %v", resp)
+		return 0, errs.ErrEtcdTxn.FastGenByArgs()
 	}
 
 	response := resp.Responses[0].GetResponseRange()
 	if response == nil || len(response.Kvs) != 1 {
-		return 0, errors.Errorf("txn returns invalid range response: %v", resp)
+		return 0, errs.ErrEtcdTxn.FastGenByArgs()
 	}
 
 	return typeutil.BytesToUint64(response.Kvs[0].Value)