*: refine the fatal log

cmd/pd-server/main.go

@@ -25,6 +25,7 @@ import (
 	"github.com/pingcap/log"
 	"github.com/tikv/pd/pkg/autoscaling"
 	"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"
@@ -54,7 +55,7 @@ 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 {
@@ -67,15 +68,15 @@ 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()
@@ -100,7 +101,7 @@ 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

@@ -22,6 +22,11 @@ var (
 )
 
 // 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 (
 	ErrGetAllocator       = errors.Normalize("get allocator failed, %s", errors.RFCCodeText("PD:tso:ErrGetAllocator"))
@@ -64,6 +69,7 @@ var (
 // scheduler errors
 var (
 	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"))
@@ -86,6 +92,11 @@ var (
 	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"))
@@ -115,6 +126,16 @@ var (
 	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"))
+)
+
 // The third-party project error.
 // url errors
 var (
@@ -228,6 +249,17 @@ var (
 
 // leveldb errors
 var (
-	ErrLevelDBClose = errors.Normalize("failed to close leveldb", errors.RFCCodeText("PD:leveldb:ErrLevelDBClose"))
-	ErrLevelDBWrite = errors.Normalize("failed to write leveldb", errors.RFCCodeText("PD:leveldb:ErrLevelDBWrite"))
+	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"))
 )

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/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"
@@ -1109,7 +1110,7 @@ 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)
@@ -1122,7 +1123,7 @@ 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

@@ -566,7 +566,7 @@ func (s *StoresInfo) ResumeLeaderTransfer(storeID uint64) {
 	store, ok := s.stores[storeID]
 	if !ok {
 		log.Fatal("try to clean a store's pause state, but it is not found",
-			zap.Uint64("store-id", storeID))
+			zap.Uint64("store-id", storeID), errs.ZapError(errs.ErrStoreNotFound.FastGenByArgs(storeID)))
 	}
 	s.stores[storeID] = store.Clone(ResumeLeaderTransfer())
 }

server/join/join.go

@@ -29,7 +29,6 @@ import (
 	"github.com/tikv/pd/server/config"
 	"go.etcd.io/etcd/clientv3"
 	"go.etcd.io/etcd/embed"
-	"go.uber.org/zap"
 )
 
 const (
@@ -94,7 +93,7 @@ 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

@@ -31,7 +31,7 @@ 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

@@ -96,7 +96,7 @@ 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
 }
@@ -105,7 +105,7 @@ 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
 	config.RegisterScheduler(typ)

server/schedulers/adjacent_region.go

@@ -234,7 +234,6 @@ func (l *balanceAdjacentRegionScheduler) process(cluster opt.Cluster) []*operato
 	defer func() {
 		if l.cacheRegions.len() < 0 {
 			log.Fatal("cache overflow", zap.String("scheduler", l.GetName()), errs.ZapError(errs.ErrCacheOverflow))
-			log.Fatal("cache overflow", zap.String("scheduler", l.GetName()), errs.ZapError(errs.ErrCacheOverflow))
 		}
 		l.cacheRegions.head = head + 1
 		l.lastKey = r2.GetStartKey()

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/config"
@@ -100,7 +101,7 @@ 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.
@@ -110,12 +111,12 @@ 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)

server/versioninfo/feature.go

@@ -16,6 +16,7 @@ package versioninfo
 import (
 	"github.com/coreos/go-semver/semver"
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	"go.uber.org/zap"
 )
 
@@ -58,7 +59,7 @@ 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/versioninfo/versioninfo.go

@@ -15,9 +15,8 @@ package versioninfo
 
 import (
 	"github.com/coreos/go-semver/semver"
-	"github.com/pingcap/errors"
 	"github.com/pingcap/log"
-	"go.uber.org/zap"
+	"github.com/tikv/pd/pkg/errs"
 )
 
 const (
@@ -44,14 +43,17 @@ func ParseVersion(v string) (*semver.Version, error) {
 		v = v[1:]
 	}
 	ver, err := semver.NewVersion(v)
-	return ver, errors.WithStack(err)
+	if err != nil {
+		return nil, errs.ErrSemverNewVersion.Wrap(err).GenWithStackByCause()
+	}
+	return ver, nil
 }
 
 // MustParseVersion wraps ParseVersion and will panic if error is not nil.
 func MustParseVersion(v string) *semver.Version {
 	ver, err := ParseVersion(v)
 	if err != nil {
-		log.Fatal("version string is illegal", zap.Error(err))
+		log.Fatal("version string is illegal", errs.ZapError(err))
 	}
 	return ver
 }