feat: add --log-level support for manager and scheduler (#4146)

Signed-off-by: Aaron Wang <aaron.wang@maiscrm.com>
Co-authored-by: Aaron Wang <aaron.wang@maiscrm.com>

build/package/nfpm/config/dfget.yaml

@@ -26,6 +26,7 @@ scheduler:
 
 # when enabled, pprof will be enabled,
 verbose: true
+log-level: debug
 console: false
 
 # current host info used for scheduler

cmd/dependency/base/option.go

@@ -19,6 +19,7 @@ package base
 type Options struct {
 	Console   bool          `yaml:"console" mapstructure:"console"`
 	Verbose   bool          `yaml:"verbose" mapstructure:"verbose"`
+	LogLevel  string        `yaml:"log-level" mapstructure:"log-level"`
 	PProfPort int           `yaml:"pprof-port" mapstructure:"pprof-port"`
 	Tracing   TracingConfig `yaml:"tracing" mapstructure:"tracing"`
 }

cmd/dependency/dependency.go

@@ -72,6 +72,7 @@ func InitCommandAndConfig(cmd *cobra.Command, useConfigFile bool, config any) {
 		flags := cmd.PersistentFlags()
 		flags.Bool("console", false, "whether logger output records to the stdout")
 		flags.Bool("verbose", false, "whether logger use debug level")
+		flags.String("log-level", "", "use specific log level(debug, info, warn, error), take precedence over verbose flag")
 		flags.Int("pprof-port", -1, "listen port for pprof, 0 represents random port")
 		flags.String("config", "", fmt.Sprintf("the path of configuration file with yaml extension name, default is %s, it can also be set by env var: %s", filepath.Join(dfpath.DefaultConfigDir, rootName+".yaml"), strings.ToUpper(rootName+"_config")))
 

cmd/dfcache/cmd/root.go

@@ -142,7 +142,7 @@ func runDfcacheSubcmd(ctx context.Context, cmdName string, args []string) error
 		MaxBackups: dfcacheConfig.LogMaxBackups}
 
 	// Initialize logger
-	if err := logger.InitDfcache(dfcacheConfig.Console, d.LogDir(), rotateConfig); err != nil {
+	if err := logger.InitDfcache(dfcacheConfig.Verbose, dfcacheConfig.LogLevel, d.LogDir(), rotateConfig); err != nil {
 		return fmt.Errorf("init client dfcache logger: %w", err)
 	}
 	logger.Infof("version:\n%s", version.Version())

cmd/dfget/cmd/daemon.go

@@ -47,8 +47,8 @@ var (
 var daemonCmd = &cobra.Command{
 	Use:   "daemon",
 	Short: "start the client daemon of dragonfly",
-	Long: `client daemon is mainly responsible for transmitting blocks between peers 
+	Long: `client daemon is mainly responsible for transmitting blocks between peers
-and putting the completed file into the specified target path. at the same time, 
+and putting the completed file into the specified target path. at the same time,
 it supports container engine, wget and other downloading tools through proxy function.`,
 	Args:               cobra.NoArgs,
 	DisableAutoGenTag:  true,
@@ -80,7 +80,7 @@ it supports container engine, wget and other downloading tools through proxy fun
 			MaxBackups: cfg.LogMaxBackups}
 
 		// Initialize logger
-		if err := logger.InitDaemon(cfg.Verbose, cfg.Console, d.LogDir(), rotateConfig); err != nil {
+		if err := logger.InitDaemon(cfg.Verbose, cfg.LogLevel, cfg.Console, d.LogDir(), rotateConfig); err != nil {
 			return fmt.Errorf("init client daemon logger: %w", err)
 		}
 		logger.RedirectStdoutAndStderr(cfg.Console, path.Join(d.LogDir(), types.DaemonName))

cmd/dfget/cmd/root.go

@@ -94,7 +94,7 @@ var rootCmd = &cobra.Command{
 			MaxBackups: dfgetConfig.LogMaxBackups}
 
 		// Initialize logger
-		if err := logger.InitDfget(dfgetConfig.Verbose, dfgetConfig.Console, d.LogDir(), rotateConfig); err != nil {
+		if err := logger.InitDfget(dfgetConfig.Verbose, dfgetConfig.LogLevel, dfgetConfig.Console, d.LogDir(), rotateConfig); err != nil {
 			return fmt.Errorf("init client dfget logger: %w", err)
 		}
 

cmd/manager/cmd/root.go

@@ -41,7 +41,7 @@ var (
 var rootCmd = &cobra.Command{
 	Use:   "manager",
 	Short: "The central manager of dragonfly.",
-	Long: `manager is a long-running process and is mainly responsible 
+	Long: `manager is a long-running process and is mainly responsible
 for managing schedulers and seed peers, offering http apis and portal, etc.`,
 	Args:              cobra.NoArgs,
 	DisableAutoGenTag: true,
@@ -72,7 +72,7 @@ for managing schedulers and seed peers, offering http apis and portal, etc.`,
 			MaxBackups: cfg.Server.LogMaxBackups}
 
 		// Initialize logger.
-		if err := logger.InitManager(cfg.Verbose, cfg.Console, d.LogDir(), rotateConfig); err != nil {
+		if err := logger.InitManager(cfg.Verbose, cfg.LogLevel, cfg.Console, d.LogDir(), rotateConfig); err != nil {
 			return fmt.Errorf("init manager logger: %w", err)
 		}
 		logger.RedirectStdoutAndStderr(cfg.Console, path.Join(d.LogDir(), types.ManagerName))

cmd/scheduler/cmd/root.go

@@ -41,7 +41,7 @@ var (
 var rootCmd = &cobra.Command{
 	Use:   "scheduler",
 	Short: "the scheduler of dragonfly",
-	Long: `Scheduler is a long-running process which receives and manages download tasks from the dfdaemon, notify the seed peer to return to the source, 
+	Long: `Scheduler is a long-running process which receives and manages download tasks from the dfdaemon, notify the seed peer to return to the source,
 generate and maintain a P2P network during the download process, and push suitable download nodes to the dfdaemon`,
 	Args:              cobra.NoArgs,
 	DisableAutoGenTag: true,
@@ -72,7 +72,7 @@ generate and maintain a P2P network during the download process, and push suitab
 			MaxBackups: cfg.Server.LogMaxBackups}
 
 		// Initialize logger.
-		if err := logger.InitScheduler(cfg.Verbose, cfg.Console, d.LogDir(), rotateConfig); err != nil {
+		if err := logger.InitScheduler(cfg.Verbose, cfg.LogLevel, cfg.Console, d.LogDir(), rotateConfig); err != nil {
 			return fmt.Errorf("init scheduler logger: %w", err)
 		}
 		logger.RedirectStdoutAndStderr(cfg.Console, path.Join(d.LogDir(), types.SchedulerName))

deploy/docker-compose/template/manager.template.yaml

@@ -165,6 +165,9 @@ console: false
 # Whether to enable debug level logger and enable pprof.
 verbose: true
 
+# Use specific log level(debug, info, warn, error), take precedence over verbose flag.
+# log-level: debug
+
 # Listen port for pprof, only valid when the verbose option is true, default is -1.
 pprof-port: -1
 

deploy/docker-compose/template/scheduler.template.yaml

@@ -181,6 +181,9 @@ console: false
 # Whether to enable debug level logger and enable pprof.
 verbose: true
 
+# Use specific log level(debug, info, warn, error), take precedence over verbose flag.
+# log-level: debug
+
 # Listen port for pprof, only valid when the verbose option is true, default is -1.
 pprof-port: -1
 

internal/dflog/logcore.go

@@ -17,6 +17,7 @@
 package logger
 
 import (
+	"fmt"
 	"strings"
 
 	"go.uber.org/atomic"
@@ -52,7 +53,7 @@ var customCoreLevel atomic.Bool
 var grpcLevel = zap.NewAtomicLevelAt(zapcore.WarnLevel)
 var customGrpcLevel atomic.Bool
 
-func CreateLogger(filePath string, compress bool, stats bool, verbose bool, config LogRotateConfig) (*zap.Logger, zap.AtomicLevel, error) {
+func CreateLogger(filePath string, compress bool, stats bool, verbose bool, logLevel string, config LogRotateConfig) (*zap.Logger, zap.AtomicLevel, error) {
 
 	rotateConfig := &lumberjack.Logger{
 		Filename:   filePath,
@@ -66,10 +67,25 @@ func CreateLogger(filePath string, compress bool, stats bool, verbose bool, conf
 
 	encoderConfig := zap.NewProductionEncoderConfig()
 	encoderConfig.EncodeTime = zapcore.TimeEncoderOfLayout(encodeTimeFormat)
-	var level = zap.NewAtomicLevel()
+	var level = zap.NewAtomicLevelAt(zap.InfoLevel)
-	if verbose {
+	// Use logLevel first, then fallback to verbose flag.
+	if logLevel != "" {
+		switch strings.ToLower(logLevel) {
+		case "debug":
+			level = zap.NewAtomicLevelAt(zapcore.DebugLevel)
+		case "info":
+			level = zap.NewAtomicLevelAt(zapcore.InfoLevel)
+		case "warn":
+			level = zap.NewAtomicLevelAt(zapcore.WarnLevel)
+		case "error":
+			level = zap.NewAtomicLevelAt(zapcore.ErrorLevel)
+		default:
+			fmt.Printf("Warning: invalid log level '%s', using 'info' instead\n", logLevel)
+		}
+	} else if verbose {
 		level = zap.NewAtomicLevelAt(zapcore.DebugLevel)
 	}
+
 	if strings.HasSuffix(filePath, GrpcLogFileName) && customGrpcLevel.Load() {
 		level = grpcLevel
 	} else if strings.HasSuffix(filePath, CoreLogFileName) && customCoreLevel.Load() {

internal/dflog/loginit.go

@@ -17,10 +17,12 @@
 package logger
 
 import (
+	"fmt"
 	"io/fs"
 	"os"
 	"path"
 	"path/filepath"
+	"strings"
 
 	"go.uber.org/zap"
 
@@ -33,9 +35,9 @@ type logInitMeta struct {
 	setLoggerFunc        func(log *zap.Logger)
 }
 
-func InitManager(verbose, console bool, dir string, rotateConfig LogRotateConfig) error {
+func InitManager(verbose bool, logLevel string, console bool, dir string, rotateConfig LogRotateConfig) error {
 	if console {
-		return createConsoleLogger(verbose)
+		return createConsoleLogger(verbose, logLevel)
 	}
 
 	logDir := filepath.Join(dir, types.ManagerName)
@@ -62,12 +64,12 @@ func InitManager(verbose, console bool, dir string, rotateConfig LogRotateConfig
 		},
 	}
 
-	return createFileLogger(verbose, meta, logDir, rotateConfig)
+	return createFileLogger(verbose, logLevel, meta, logDir, rotateConfig)
 }
 
-func InitScheduler(verbose, console bool, dir string, rotateConfig LogRotateConfig) error {
+func InitScheduler(verbose bool, logLevel string, console bool, dir string, rotateConfig LogRotateConfig) error {
 	if console {
-		return createConsoleLogger(verbose)
+		return createConsoleLogger(verbose, logLevel)
 	}
 
 	logDir := filepath.Join(dir, types.SchedulerName)
@@ -90,12 +92,12 @@ func InitScheduler(verbose, console bool, dir string, rotateConfig LogRotateConf
 		},
 	}
 
-	return createFileLogger(verbose, meta, logDir, rotateConfig)
+	return createFileLogger(verbose, logLevel, meta, logDir, rotateConfig)
 }
 
-func InitDaemon(verbose, console bool, dir string, rotateConfig LogRotateConfig) error {
+func InitDaemon(verbose bool, logLevel string, console bool, dir string, rotateConfig LogRotateConfig) error {
 	if console {
-		return createConsoleLogger(verbose)
+		return createConsoleLogger(verbose, logLevel)
 	}
 
 	logDir := filepath.Join(dir, types.DaemonName)
@@ -118,12 +120,12 @@ func InitDaemon(verbose, console bool, dir string, rotateConfig LogRotateConfig)
 		},
 	}
 
-	return createFileLogger(verbose, meta, logDir, rotateConfig)
+	return createFileLogger(verbose, logLevel, meta, logDir, rotateConfig)
 }
 
-func InitDfget(verbose, console bool, dir string, rotateConfig LogRotateConfig) error {
+func InitDfget(verbose bool, logLevel string, console bool, dir string, rotateConfig LogRotateConfig) error {
 	if console {
-		return createConsoleLogger(verbose)
+		return createConsoleLogger(verbose, logLevel)
 	}
 
 	logDir := filepath.Join(dir, types.DfgetName)
@@ -138,10 +140,10 @@ func InitDfget(verbose, console bool, dir string, rotateConfig LogRotateConfig)
 		},
 	}
 
-	return createFileLogger(verbose, meta, logDir, rotateConfig)
+	return createFileLogger(verbose, logLevel, meta, logDir, rotateConfig)
 }
 
-func InitDfcache(console bool, dir string, rotateConfig LogRotateConfig) error {
+func InitDfcache(verbose bool, logLevel string, dir string, rotateConfig LogRotateConfig) error {
 	logDir := filepath.Join(dir, types.DfcacheName)
 	var meta = []logInitMeta{
 		{
@@ -154,16 +156,31 @@ func InitDfcache(console bool, dir string, rotateConfig LogRotateConfig) error {
 		},
 	}
 
-	return createFileLogger(console, meta, logDir, rotateConfig)
+	return createFileLogger(verbose, logLevel, meta, logDir, rotateConfig)
 }
 
-func createConsoleLogger(verbose bool) error {
+func createConsoleLogger(verbose bool, logLevel string) error {
 	levels = nil
 	config := zap.NewDevelopmentConfig()
 	config.Level = zap.NewAtomicLevelAt(zap.InfoLevel)
-	if verbose {
+	// Use logLevel first, then fallback to verbose flag.
+	if logLevel != "" {
+		switch strings.ToLower(logLevel) {
+		case "debug":
+			config.Level = zap.NewAtomicLevelAt(zap.DebugLevel)
+		case "info":
+			config.Level = zap.NewAtomicLevelAt(zap.InfoLevel)
+		case "warn":
+			config.Level = zap.NewAtomicLevelAt(zap.WarnLevel)
+		case "error":
+			config.Level = zap.NewAtomicLevelAt(zap.ErrorLevel)
+		default:
+			fmt.Printf("Warning: invalid log level '%s', using 'info' instead\n", logLevel)
+		}
+	} else if verbose {
 		config.Level = zap.NewAtomicLevelAt(zap.DebugLevel)
 	}
+
 	log, err := config.Build(zap.AddCaller(), zap.AddStacktrace(zap.WarnLevel), zap.AddCallerSkip(1))
 	if err == nil {
 		sugar := log.Sugar()
@@ -182,13 +199,13 @@ func createConsoleLogger(verbose bool) error {
 	return nil
 }
 
-func createFileLogger(verbose bool, meta []logInitMeta, logDir string, rotateConfig LogRotateConfig) error {
+func createFileLogger(verbose bool, logLevel string, meta []logInitMeta, logDir string, rotateConfig LogRotateConfig) error {
 	levels = nil
 	// create parent dir first
 	_ = os.MkdirAll(logDir, fs.FileMode(0700))
 
 	for _, m := range meta {
-		log, level, err := CreateLogger(path.Join(logDir, m.fileName), false, false, verbose, rotateConfig)
+		log, level, err := CreateLogger(path.Join(logDir, m.fileName), false, false, verbose, logLevel, rotateConfig)
 		if err != nil {
 			return err
 		}

pkg/log/log.go

@@ -34,7 +34,7 @@ func SetGrpcLevel(level zapcore.Level) {
 }
 
 // SetupDaemon sets daemon log config: path, console
-func SetupDaemon(logDir string, verbose bool, console bool, rotateConfig logger.LogRotateConfig) error {
+func SetupDaemon(logDir string, verbose bool, logLevel string, console bool, rotateConfig logger.LogRotateConfig) error {
 	var options []dfpath.Option
 	if logDir != "" {
 		options = append(options, dfpath.WithLogDir(logDir))
@@ -45,5 +45,5 @@ func SetupDaemon(logDir string, verbose bool, console bool, rotateConfig logger.
 		return err
 	}
 
-	return logger.InitDaemon(verbose, console, d.LogDir(), rotateConfig)
+	return logger.InitDaemon(verbose, logLevel, console, d.LogDir(), rotateConfig)
 }

scheduler/resource/standard/peer_manager.go

@@ -212,7 +212,7 @@ func (p *peerManager) RunGC(ctx context.Context) error {
 		if elapsed > p.hostTTL {
 			peer.Log.Info("peer elapsed exceeds the host ttl, causing the peer to leave")
 			if err := peer.FSM.Event(ctx, PeerEventLeave); err != nil {
-				peer.Log.Errorf("peer fsm event failed: %s", err.Error())
+				peer.Log.Infof("peer fsm event failed: %s", err.Error())
 				return true
 			}