Switch to slog, remove logrus (#1672)

cmd/nebula-service/logs_generic.go

@@ -3,8 +3,15 @@
 
 package main
 
-import "github.com/sirupsen/logrus"
+import (
+	"log/slog"
+	"os"
 
-func HookLogger(l *logrus.Logger) {
-	// Do nothing, let the logs flow to stdout/stderr
+	"github.com/slackhq/nebula/logging"
+)
+
+// newPlatformLogger returns a *slog.Logger that writes to stdout. Non-Windows
+// platforms have no special sink to integrate with.
+func newPlatformLogger() *slog.Logger {
+	return logging.NewLogger(os.Stdout)
 }

cmd/nebula-service/logs_windows.go

@@ -1,54 +1,86 @@
 package main
 
 import (
-	"fmt"
-	"io/ioutil"
-	"os"
+	"context"
+	"log/slog"
+	"strings"
+	"sync"
 
-	"github.com/kardianos/service"
-	"github.com/sirupsen/logrus"
+	"github.com/slackhq/nebula/logging"
 )
 
-// HookLogger routes the logrus logs through the service logger so that they end up in the Windows Event Viewer
-// logrus output will be discarded
-func HookLogger(l *logrus.Logger) {
-	l.AddHook(newLogHook(logger))
-	l.SetOutput(ioutil.Discard)
+// newPlatformLogger returns a *slog.Logger that routes every log record
+// through the Windows service logger so records end up in the Windows
+// Event Log. All the heavy lifting (level management, format swap,
+// timestamp toggle, WithAttrs/WithGroup) comes from logging.NewHandler;
+// this file only contributes:
+//
+//   - an io.Writer that forwards each formatted line to the service
+//     logger at the current record's Event Log severity, and
+//   - a thin severityTag that embeds *logging.Handler and overrides
+//     only Handle / WithAttrs / WithGroup, so Event Viewer's severity
+//     column and severity-based filters keep working the way they did
+//     before the slog migration.
+//
+// Format (text vs json) is carried by the embedded *logging.Handler, so
+// logging.format: json in config still produces JSON lines in Event
+// Viewer, same as the pre-slog logrus setup.
+func newPlatformLogger() *slog.Logger {
+	w := &eventLogWriter{}
+	return slog.New(&severityTag{Handler: logging.NewHandler(w), w: w})
 }
 
-type logHook struct {
-	sl service.Logger
+// eventLogWriter forwards slog-formatted lines to the Windows service
+// logger at the severity most recently stashed by severityTag.Handle.
+// The mutex serializes the stash + inner.Handle + Write cycle per record
+// across all concurrent goroutines; slog's builtin text/json handlers
+// each hold their own mutex around Write, but that only protects the
+// Write call itself, not our stash-then-handle sequence.
+type eventLogWriter struct {
+	mu    sync.Mutex
+	level slog.Level
 }
 
-func newLogHook(sl service.Logger) *logHook {
-	return &logHook{sl: sl}
-}
-
-func (h *logHook) Fire(entry *logrus.Entry) error {
-	line, err := entry.String()
-	if err != nil {
-		fmt.Fprintf(os.Stderr, "Unable to read entry, %v", err)
-		return err
-	}
-
-	switch entry.Level {
-	case logrus.PanicLevel:
-		return h.sl.Error(line)
-	case logrus.FatalLevel:
-		return h.sl.Error(line)
-	case logrus.ErrorLevel:
-		return h.sl.Error(line)
-	case logrus.WarnLevel:
-		return h.sl.Warning(line)
-	case logrus.InfoLevel:
-		return h.sl.Info(line)
-	case logrus.DebugLevel:
-		return h.sl.Info(line)
+func (w *eventLogWriter) Write(p []byte) (int, error) {
+	line := strings.TrimRight(string(p), "\n")
+	switch {
+	case w.level >= slog.LevelError:
+		return len(p), logger.Error(line)
+	case w.level >= slog.LevelWarn:
+		return len(p), logger.Warning(line)
 	default:
-		return nil
+		return len(p), logger.Info(line)
 	}
 }
 
-func (h *logHook) Levels() []logrus.Level {
-	return logrus.AllLevels
+// severityTag embeds *logging.Handler to pick up everything it does for
+// free (Enabled, SetLevel, GetLevel, SetFormat, GetFormat,
+// SetDisableTimestamp) and overrides only Handle / WithAttrs / WithGroup
+// so each record's slog.Level is stashed on the writer before formatting
+// and so derived handlers stay wrapped as severityTag rather than
+// downgrading to bare *logging.Handler.
+type severityTag struct {
+	*logging.Handler
+	w *eventLogWriter
+}
+
+func (s *severityTag) Handle(ctx context.Context, r slog.Record) error {
+	s.w.mu.Lock()
+	defer s.w.mu.Unlock()
+	s.w.level = r.Level
+	return s.Handler.Handle(ctx, r)
+}
+
+func (s *severityTag) WithAttrs(attrs []slog.Attr) slog.Handler {
+	if len(attrs) == 0 {
+		return s
+	}
+	return &severityTag{Handler: s.Handler.WithAttrs(attrs).(*logging.Handler), w: s.w}
+}
+
+func (s *severityTag) WithGroup(name string) slog.Handler {
+	if name == "" {
+		return s
+	}
+	return &severityTag{Handler: s.Handler.WithGroup(name).(*logging.Handler), w: s.w}
 }

cmd/nebula-service/main.go

@@ -7,9 +7,9 @@ import (
 	"runtime/debug"
 	"strings"
 
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula"
 	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/logging"
 	"github.com/slackhq/nebula/util"
 )
 
@@ -50,12 +50,11 @@ func main() {
 		os.Exit(0)
 	}
 
-	l := logrus.New()
-	l.Out = os.Stdout
+	l := logging.NewLogger(os.Stdout)
 
 	if *serviceFlag != "" {
 		if err := doService(configPath, configTest, Build, serviceFlag); err != nil {
-			l.WithError(err).Error("Service command failed")
+			l.Error("Service command failed", "error", err)
 			os.Exit(1)
 		}
 		return
@@ -74,6 +73,16 @@ func main() {
 		os.Exit(1)
 	}
 
+	if err := logging.ApplyConfig(l, c); err != nil {
+		fmt.Printf("failed to apply logging config: %s", err)
+		os.Exit(1)
+	}
+	c.RegisterReloadCallback(func(c *config.C) {
+		if err := logging.ApplyConfig(l, c); err != nil {
+			l.Error("Failed to reconfigure logger on reload", "error", err)
+		}
+	})
+
 	ctrl, err := nebula.Main(c, *configTest, Build, l, nil)
 	if err != nil {
 		util.LogWithContextIfNeeded("Failed to start", err, l)
@@ -90,7 +99,7 @@ func main() {
 		go ctrl.ShutdownBlock()
 
 		if err := wait(); err != nil {
-			l.WithError(err).Error("Nebula stopped due to fatal error")
+			l.Error("Nebula stopped due to fatal error", "error", err)
 			os.Exit(2)
 		}
 

cmd/nebula-service/service.go

@@ -7,9 +7,9 @@ import (
 	"path/filepath"
 
 	"github.com/kardianos/service"
-	"github.com/sirupsen/logrus"
 	"github.com/slackhq/nebula"
 	"github.com/slackhq/nebula/config"
+	"github.com/slackhq/nebula/logging"
 )
 
 var logger service.Logger
@@ -25,8 +25,7 @@ func (p *program) Start(s service.Service) error {
 	// Start should not block.
 	logger.Info("Nebula service starting.")
 
-	l := logrus.New()
-	HookLogger(l)
+	l := newPlatformLogger()
 
 	c := config.NewC(l)
 	err := c.Load(*p.configPath)
@@ -34,6 +33,15 @@ func (p *program) Start(s service.Service) error {
 		return fmt.Errorf("failed to load config: %s", err)
 	}
 
+	if err := logging.ApplyConfig(l, c); err != nil {
+		return fmt.Errorf("failed to apply logging config: %s", err)
+	}
+	c.RegisterReloadCallback(func(c *config.C) {
+		if err := logging.ApplyConfig(l, c); err != nil {
+			l.Error("Failed to reconfigure logger on reload", "error", err)
+		}
+	})
+
 	p.control, err = nebula.Main(c, *p.configTest, Build, l, nil)
 	if err != nil {
 		return err
@@ -85,7 +93,7 @@ func doService(configPath *string, configTest *bool, build string, serviceFlag *
 	// Here are what the different loggers are doing:
 	// - `log` is the standard go log utility, meant to be used while the process is still attached to stdout/stderr
 	// - `logger` is the service log utility that may be attached to a special place depending on OS (Windows will have it attached to the event log)
-	// - above, in `Run` we create a `logrus.Logger` which is what nebula expects to use
+	// - in program.Start we build a *slog.Logger via newPlatformLogger; on non-Windows that is a stdout-backed slog logger, on Windows it routes records through the service logger
 	s, err := service.New(prg, svcConfig)
 	if err != nil {
 		return err