From 703bfccdc0145c51519b39381dad3b1cc5479b59 Mon Sep 17 00:00:00 2001
From: Marco Primi <marco@mpri.me>
Date: Thu, 23 Jan 2025 14:28:47 -0800
Subject: [PATCH] [POC/WIP/DRAFT] Standard exit code when shutdown via signal

When receiving <signal>, exit with code 128+<signal>.

Pros:
 - Implement de-facto convention
 - Easier integration for any tool/script that expects this convention
 - Easier investigation of "why did server shutdown?" mystery scenarios

Cons:
 - May break some existing integration/tools/script (Hyrum's Law)

Sources:
 - https://tldp.org/LDP/abs/html/exitcodes.html
 - https://en.wikipedia.org/wiki/Exit_status#Shell_and_scripts
 - https://www.baeldung.com/linux/status-codes#fatal-error-signal
---
 main.go                  |  8 +++++---
 server/server.go         | 20 ++++++++++++++------
 server/signal.go         |  6 ++++--
 server/signal_windows.go |  1 +
 4 files changed, 24 insertions(+), 11 deletions(-)

diff --git a/main.go b/main.go
index a85eab97004..b3ec8addef8 100644
--- a/main.go
+++ b/main.go
@@ -108,7 +108,8 @@ func main() {
 		fs.Usage,
 		server.PrintTLSHelpAndDie)
 	if err != nil {
-		server.PrintAndDie(fmt.Sprintf("%s: %s", exe, err))
+		fmt.Fprintln(os.Stderr, fmt.Sprintf("%s: %s", exe, err))
+		os.Exit(1)
 	} else if opts.CheckConfig {
 		fmt.Fprintf(os.Stderr, "%s: configuration file %s is valid (%s)\n", exe, opts.ConfigFile, opts.ConfigDigest())
 		os.Exit(0)
@@ -117,7 +118,8 @@ func main() {
 	// Create the server with appropriate options.
 	s, err := server.NewServer(opts)
 	if err != nil {
-		server.PrintAndDie(fmt.Sprintf("%s: %s", exe, err))
+		fmt.Fprintln(os.Stderr, fmt.Sprintf("%s: %s", exe, err))
+		os.Exit(1)
 	}
 
 	// Configure the logger based on the flags.
@@ -125,7 +127,7 @@ func main() {
 
 	// Start things up. Block here until done.
 	if err := server.Run(s); err != nil {
-		server.PrintAndDie(err.Error())
+		s.PrintAndDie(err.Error())
 	}
 
 	// Adjust MAXPROCS if running under linux/cgroups quotas.
diff --git a/server/server.go b/server/server.go
index ce878d74d36..bd2ecc217e6 100644
--- a/server/server.go
+++ b/server/server.go
@@ -178,6 +178,7 @@ type Server struct {
 	opts                *Options
 	running             atomic.Bool
 	shutdown            atomic.Bool
+	shutdownSignal      atomic.Int32
 	listener            net.Listener
 	listenerErr         error
 	gacc                *Account
@@ -1544,6 +1545,19 @@ func (s *Server) processTrustedKeys() bool {
 	return true
 }
 
+// PrintAndDie print error message and exit
+func (s *Server) PrintAndDie(msg string) {
+
+	sig := s.shutdownSignal.Load()
+	if sig == 0 {
+		fmt.Fprintln(os.Stderr, msg)
+		os.Exit(1)
+	} else {
+		fmt.Fprintln(os.Stderr, fmt.Sprintf("Shutting down due to signal: %d", sig))
+		os.Exit(int(sig + 128))
+	}
+}
+
 // checkTrustedKeyString will check that the string is a valid array
 // of public operator nkeys.
 func checkTrustedKeyString(keys string) []string {
@@ -1576,12 +1590,6 @@ func (s *Server) initStampedTrustedKeys() bool {
 	return true
 }
 
-// PrintAndDie is exported for access in other packages.
-func PrintAndDie(msg string) {
-	fmt.Fprintln(os.Stderr, msg)
-	os.Exit(1)
-}
-
 // PrintServerAndExit will print our version and exit.
 func PrintServerAndExit() {
 	fmt.Printf("nats-server: v%s\n", VERSION)
diff --git a/server/signal.go b/server/signal.go
index aef50a5b4b9..7527f95120e 100644
--- a/server/signal.go
+++ b/server/signal.go
@@ -49,10 +49,12 @@ func (s *Server) handleSignals() {
 				s.Noticef("Trapped %q signal", sig)
 				switch sig {
 				case syscall.SIGINT:
+					s.shutdownSignal.CompareAndSwap(0, int32(syscall.SIGINT))
 					s.Shutdown()
 					s.WaitForShutdown()
-					os.Exit(0)
+					os.Exit(128 + int(syscall.SIGINT))
 				case syscall.SIGTERM:
+					s.shutdownSignal.CompareAndSwap(0, int32(syscall.SIGTERM))
 					// Shutdown unless graceful shutdown already in progress.
 					s.mu.Lock()
 					ldm := s.ldm
@@ -61,7 +63,7 @@ func (s *Server) handleSignals() {
 					if !ldm {
 						s.Shutdown()
 						s.WaitForShutdown()
-						os.Exit(0)
+						os.Exit(128 + int(syscall.SIGTERM))
 					}
 				case syscall.SIGUSR1:
 					// File log re-open for rotating file logs.
diff --git a/server/signal_windows.go b/server/signal_windows.go
index 2f5a27c51d7..1751995bbb7 100644
--- a/server/signal_windows.go
+++ b/server/signal_windows.go
@@ -37,6 +37,7 @@ func (s *Server) handleSignals() {
 		for {
 			select {
 			case sig := <-c:
+				s.shutdownSignal.CompareAndSwap(0, int32(syscall.SIGTERM))
 				s.Debugf("Trapped %q signal", sig)
 				s.Shutdown()
 				os.Exit(0)