daemon.go

@@ -7,18 +7,15 @@ import (
 	"os"
 	"os/signal"
 	"path/filepath"
+	"sync"
 	"syscall"
+	"time"
 )
 
 type CommandPayload struct {
 	Command string `json:"command"`
 }
 
-type DaemonServer struct {
-	cfg         *AppConfig
-	procManager *ProcessManager
-}
-
 type InstanceStatusResponse struct {
 	Name    string `json:"name"`
 	Version string `json:"version"`
@@ -26,6 +23,11 @@ type InstanceStatusResponse struct {
 	Status  string `json:"status"`
 }
 
+type DaemonServer struct {
+	cfg         *AppConfig
+	procManager *ProcessManager
+}
+
 func StartDaemon(cfg *AppConfig) error {
 	ds := &DaemonServer{
 		cfg:         cfg,
@@ -50,15 +52,73 @@ func StartDaemon(cfg *AppConfig) error {
 	go func() {
 		fmt.Printf("Engine daemon actively listening on http://%s\n", cfg.Daemon.ListenAddress)
 		if err := server.ListenAndServe(); err != http.ErrServerClosed {
-			fmt.Errorf("Daemon runtime failure: %v", err)
+			fmt.Printf("Daemon runtime failure: %v\n", err)
 		}
 	}()
 
 	<-sigChan
-	fmt.Println("\nShutting down supervisor daemon threads...")
+	fmt.Println("\n[Daemon] Shutdown signal caught! Initializing graceful teardown sequence...")
+	_ = server.Close()
+	ds.shutdownAllRunningServers()
+	fmt.Println("[Daemon] All threads gracefully shut down. Exiting supervisor cleanly.")
 	return nil
 }
 
+func (ds *DaemonServer) shutdownAllRunningServers() {
+	ds.procManager.Lock()
+
+	var activeNames []string
+	for name := range ds.procManager.ActiveInstances {
+		activeNames = append(activeNames, name)
+	}
+	ds.procManager.Unlock()
+
+	if len(activeNames) == 0 {
+		fmt.Println("[Daemon] No active server instances to tear down.")
+		return
+	}
+
+	fmt.Printf("[Daemon] Flushing stop instructions to %d running instance(s)...\n", len(activeNames))
+
+	var wg sync.WaitGroup
+	for _, name := range activeNames {
+		wg.Add(1)
+		go func(instanceName string) {
+			defer wg.Done()
+
+			fmt.Printf("[Daemon] Sending graceful /stop to instance '%s'...\n", instanceName)
+			err := ds.procManager.SendCommand(instanceName, "/stop")
+			if err != nil {
+				fmt.Printf("[Daemon Error] Could not send stop to %s: %v\n", instanceName, err)
+				return
+			}
+
+			ticker := time.NewTicker(250 * time.Millisecond)
+			defer ticker.Stop()
+			timeout := time.After(15 * time.Second)
+
+			for {
+				select {
+				case <-ticker.C:
+					ds.procManager.RLock()
+					_, running := ds.procManager.ActiveInstances[instanceName]
+					ds.procManager.RUnlock()
+
+					if !running {
+						fmt.Printf("[Daemon] Instance '%s' has successfully exited.\n", instanceName)
+						return
+					}
+				case <-timeout:
+					fmt.Printf("[Daemon Warning] Instance '%s' timed out while trying to stop safely.\n", instanceName)
+					return
+				}
+			}
+		}(name)
+	}
+
+	wg.Wait()
+}
+
 func (ds *DaemonServer) handleCreate(w http.ResponseWriter, r *http.Request) {
 	if r.Method != http.MethodPost {
 		http.Error(w, "Method not allowed", http.StatusMethodNotAllowed)