fix(server-manager): kill orphaned cartesi-machine

Adds a workaround for a problem in server-manager where the cartesi-machine keeps running after server-manager exits. For more information, check #201.

internal/services/server-manager.go

@@ -29,6 +29,11 @@ type ServerManager struct {
 
 	// Environment variables.
 	Env []string
+
+	// PID of the Cartesi Machine spawned by the server-manager. Used to manually
+	// stop the orphaned process after server-manager exits. For more information,
+	// check https://github.com/cartesi/server-manager/issues/18
+	machinePid int
 }
 
 const waitDelay = 200 * time.Millisecond
@@ -38,25 +43,39 @@ func (s ServerManager) Start(ctx context.Context, ready chan<- struct{}) error {
 	cmd.Env = s.Env
 	cmd.Stderr = newLineWriter(commandLogger{s.Name})
 	cmd.Stdout = newLineWriter(commandLogger{s.Name})
+	// Without a delay, cmd.Wait() will block forever waiting for the I/O pipes
+	// to be closed
 	cmd.WaitDelay = waitDelay
 	cmd.Cancel = func() error {
 		err := cmd.Process.Signal(syscall.SIGTERM)
 		if err != nil {
-			msg := "failed to send SIGTERM to %v: %v\n"
-			config.WarningLogger.Printf(msg, s, err)
+			config.WarningLogger.Printf("failed to send SIGTERM to %v: %v\n", s, err)
+		}
+		// If we successfully obtained the PID, kill the orphaned cartesi-machine process
+		if s.machinePid != 0 {
+			if err := syscall.Kill(s.machinePid, syscall.SIGTERM); err != nil {
+				config.WarningLogger.Printf("failed to kill cartesi-machine process: %v\n", err)
+			}
 		}
 		return err
 	}
+
+	if err := cmd.Start(); err != nil {
+		return err
+	}
+
 	go s.pollTcp(ctx, ready)
-	err := cmd.Run()
+	go s.storeCartesiMachinePid(ctx, cmd.Process.Pid)
+
+	err := cmd.Wait()
 
 	if ctx.Err() != nil {
 		return ctx.Err()
 	}
 	return err
 }
 
-// Blocks until the service is ready or the context is canceled.
+// Blocks until the service is ready or the context is canceled
 func (s ServerManager) pollTcp(ctx context.Context, ready chan<- struct{}) {
 	ctx, cancel := context.WithCancel(ctx)
 	defer cancel()
@@ -76,6 +95,40 @@ func (s ServerManager) pollTcp(ctx context.Context, ready chan<- struct{}) {
 	}
 }
 
+// Blocks until it successfully stores Cartesi Machine's PID,
+// the context is canceled or it times out, whichever comes first
+func (s *ServerManager) storeCartesiMachinePid(ctx context.Context, ppid int) {
+	ctx, cancel := context.WithTimeout(ctx, DefaultServiceTimeout)
+	defer cancel()
+	for {
+		childPid, err := getChildPid(ppid)
+		if err == nil {
+			config.DebugLogger.Println("stored cartesi-machine PID")
+			s.machinePid = childPid
+			return
+		}
+		select {
+		case <-ctx.Done():
+			config.DebugLogger.Println("failed to store cartesi-machine PID")
+			return
+		case <-time.After(time.Second):
+		}
+	}
+}
+
 func (s ServerManager) String() string {
 	return s.Name
 }
+
+func getChildPid(pid int) (int, error) {
+	output, err := exec.Command("pgrep", "-P", fmt.Sprint(pid)).CombinedOutput()
+	if err != nil {
+		return 0, fmt.Errorf("failed to exec pgrep: %v: %v", err, string(output))
+	}
+	var childPid int
+	_, err = fmt.Sscanf(string(output), "%d\n", &childPid)
+	if err != nil {
+		return 0, fmt.Errorf("failed to parse pid: %v", err)
+	}
+	return childPid, err
+}