fix(core): exit non-zero when nylon stops on a fatal error

core/nylon.go

@@ -329,6 +329,8 @@ func (n *Nylon) Init() error {
 	return nil
 }
 
+var errShutdownSignal = errors.New("received shutdown signal")
+
 func (n *Nylon) Start() error {
 	n.Log.Info("init modules complete")
 
@@ -343,7 +345,7 @@ func (n *Nylon) Start() error {
 	go func() {
 		select {
 		case _ = <-c:
-			n.Cancel(errors.New("received shutdown signal"))
+			n.Cancel(errShutdownSignal)
 		case <-n.Context.Done():
 			return
 		}
@@ -402,7 +404,12 @@ endLoop:
 		n.Log.Error("error occurred during Stop: ", "error", err)
 	}
 	n.Log.Info("stopped")
-	return nil
+	// report fatal errors so the process exits non-zero
+	cause := context.Cause(n.Context)
+	if errors.Is(cause, context.Canceled) || errors.Is(cause, errShutdownSignal) {
+		return nil
+	}
+	return cause
 }
 
 func (n *Nylon) Cleanup() error {

integration/harness.go

@@ -75,6 +75,9 @@ type Harness struct {
 	// UntrackedRouting turns off UseSystemRouting on every node, so packets
 	// for peers stay inside polyamide instead of coming out of the TUN.
 	UntrackedRouting bool
+	// OnStartError, if set, receives errors returned by a node's Start
+	// instead of failing the test.
+	OnStartError func(id state.NodeId, err error)
 
 	ctx    context.Context
 	cancel context.CancelCauseFunc
@@ -238,7 +241,11 @@ func (h *Harness) launch(nd *node, n *core.Nylon, delay time.Duration) {
 		}
 		pprof.Do(context.Background(), pprof.Labels("nylon node", string(n.LocalCfg.Id)), func(context.Context) {
 			if err := n.Start(); err != nil {
-				h.fail("node %s: %v", n.LocalCfg.Id, err)
+				if h.OnStartError != nil {
+					h.OnStartError(n.LocalCfg.Id, err)
+				} else {
+					h.fail("node %s: %v", n.LocalCfg.Id, err)
+				}
 			}
 		})
 	}()

integration/ipc_test.go

@@ -279,3 +279,23 @@ func TestIPCConcurrentRequestsAndCancel(t *testing.T) {
 		require.Less(t, time.Since(start), 4*time.Second, "the cancelled probe should end before its timeout")
 	})
 }
+
+// A node stopped by a fatal error must report it so nylon run exits non-zero.
+func TestStartReturnsFatalDispatchError(t *testing.T) {
+	synctest.Test(t, func(t *testing.T) {
+		h := NewHarness(t)
+		errs := make(chan error, 2)
+		h.OnStartError = func(_ state.NodeId, err error) { errs <- err }
+		a1, b1 := "192.168.54.1:1234", "192.168.54.2:1234"
+		h.AddNode("a", "10.0.0.1/32", a1)
+		h.AddNode("b", "10.0.0.2/32", b1)
+		h.Central.Graph = []string{"a, b"}
+		h.Connect(a1, b1)
+		h.Start()
+		time.Sleep(time.Second) // let both nodes start
+
+		h.Nylon("a").Dispatch(func() error { return errors.New("fatal") })
+		err := await(h, errs, 5*time.Second, "a's Start to return")
+		require.ErrorContains(t, err, "fatal")
+	})
+}