feat: add host preflights for all needed ports

cmd/embedded-cluster/reset.go

@@ -365,11 +365,6 @@ var resetCommand = &cli.Command{
 			Usage: "Disable interactive prompts",
 			Value: false,
 		},
-		&cli.BoolFlag{
-			Name:  "reboot",
-			Usage: "Reboot system after resetting the node",
-			Value: false,
-		},
 	},
 	Usage: fmt.Sprintf("Remove %s from the current node", binName),
 	Action: func(c *cli.Context) error {
@@ -378,7 +373,7 @@ var resetCommand = &cli.Command{
 		}
 
 		logrus.Info("This will remove this node from the cluster and completely reset it, removing all data stored on the node.")
-		logrus.Info("Do not reset another node until this is complete.")
+		logrus.Info("This action will cause the node to reboot. Do not reset another node until this is complete.")
 		if !c.Bool("force") && !c.Bool("no-prompt") && !prompts.New().Confirm("Do you want to continue?", false) {
 			return fmt.Errorf("Aborting")
 		}
@@ -449,10 +444,6 @@ var resetCommand = &cli.Command{
 			return err
 		}
 
-		if !c.Bool("reboot") {
-			logrus.Infof("Node has been reset. Please reboot to ensure transient configuration is also reset.")
-		}
-
 		if err := helpers.RemoveAll(defaults.PathToK0sConfig()); err != nil {
 			return fmt.Errorf("failed to remove k0s config: %w", err)
 		}
@@ -501,10 +492,8 @@ var resetCommand = &cli.Command{
 			return fmt.Errorf("failed to remove k0s binary: %w", err)
 		}
 
-		if c.Bool("reboot") {
-			if _, err := exec.Command("reboot").Output(); err != nil {
-				return err
-			}
+		if _, err := exec.Command("reboot").Output(); err != nil {
+			return err
 		}
 
 		return nil

e2e/cluster/cluster.go

@@ -15,6 +15,7 @@ import (
 	lxd "github.com/canonical/lxd/client"
 	"github.com/canonical/lxd/shared/api"
 	"github.com/google/uuid"
+	"github.com/stretchr/testify/require"
 )
 
 var networkaddr chan string
@@ -131,6 +132,26 @@ func (o *Output) Destroy() {
 	networkaddr <- o.network
 }
 
+// WaitForNodeRunning waits until the node is in "running" state.
+func (o *Output) WaitForNodeRunning(node int) {
+	client, err := lxd.ConnectLXDUnix(lxdSocket, nil)
+	require.NoError(o.T, err, "failed to connect to lxd")
+
+	var counter int
+	state := &api.InstanceState{}
+	for state.Status != "Running" {
+		time.Sleep(5 * time.Second)
+		o.T.Logf("waiting for node %d to start (running)", node)
+
+		state, _, err = client.GetInstanceState(o.Nodes[node])
+		require.NoError(o.T, err, "failed to get node state %d", node)
+		o.T.Logf("node %d is in state %q", node, state.Status)
+
+		require.True(o.T, counter < 10, "node %d failed to start", node)
+		counter++
+	}
+}
+
 // Command is a command to be run in a node.
 type Command struct {
 	Node        string

e2e/install_test.go

@@ -624,6 +624,10 @@ func TestResetAndReinstall(t *testing.T) {
 		t.Fatalf("fail to reset the installation: %v", err)
 	}
 
+	// reset will reboot the node, this waits until the node reports back
+	// as running.
+	tc.WaitForNodeRunning(0)
+
 	t.Logf("%s: installing embedded-cluster on node 0 after reset", time.Now().Format(time.RFC3339))
 	line = []string{"single-node-install.sh", "ui"}
 	if _, _, err := RunCommandOnNode(t, tc, 0, line); err != nil {
@@ -698,6 +702,10 @@ func TestResetAndReinstallAirgap(t *testing.T) {
 		t.Fatalf("fail to reset the installation: %v", err)
 	}
 
+	// reset will reboot the node, this waits until the node reports back
+	// as running.
+	tc.WaitForNodeRunning(0)
+
 	t.Logf("%s: installing embedded-cluster on node 0", time.Now().Format(time.RFC3339))
 	line = []string{"single-node-airgap-install.sh"}
 	if _, _, err := RunCommandOnNode(t, tc, 0, line); err != nil {

e2e/preflights_test.go

@@ -29,12 +29,24 @@ func TestPreflights(t *testing.T) {
 	})
 
 	_, stderr, err := container.Exec(cli,
-		"apt-get update && apt-get install -y apt-utils kmod",
+		"apt-get update && apt-get install -y apt-utils kmod netcat-traditional",
 	)
 	if err != nil {
 		t.Fatalf("failed to install deps: err=%v, stderr=%s", err, stderr)
 	}
 
+	if _, stderr, err = container.Exec(cli, "nohup netcat -l -p 10250 &"); err != nil {
+		t.Fatalf("failed to start netcat: err=%v, stderr=%s", err, stderr)
+	}
+
+	if _, stderr, err = container.Exec(cli, "nohup netcat -l 127.0.0.1 -p 50000 &"); err != nil {
+		t.Fatalf("failed to start netcat: err=%v, stderr=%s", err, stderr)
+	}
+
+	if _, stderr, err = container.Exec(cli, "nohup netcat -l -u -p 4789 &"); err != nil {
+		t.Fatalf("failed to start netcat: err=%v, stderr=%s", err, stderr)
+	}
+
 	runCmd := fmt.Sprintf("%s install run-preflights --no-prompt", container.GetECBinaryPath())
 	if os.Getenv("LICENSE_PATH") != "" {
 		runCmd = fmt.Sprintf("%s --license %s", runCmd, container.GetLicensePath())
@@ -93,10 +105,13 @@ func TestPreflights(t *testing.T) {
 			assert: func(t *testing.T, results *preflights.Output) {
 				expected := map[string]bool{
 					// TODO: work to remove these
-					"System Clock":                true,
-					"'devices' Cgroup Controller": true,
-					"API Access":                  true,
-					"Proxy Registry Access":       true,
+					"System Clock":                            true,
+					"'devices' Cgroup Controller":             true,
+					"API Access":                              true,
+					"Proxy Registry Access":                   true,
+					"Kubelet Port Availability":               true,
+					"Calico Communication Port Availability":  true,
+					"Local Artifact Mirror Port Availability": true,
 					// as long as fio ran successfully, we're good
 					"Filesystem Write Latency": true,
 				}
@@ -124,6 +139,26 @@ func TestPreflights(t *testing.T) {
 				}
 			},
 		},
+		{
+			name: "Should contain port failures",
+			assert: func(t *testing.T, results *preflights.Output) {
+				expected := map[string]bool{
+					"Kubelet Port Availability":               false,
+					"Calico Communication Port Availability":  false,
+					"Local Artifact Mirror Port Availability": false,
+				}
+				for _, res := range results.Fail {
+					if _, ok := expected[res.Title]; ok {
+						expected[res.Title] = true
+					}
+				}
+				for title, found := range expected {
+					if !found {
+						t.Errorf("expected port failure not found: %q", title)
+					}
+				}
+			},
+		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {