fix(ci): fix manager integration test (#1730)

fix(ci): fix manager integration tests

Author: emosbaugh

cmd/manager/start.go

@@ -14,8 +14,6 @@ import (
 
 func StartCmd(ctx context.Context, name string) *cobra.Command {
 	var (
-		dataDir string
-
 		disableWebsocket bool
 	)
 
@@ -24,7 +22,10 @@ func StartCmd(ctx context.Context, name string) *cobra.Command {
 		Short: fmt.Sprintf("Start the %s cluster manager", name),
 		PreRun: func(cmd *cobra.Command, args []string) {
 			// init runtime config and relevant env vars
-			runtimeconfig.ApplyFlags(cmd.Flags())
+			if dataDir := os.Getenv("DATA_DIR"); dataDir != "" {
+				runtimeconfig.SetDataDir(dataDir)
+			}
+
 			os.Setenv("KUBECONFIG", runtimeconfig.PathToKubeConfig())
 			os.Setenv("TMPDIR", runtimeconfig.EmbeddedClusterTmpSubDir())
 		},
@@ -52,8 +53,6 @@ func StartCmd(ctx context.Context, name string) *cobra.Command {
 		},
 	}
 
-	cmd.Flags().StringVar(&dataDir, "data-dir", "", "Path to the data directory")
-
 	// flags to enable running in test mode
 	cmd.Flags().BoolVar(&disableWebsocket, "disable-websocket", false, "When set, don't connect to the KOTS webscoket")
 	cmd.Flags().MarkHidden("disable-websocket")

pkg/manager/install.go

@@ -15,6 +15,8 @@ var (
 	_systemdUnitFileContents []byte
 
 	managerDropInFileContents = `[Service]
+Environment="DATA_DIR=%s"
+
 # Empty ExecStart= will clear out the previous ExecStart value
 ExecStart=
 ExecStart=%s start
@@ -138,6 +140,7 @@ func writeSystemdUnitFile() error {
 func writeDropInFile() error {
 	contents := fmt.Sprintf(
 		managerDropInFileContents,
+		runtimeconfig.EmbeddedClusterHomeDirectory(),
 		runtimeconfig.PathToEmbeddedClusterBinary("manager"),
 	)
 	err := systemd.WriteDropInFile(UnitName(), "embedded-cluster.conf", []byte(contents))

tests/integration/manager/Makefile

@@ -28,7 +28,7 @@ test:
 
 .PHONY: clean
 clean:
-	rm -f manager.test
+	rm -f manager.test ../../../pkg/goods/bins/manager
 	-docker rm -f $(CONTAINER_NAME) 2>/dev/null ; true
 
 .PHONY: build

tests/integration/manager/install_test.go

@@ -4,6 +4,8 @@ import (
 	"context"
 	"os"
 	"os/exec"
+	"path/filepath"
+	"strings"
 	"testing"
 	"time"
 
@@ -13,13 +15,42 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+const (
+	kubeconfigContents = `apiVersion: v1
+kind: Config
+clusters:
+- name: dummy
+  cluster:
+    api-version: v1
+    server: http://example.com
+contexts:
+- name: dummy
+  context:
+    cluster: dummy
+    namespace: dummy
+    user: dummy
+users:
+- name: dummy
+  user:
+    token: dummy
+current-context: dummy
+`
+)
+
 func TestManagerInstall(t *testing.T) {
 	ctx := context.Background()
 
-	runtimeconfig.SetDataDir(getDataDir(t))
+	dataDir := getDataDir(t)
+	runtimeconfig.SetDataDir(dataDir)
+
+	// Write a dummy kubeconfig to the data dir
+	err := os.MkdirAll(filepath.Dir(runtimeconfig.PathToKubeConfig()), 0755)
+	require.NoError(t, err, "failed to create kubeconfig directory")
+	err = os.WriteFile(runtimeconfig.PathToKubeConfig(), []byte(kubeconfigContents), 0644)
+	require.NoError(t, err, "failed to write kubeconfig")
 
 	manager.SetServiceName("ec")
-	err := manager.Install(ctx, t.Logf)
+	err = manager.Install(ctx, t.Logf)
 	require.NoError(t, err, "failed to install manager")
 
 	// Verify service files exists
@@ -28,14 +59,16 @@ func TestManagerInstall(t *testing.T) {
 	dropInDirExists := checkFileExists(t, "/etc/systemd/system/ec-manager.service.d")
 	assert.True(t, dropInDirExists, "ec-manager.service.d drop-in directory should exist")
 
-	// Wait for service to start and become ready
-	// TODO: this should be added to the manager package
-	time.Sleep(5 * time.Second)
-
 	// Verify service is enabled and running
 	status := getServiceStatus(t, "ec-manager.service")
 	assert.Contains(t, status, "enabled", "service should be enabled")
-	assert.Contains(t, status, "active (running)", "service should be running")
+
+	// Wait for service to start and become ready
+	// TODO: this should be added to the manager package
+	assert.Eventually(t, func() bool {
+		status := getServiceStatus(t, "ec-manager.service")
+		return strings.Contains(status, "active (running)")
+	}, 10*time.Second, 1*time.Second, "service should be running")
 
 	err = manager.Uninstall(ctx, t.Logf)
 	require.NoError(t, err, "failed to uninstall manager")