Not fail csi-init on network problems (#2042)

src/cmd/csi/init/builder.go

@@ -10,6 +10,7 @@ import (
 	"github.com/spf13/cobra"
 	"golang.org/x/sys/unix"
 	ctrl "sigs.k8s.io/controller-runtime"
+	"sigs.k8s.io/controller-runtime/pkg/client"
 )
 
 const use = "csi-init"
@@ -56,7 +57,7 @@ func (builder CommandBuilder) buildRun() func(*cobra.Command, []string) error {
 
 		csiManager, err := createManager(builder.namespace, kubeConfig)
 		if err != nil {
-			return err
+			log.Info("failed to create/configure kubernetes client, will only run non-network related corrections and checks", "err", err.Error())
 		}
 
 		err = createCsiDataPath(afero.NewOsFs())
@@ -76,7 +77,11 @@ func (builder CommandBuilder) buildRun() func(*cobra.Command, []string) error {
 			RootDir:  dtcsi.DataPath,
 		}
 
-		err = metadata.NewCorrectnessChecker(csiManager.GetClient(), access, csiOptions).CorrectCSI(signalHandler)
+		var apiReader client.Reader
+		if csiManager != nil {
+			apiReader = csiManager.GetAPIReader()
+		}
+		err = metadata.NewCorrectnessChecker(apiReader, access, csiOptions).CorrectCSI(signalHandler)
 		if err != nil {
 			return err
 		}

src/cmd/csi/init/config.go

@@ -0,0 +1,5 @@
+package init
+
+import "github.com/Dynatrace/dynatrace-operator/src/logger"
+
+var log = logger.Factory.GetLogger("csi-init")

src/controllers/csi/metadata/correctness.go

@@ -15,18 +15,18 @@ import (
 )
 
 type CorrectnessChecker struct {
-	cl     client.Client
-	fs     afero.Fs
-	path   PathResolver
-	access Access
+	apiReader client.Reader
+	fs        afero.Fs
+	path      PathResolver
+	access    Access
 }
 
-func NewCorrectnessChecker(cl client.Client, access Access, opts dtcsi.CSIOptions) *CorrectnessChecker {
+func NewCorrectnessChecker(cl client.Reader, access Access, opts dtcsi.CSIOptions) *CorrectnessChecker {
 	return &CorrectnessChecker{
-		cl:     cl,
-		fs:     afero.NewOsFs(),
-		path:   PathResolver{RootDir: opts.RootDir},
-		access: access,
+		apiReader: cl,
+		fs:        afero.NewOsFs(),
+		path:      PathResolver{RootDir: opts.RootDir},
+		access:    access,
 	}
 }
 
@@ -49,14 +49,18 @@ func (checker *CorrectnessChecker) CorrectCSI(ctx context.Context) error {
 
 // Removes volume entries if their pod is no longer exists
 func (checker *CorrectnessChecker) removeVolumesForMissingPods(ctx context.Context) error {
+	if checker.apiReader == nil {
+		log.Info("no kubernetes client configured, skipping orphaned volume metadata cleanup")
+		return nil
+	}
 	podNames, err := checker.access.GetPodNames(ctx)
 	if err != nil {
 		return err
 	}
 	pruned := []string{}
 	for podName := range podNames {
 		var pod corev1.Pod
-		if err := checker.cl.Get(context.TODO(), client.ObjectKey{Name: podName}, &pod); !k8serrors.IsNotFound(err) {
+		if err := checker.apiReader.Get(ctx, client.ObjectKey{Name: podName}, &pod); !k8serrors.IsNotFound(err) {
 			continue
 		}
 		volumeID := podNames[podName]
@@ -71,14 +75,18 @@ func (checker *CorrectnessChecker) removeVolumesForMissingPods(ctx context.Conte
 
 // Removes dynakube entries if their Dynakube instance no longer exists in the cluster
 func (checker *CorrectnessChecker) removeMissingDynakubes(ctx context.Context) error {
+	if checker.apiReader == nil {
+		log.Info("no kubernetes client configured, skipping orphaned dynakube metadata cleanup")
+		return nil
+	}
 	dynakubes, err := checker.access.GetTenantsToDynakubes(ctx)
 	if err != nil {
 		return err
 	}
 	pruned := []string{}
 	for dynakubeName := range dynakubes {
 		var dynakube dynatracev1beta1.DynaKube
-		if err := checker.cl.Get(context.TODO(), client.ObjectKey{Name: dynakubeName}, &dynakube); !k8serrors.IsNotFound(err) {
+		if err := checker.apiReader.Get(ctx, client.ObjectKey{Name: dynakubeName}, &dynakube); !k8serrors.IsNotFound(err) {
 			continue
 		}
 		if err := checker.access.DeleteDynakube(ctx, dynakubeName); err != nil {

src/controllers/csi/metadata/correctness_test.go

@@ -54,12 +54,36 @@ func TestCorrectCSI(t *testing.T) {
 		assert.NoError(t, err)
 	})
 
+	t.Run("no error on nil apiReader, database is not cleaned", func(t *testing.T) {
+		ctx := context.TODO()
+		testVolume1 := createTestVolume(1)
+		testDynakube1 := createTestDynakube(1)
+		db := FakeMemoryDB()
+		db.InsertVolume(ctx, &testVolume1)
+		db.InsertDynakube(ctx, &testDynakube1)
+
+		checker := NewCorrectnessChecker(nil, db, dtcsi.CSIOptions{})
+
+		err := checker.CorrectCSI(context.TODO())
+
+		assert.NoError(t, err)
+		vol, err := db.GetVolume(ctx, testVolume1.VolumeID)
+		assert.NoError(t, err)
+		assert.Equal(t, &testVolume1, vol)
+
+		assert.NoError(t, err)
+		dk, err := db.GetDynakube(ctx, testDynakube1.Name)
+		assert.NoError(t, err)
+		assert.Equal(t, &testDynakube1, dk)
+	})
+
 	t.Run("nothing to remove, everything is still correct", func(t *testing.T) {
 		ctx := context.TODO()
 		testVolume1 := createTestVolume(1)
 		testDynakube1 := createTestDynakube(1)
 		db := FakeMemoryDB()
 		db.InsertVolume(ctx, &testVolume1)
+		db.InsertDynakube(ctx, &testDynakube1)
 		client := fake.NewClient(
 			&corev1.Pod{ObjectMeta: metav1.ObjectMeta{Name: testVolume1.PodName}},
 			&dynatracev1beta1.DynaKube{ObjectMeta: metav1.ObjectMeta{Name: testDynakube1.Name}},
@@ -73,6 +97,11 @@ func TestCorrectCSI(t *testing.T) {
 		vol, err := db.GetVolume(ctx, testVolume1.VolumeID)
 		assert.NoError(t, err)
 		assert.Equal(t, &testVolume1, vol)
+
+		assert.NoError(t, err)
+		dk, err := db.GetDynakube(ctx, testDynakube1.Name)
+		assert.NoError(t, err)
+		assert.Equal(t, &testDynakube1, dk)
 	})
 	t.Run("remove unnecessary entries in the filesystem", func(t *testing.T) {
 		ctx := context.TODO()

test/testdata/sample-app/restricted-csi.yaml

@@ -1,7 +1,7 @@
 apiVersion: security.openshift.io/v1
 kind: SecurityContextConstraints
 metadata:
-  name: restricted-csi
+  name: operator-e2e-restricted-csi
 runAsUser:
   type: RunAsAny
 seLinuxContext:
@@ -30,3 +30,5 @@ volumes:
 - emptyDir
 - secret
 - csi
+- projected
+- downwardAPI