Fix transient condition when device path is not yet available on host (…

…#415)

We currently fail with an internal error if the device path is not ready
on the host yet. The error occurs when we try to mount the volume. This
leads to unnecessary error events.

In this PR we try to address this issue by return an `Unavailable` error
in case the device on the host is not ready yet. This will trigger a
re-try.

hack/license-header.txt

@@ -1,2 +1,2 @@
-SPDX-FileCopyrightText: 2023 SAP SE or an SAP affiliate company and IronCore contributors
+SPDX-FileCopyrightText: 2024 SAP SE or an SAP affiliate company and IronCore contributors
 SPDX-License-Identifier: Apache-2.0

pkg/driver/node.go

@@ -7,6 +7,7 @@ import (
 	"context"
 	"fmt"
 	"io"
+	"os"
 	"path/filepath"
 
 	"github.com/container-storage-interface/spec/lib/go/csi"
@@ -35,6 +36,16 @@ func (d *driver) NodeStageVolume(_ context.Context, req *csi.NodeStageVolumeRequ
 	fstype := req.GetVolumeContext()["fstype"]
 	devicePath := req.PublishContext["device_name"]
 
+	klog.InfoS("Check if the device path exists")
+	if _, err := d.os.Stat(devicePath); err != nil {
+		if os.IsNotExist(err) {
+			// We will requeue here since the device path is not ready yet.
+			return nil, status.Errorf(codes.Unavailable, "Device path %s does not exist: %v", devicePath, err)
+		} else {
+			return nil, status.Errorf(codes.Internal, "Failed to determine wether the device path %s exists: %v", devicePath, err)
+		}
+	}
+
 	readOnly := false
 	if req.GetVolumeContext()["readOnly"] == "true" {
 		readOnly = true
@@ -45,7 +56,7 @@ func (d *driver) NodeStageVolume(_ context.Context, req *csi.NodeStageVolumeRequ
 	klog.InfoS("Validate mount point", "MountPoint", targetPath)
 	notMnt, err := d.mounter.IsLikelyNotMountPoint(targetPath)
 	if err != nil && !d.os.IsNotExist(err) {
-		return nil, status.Errorf(codes.Internal, "Failed to verify mount point %s: %v", devicePath, err)
+		return nil, status.Errorf(codes.Internal, "Failed to verify mount point %s: %v", targetPath, err)
 	}
 	klog.InfoS("Check if volume is already mounted")
 	if !notMnt {

pkg/driver/node_test.go

@@ -79,6 +79,8 @@ var _ = Describe("Node", func() {
 					},
 				},
 			}
+
+			mockOS.EXPECT().Stat(devicePath).Return(nil, nil)
 		})
 
 		It("should fail if the volume is already mounted", func(ctx SpecContext) {
@@ -107,7 +109,6 @@ var _ = Describe("Node", func() {
 			statusErr, ok := status.FromError(err)
 			Expect(ok).To(BeTrue())
 			Expect(statusErr.Code()).To(Equal(codes.Internal))
-
 		})
 
 		It("should fail if the mount operation fails", func(ctx SpecContext) {
@@ -129,7 +130,6 @@ var _ = Describe("Node", func() {
 			_, err := drv.NodeStageVolume(ctx, req)
 			Expect(err).NotTo(HaveOccurred())
 		})
-
 	})
 
 	Describe("NodePublishVolume", func() {
@@ -598,7 +598,6 @@ var _ = Describe("Node", func() {
 			Expect(err).NotTo(HaveOccurred())
 			Expect(res).To(SatisfyAll(
 				HaveField("Usage", ContainElements([]*csi.VolumeUsage{
-
 					{
 						Available: 0,
 						Total:     0,
@@ -615,6 +614,5 @@ var _ = Describe("Node", func() {
 				)),
 			))
 		})
-
 	})
 })