Merge pull request #321 from harshika-kashyap/concurrent_changes

fix: Handle concurrent operation of NodeUnstage with NodeStage and NodeUnpublish with NodePublish

Author: k8s-ci-robot

pkg/blob/blob.go

@@ -94,6 +94,9 @@ type Driver struct {
 	cloud      *azure.Cloud
 	mounter    *mount.SafeFormatAndMount
 	volLockMap *util.LockMap
+	// A map storing all volumes with ongoing operations so that additional operations
+	// for that same volume (as defined by VolumeID) return an Aborted error
+	volumeLocks *volumeLocks
 }
 
 // NewDriver Creates a NewCSIDriver object. Assumes vendor version is equal to driver version &
@@ -104,6 +107,7 @@ func NewDriver(nodeID string) *Driver {
 	driver.Version = driverVersion
 	driver.NodeID = nodeID
 	driver.volLockMap = util.NewLockMap()
+	driver.volumeLocks = newVolumeLocks()
 	return &driver
 }
 

pkg/blob/nodeserver.go

@@ -60,6 +60,11 @@ func (d *Driver) NodePublishVolume(ctx context.Context, req *csi.NodePublishVolu
 		return nil, status.Error(codes.InvalidArgument, "Target path not provided")
 	}
 
+	if acquired := d.volumeLocks.TryAcquire(volumeID); !acquired {
+		return nil, status.Errorf(codes.Aborted, volumeOperationAlreadyExistsFmt, volumeID)
+	}
+	defer d.volumeLocks.Release(volumeID)
+
 	mountOptions := []string{"bind"}
 	if req.GetReadonly() {
 		mountOptions = append(mountOptions, "ro")
@@ -97,6 +102,11 @@ func (d *Driver) NodeUnpublishVolume(ctx context.Context, req *csi.NodeUnpublish
 	targetPath := req.GetTargetPath()
 	volumeID := req.GetVolumeId()
 
+	if acquired := d.volumeLocks.TryAcquire(volumeID); !acquired {
+		return nil, status.Errorf(codes.Aborted, volumeOperationAlreadyExistsFmt, volumeID)
+	}
+	defer d.volumeLocks.Release(volumeID)
+
 	klog.V(2).Infof("NodeUnpublishVolume: unmounting volume %s on %s", volumeID, targetPath)
 	err := mount.CleanupMountPoint(targetPath, d.mounter, false)
 	if err != nil {
@@ -122,6 +132,11 @@ func (d *Driver) NodeStageVolume(ctx context.Context, req *csi.NodeStageVolumeRe
 		return nil, status.Error(codes.InvalidArgument, "Volume capability not provided")
 	}
 
+	if acquired := d.volumeLocks.TryAcquire(volumeID); !acquired {
+		return nil, status.Errorf(codes.Aborted, volumeOperationAlreadyExistsFmt, volumeID)
+	}
+	defer d.volumeLocks.Release(volumeID)
+
 	mnt, err := d.ensureMountPoint(targetPath)
 	if err != nil {
 		return nil, status.Errorf(codes.Internal, "Could not mount target %q: %v", targetPath, err)
@@ -239,6 +254,11 @@ func (d *Driver) NodeUnstageVolume(ctx context.Context, req *csi.NodeUnstageVolu
 		return nil, status.Error(codes.InvalidArgument, "Staging target not provided")
 	}
 
+	if acquired := d.volumeLocks.TryAcquire(volumeID); !acquired {
+		return nil, status.Errorf(codes.Aborted, volumeOperationAlreadyExistsFmt, volumeID)
+	}
+	defer d.volumeLocks.Release(volumeID)
+
 	klog.V(2).Infof("NodeUnstageVolume: volume %s unmounting on %s", volumeID, stagingTargetPath)
 	err := mount.CleanupMountPoint(stagingTargetPath, d.mounter, false)
 	if err != nil {

pkg/blob/nodeserver_test.go

@@ -143,8 +143,10 @@ func TestNodePublishVolume(t *testing.T) {
 	}
 	tests := []struct {
 		desc        string
+		setup       func(*Driver)
 		req         csi.NodePublishVolumeRequest
 		expectedErr error
+		cleanup     func(*Driver)
 	}{
 		{
 			desc:        "Volume capabilities missing",
@@ -169,6 +171,21 @@ func TestNodePublishVolume(t *testing.T) {
 				StagingTargetPath: sourceTest},
 			expectedErr: status.Error(codes.InvalidArgument, "Target path not provided"),
 		},
+		{
+			desc: "[Error] Volume operation in progress",
+			setup: func(d *Driver) {
+				d.volumeLocks.TryAcquire("vol_1")
+			},
+			req: csi.NodePublishVolumeRequest{VolumeCapability: &csi.VolumeCapability{AccessMode: &volumeCap},
+				VolumeId:          "vol_1",
+				TargetPath:        targetTest,
+				StagingTargetPath: sourceTest,
+				Readonly:          true},
+			expectedErr: status.Error(codes.Aborted, fmt.Sprintf(volumeOperationAlreadyExistsFmt, "vol_1")),
+			cleanup: func(d *Driver) {
+				d.volumeLocks.Release("vol_1")
+			},
+		},
 		{
 			desc: "Valid request read only",
 			req: csi.NodePublishVolumeRequest{VolumeCapability: &csi.VolumeCapability{AccessMode: &volumeCap},
@@ -210,11 +227,17 @@ func TestNodePublishVolume(t *testing.T) {
 	}
 
 	for _, test := range tests {
+		if test.setup != nil {
+			test.setup(d)
+		}
 		_, err := d.NodePublishVolume(context.Background(), &test.req)
 
 		if !reflect.DeepEqual(err, test.expectedErr) {
 			t.Errorf("Desc: %s - Unexpected error: %v - Expected: %v", test.desc, err, test.expectedErr)
 		}
+		if test.cleanup != nil {
+			test.cleanup(d)
+		}
 	}
 
 	// Clean up
@@ -230,8 +253,10 @@ func TestNodePublishVolume(t *testing.T) {
 func TestNodeUnpublishVolume(t *testing.T) {
 	tests := []struct {
 		desc        string
+		setup       func(*Driver)
 		req         csi.NodeUnpublishVolumeRequest
 		expectedErr error
+		cleanup     func(*Driver)
 	}{
 		{
 			desc:        "Volume ID missing",
@@ -243,6 +268,17 @@ func TestNodeUnpublishVolume(t *testing.T) {
 			req:         csi.NodeUnpublishVolumeRequest{VolumeId: "vol_1"},
 			expectedErr: status.Error(codes.InvalidArgument, "Target path missing in request"),
 		},
+		{
+			desc: "[Error] Volume operation in progress",
+			setup: func(d *Driver) {
+				d.volumeLocks.TryAcquire("vol_1")
+			},
+			req:         csi.NodeUnpublishVolumeRequest{TargetPath: targetTest, VolumeId: "vol_1"},
+			expectedErr: status.Error(codes.Aborted, fmt.Sprintf(volumeOperationAlreadyExistsFmt, "vol_1")),
+			cleanup: func(d *Driver) {
+				d.volumeLocks.Release("vol_1")
+			},
+		},
 		{
 			desc:        "Valid request",
 			req:         csi.NodeUnpublishVolumeRequest{TargetPath: "./abc.go", VolumeId: "vol_1"},
@@ -263,11 +299,17 @@ func TestNodeUnpublishVolume(t *testing.T) {
 	}
 
 	for _, test := range tests {
+		if test.setup != nil {
+			test.setup(d)
+		}
 		_, err := d.NodeUnpublishVolume(context.Background(), &test.req)
 
 		if !reflect.DeepEqual(err, test.expectedErr) {
 			t.Errorf("Unexpected error: %v", err)
 		}
+		if test.cleanup != nil {
+			test.cleanup(d)
+		}
 	}
 
 	//Clean up
@@ -280,6 +322,7 @@ func TestNodeUnpublishVolume(t *testing.T) {
 }
 
 func TestNodeStageVolume(t *testing.T) {
+	volumeCap := csi.VolumeCapability_AccessMode{Mode: csi.VolumeCapability_AccessMode_MULTI_NODE_MULTI_WRITER}
 	testCases := []struct {
 		name     string
 		testFunc func(t *testing.T)
@@ -325,13 +368,32 @@ func TestNodeStageVolume(t *testing.T) {
 				}
 			},
 		},
+		{
+			name: "Volume operation in progress",
+			testFunc: func(t *testing.T) {
+				req := &csi.NodeStageVolumeRequest{
+					VolumeId:          "unit-test",
+					StagingTargetPath: "unit-test",
+					VolumeCapability:  &csi.VolumeCapability{AccessMode: &volumeCap},
+				}
+				d := NewFakeDriver()
+				d.volumeLocks.TryAcquire("unit-test")
+				defer d.volumeLocks.Release("unit-test")
+				_, err := d.NodeStageVolume(context.TODO(), req)
+				expectedErr := status.Error(codes.Aborted, fmt.Sprintf(volumeOperationAlreadyExistsFmt, "unit-test"))
+				if !reflect.DeepEqual(err, expectedErr) {
+					t.Errorf("actualErr: (%v), expectedErr: (%v)", err, expectedErr)
+				}
+			},
+		},
 	}
 	for _, tc := range testCases {
 		t.Run(tc.name, tc.testFunc)
 	}
 }
 
 func TestNodeUnstageVolume(t *testing.T) {
+	volumeCap := csi.VolumeCapability_AccessMode{Mode: csi.VolumeCapability_AccessMode_MULTI_NODE_MULTI_WRITER}
 	testCases := []struct {
 		name     string
 		testFunc func(t *testing.T)
@@ -362,6 +424,24 @@ func TestNodeUnstageVolume(t *testing.T) {
 				}
 			},
 		},
+		{
+			name: "Volume operation in progress",
+			testFunc: func(t *testing.T) {
+				req := &csi.NodeStageVolumeRequest{
+					VolumeId:          "unit-test",
+					StagingTargetPath: "unit-test",
+					VolumeCapability:  &csi.VolumeCapability{AccessMode: &volumeCap},
+				}
+				d := NewFakeDriver()
+				d.volumeLocks.TryAcquire("unit-test")
+				defer d.volumeLocks.Release("unit-test")
+				_, err := d.NodeStageVolume(context.TODO(), req)
+				expectedErr := status.Error(codes.Aborted, fmt.Sprintf(volumeOperationAlreadyExistsFmt, "unit-test"))
+				if !reflect.DeepEqual(err, expectedErr) {
+					t.Errorf("actualErr: (%v), expectedErr: (%v)", err, expectedErr)
+				}
+			},
+		},
 		{
 			name: "mount point not exist ",
 			testFunc: func(t *testing.T) {

pkg/blob/volume_lock.go

@@ -0,0 +1,58 @@
+/*
+Copyright 2021 The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package blob
+
+import (
+	"sync"
+
+	"k8s.io/apimachinery/pkg/util/sets"
+)
+
+const (
+	volumeOperationAlreadyExistsFmt = "An operation with the given Volume ID %s already exists"
+)
+
+// VolumeLocks implements a map with atomic operations. It stores a set of all volume IDs
+// with an ongoing operation.
+type volumeLocks struct {
+	locks sets.String
+	mux   sync.Mutex
+}
+
+func newVolumeLocks() *volumeLocks {
+	return &volumeLocks{
+		locks: sets.NewString(),
+	}
+}
+
+// TryAcquire tries to acquire the lock for operating on volumeID and returns true if successful.
+// If another operation is already using volumeID, returns false.
+func (vl *volumeLocks) TryAcquire(volumeID string) bool {
+	vl.mux.Lock()
+	defer vl.mux.Unlock()
+	if vl.locks.Has(volumeID) {
+		return false
+	}
+	vl.locks.Insert(volumeID)
+	return true
+}
+
+func (vl *volumeLocks) Release(volumeID string) {
+	vl.mux.Lock()
+	defer vl.mux.Unlock()
+	vl.locks.Delete(volumeID)
+}