Merge pull request #754 from andyzhangx/fix-delete-error-archived

fix: delete volume error in archive deletion mode
kubernetes-csi · Sep 1, 2024 · 6c93c6a · 6c93c6a
2 parents 3b8d72d + e0ea084
commit 6c93c6a
Show file tree

Hide file tree

Showing 2 changed files with 26 additions and 0 deletions.
diff --git a/pkg/nfs/controllerserver.go b/pkg/nfs/controllerserver.go
@@ -32,6 +32,7 @@ import (
 	"google.golang.org/grpc/codes"
 	"google.golang.org/grpc/status"
 	"google.golang.org/protobuf/types/known/timestamppb"
+	azcache "sigs.k8s.io/cloud-provider-azure/pkg/cache"
 
 	"k8s.io/klog/v2"
 )
@@ -151,6 +152,11 @@ func (cs *ControllerServer) CreateVolume(ctx context.Context, req *csi.CreateVol
 		}
 	}
 
+	if acquired := cs.Driver.volumeLocks.TryAcquire(name); !acquired {
+		return nil, status.Errorf(codes.Aborted, volumeOperationAlreadyExistsFmt, name)
+	}
+	defer cs.Driver.volumeLocks.Release(name)
+
 	nfsVol, err := newNFSVolume(name, reqCapacity, parameters, cs.Driver.defaultOnDeletePolicy)
 	if err != nil {
 		return nil, status.Error(codes.InvalidArgument, err.Error())
@@ -230,7 +236,21 @@ func (cs *ControllerServer) DeleteVolume(ctx context.Context, req *csi.DeleteVol
 		nfsVol.onDelete = cs.Driver.defaultOnDeletePolicy
 	}
 
+	if acquired := cs.Driver.volumeLocks.TryAcquire(volumeID); !acquired {
+		return nil, status.Errorf(codes.Aborted, volumeOperationAlreadyExistsFmt, volumeID)
+	}
+	defer cs.Driver.volumeLocks.Release(volumeID)
+
 	if !strings.EqualFold(nfsVol.onDelete, retain) {
+		// check whether volumeID is in the cache
+		cache, err := cs.Driver.volDeletionCache.Get(volumeID, azcache.CacheReadTypeDefault)
+		if err != nil {
+			return nil, status.Errorf(codes.Internal, err.Error())
+		}
+		if cache != nil {
+			klog.V(2).Infof("DeleteVolume: volume %s is already deleted", volumeID)
+			return &csi.DeleteVolumeResponse{}, nil
+		}
 		// mount nfs base share so we can delete the subdirectory
 		if err = cs.internalMount(ctx, nfsVol, nil, volCap); err != nil {
 			return nil, status.Errorf(codes.Internal, "failed to mount nfs server: %v", err.Error())
@@ -281,6 +301,7 @@ func (cs *ControllerServer) DeleteVolume(ctx context.Context, req *csi.DeleteVol
 		klog.V(2).Infof("DeleteVolume: volume(%s) is set to retain, not deleting/archiving subdirectory", volumeID)
 	}
 
+	cs.Driver.volDeletionCache.Set(volumeID, "")
 	return &csi.DeleteVolumeResponse{}, nil
 }
 

diff --git a/pkg/nfs/nfs.go b/pkg/nfs/nfs.go
@@ -59,6 +59,8 @@ type Driver struct {
 	// a timed cache storing volume stats <volumeID, volumeStats>
 	volStatsCache                azcache.Resource
 	volStatsCacheExpireInMinutes int
+	// a timed cache storing volume deletion records <volumeID, "">
+	volDeletionCache azcache.Resource
 }
 
 const (
@@ -119,6 +121,9 @@ func NewDriver(options *DriverOptions) *Driver {
 	if n.volStatsCache, err = azcache.NewTimedCache(time.Duration(options.VolStatsCacheExpireInMinutes)*time.Minute, getter, false); err != nil {
 		klog.Fatalf("%v", err)
 	}
+	if n.volDeletionCache, err = azcache.NewTimedCache(time.Minute, getter, false); err != nil {
+		klog.Fatalf("%v", err)
+	}
 	return n
 }