rbd: fix resync issue

During the Demote volume store
the image creation timestamp.

During Resync do below operation

* Check image creation timestamp
stored during Demote operation
and current creation timestamp during Resync
and check both are equal and its for
force resync then issue resync
* If the image on both sides is
not in unknown state, check
last_snapshot_timestamp on the
local mirror description, if its present
send volumeReady as false or else return
error message.

If both the images are in up+unknown the
send volumeReady as true.

Signed-off-by: Madhu Rajanna <[email protected]>

internal/csi-addons/rbd/replication.go

@@ -24,6 +24,7 @@ import (
 	"regexp"
 	"strconv"
 	"strings"
+	"sync"
 	"time"
 
 	corerbd "github.com/ceph/ceph-csi/internal/rbd"
@@ -40,6 +41,12 @@ import (
 	"google.golang.org/protobuf/types/known/timestamppb"
 )
 
+// imageResyncMap is used to store the volumeID and image creation time for resync.
+// The volumeID gets added to the map when the DemoteVolume RPC is called and
+// removed from the map when the PromoteVolume RPC is called or
+// DisableVolumeReplication RPC is called.
+var imageResyncMap = sync.Map{}
+
 // imageMirroringMode is used to indicate the mirroring mode for an RBD image.
 type imageMirroringMode string
 
@@ -337,11 +344,15 @@ func (rs *ReplicationServer) DisableVolumeReplication(ctx context.Context,
 		if err != nil {
 			return nil, getGRPCError(err)
 		}
+		imageResyncMap.Delete(volumeID)
 
 		return &replication.DisableVolumeReplicationResponse{}, nil
 	default:
 		return nil, status.Errorf(codes.InvalidArgument, "image is in %s Mode", mirroringInfo.State)
 	}
+	// remove the volume from the resync map as the image mirroring is disabled and
+	// during next Demote operation cephcsi will re-add the image to the sync map
+	imageResyncMap.Delete(volumeID)
 
 	return &replication.DisableVolumeReplicationResponse{}, nil
 }
@@ -439,6 +450,13 @@ func (rs *ReplicationServer) PromoteVolume(ctx context.Context,
 			rbdVol)
 	}
 
+	// remove the volume from the resync map as the image is promoted and
+	// during next Demote operation cephcsi will re-add the image to the sync
+	// map.
+	// This is required if someone does not do DisableVolumeReplication but try
+	// to toggle between PromoteVolume and DemoteVolume.
+	imageResyncMap.Delete(volumeID)
+
 	return &replication.PromoteVolumeResponse{}, nil
 }
 
@@ -480,6 +498,18 @@ func (rs *ReplicationServer) DemoteVolume(ctx context.Context,
 
 		return nil, err
 	}
+
+	err = rbdVol.GetImageInfo()
+	if err != nil {
+		log.ErrorLog(ctx, err.Error())
+
+		return nil, status.Error(codes.Internal, err.Error())
+	}
+	// store the image creation time for resync
+	if _, ok := imageResyncMap.Load(volumeID); !ok {
+		imageResyncMap.Store(volumeID, rbdVol.CreatedAt.AsTime())
+	}
+
 	mirroringInfo, err := rbdVol.GetImageMirroringInfo()
 	if err != nil {
 		log.ErrorLog(ctx, err.Error())
@@ -538,6 +568,8 @@ func checkRemoteSiteStatus(ctx context.Context, mirrorStatus *librbd.GlobalMirro
 // ResyncVolume extracts the RBD volume information from the volumeID, If the
 // image is present, mirroring is enabled and the image is in demoted state.
 // If yes it will resync the image to correct the split-brain.
+//
+//nolint:gocyclo,cyclop // TODO: reduce complexity
 func (rs *ReplicationServer) ResyncVolume(ctx context.Context,
 	req *replication.ResyncVolumeRequest,
 ) (*replication.ResyncVolumeResponse, error) {
@@ -572,22 +604,11 @@ func (rs *ReplicationServer) ResyncVolume(ctx context.Context,
 		return nil, err
 	}
 
-	mirroringInfo, err := rbdVol.GetImageMirroringInfo()
+	err = rbdVol.CheckImageIsPrimary()
 	if err != nil {
-		// in case of Resync the image will get deleted and gets recreated and
-		// it takes time for this operation.
 		log.ErrorLog(ctx, err.Error())
 
-		return nil, status.Error(codes.Aborted, err.Error())
-	}
-
-	if mirroringInfo.State != librbd.MirrorImageEnabled {
-		return nil, status.Error(codes.InvalidArgument, "image mirroring is not enabled")
-	}
-
-	// return error if the image is still primary
-	if mirroringInfo.Primary {
-		return nil, status.Error(codes.InvalidArgument, "image is in primary state")
+		return nil, getGRPCError(err)
 	}
 
 	mirrorStatus, err := rbdVol.GetImageMirroringStatus()
@@ -637,14 +658,28 @@ func (rs *ReplicationServer) ResyncVolume(ctx context.Context,
 		ready = checkRemoteSiteStatus(ctx, mirrorStatus)
 	}
 
-	err = rbdVol.ResyncVol(localStatus, req.Force)
+	err = rbdVol.GetImageInfo()
 	if err != nil {
-		return nil, getGRPCError(err)
+		return nil, status.Errorf(codes.Internal, "failed to get image info: %s", err.Error())
 	}
 
-	err = checkVolumeResyncStatus(localStatus)
+	ok, err := resyncVolume(ctx, volumeID, rbdVol.CreatedAt)
 	if err != nil {
-		return nil, status.Error(codes.Internal, err.Error())
+		return nil, status.Errorf(codes.Internal, "failed to check image needs resync: %s", err.Error())
+	}
+
+	if req.Force && ok {
+		err = rbdVol.ResyncVol(localStatus)
+		if err != nil {
+			return nil, getGRPCError(err)
+		}
+	}
+
+	if !ready {
+		err = checkVolumeResyncStatus(localStatus)
+		if err != nil {
+			return nil, status.Error(codes.Internal, err.Error())
+		}
 	}
 
 	err = rbdVol.RepairResyncedImageID(ctx, ready)
@@ -659,6 +694,29 @@ func (rs *ReplicationServer) ResyncVolume(ctx context.Context,
 	return resp, nil
 }
 
+func resyncVolume(
+	ctx context.Context,
+	volumeID string,
+	currentImageTime *timestamppb.Timestamp,
+) (bool, error) {
+	savedImageTime, ok := imageResyncMap.Load(volumeID)
+	if !ok {
+		return false, errors.New("failed to get image creation time")
+	}
+
+	log.UsefulLog(ctx, "savedImageTime=%v, currentImageTime=%v", savedImageTime, currentImageTime)
+	st, ok := savedImageTime.(time.Time)
+	if !ok {
+		return false, errors.New("failed to convert image creation time")
+	}
+
+	if st.Equal(currentImageTime.AsTime()) {
+		return true, nil
+	}
+
+	return false, nil
+}
+
 func getGRPCError(err error) error {
 	if err == nil {
 		return status.Error(codes.OK, codes.OK.String())
@@ -854,20 +912,17 @@ func getLastSyncInfo(description string) (*replication.GetVolumeReplicationInfoR
 }
 
 func checkVolumeResyncStatus(localStatus librbd.SiteMirrorImageStatus) error {
-	// we are considering 2 states to check resync started and resync completed
-	// as below. all other states will be considered as an error state so that
-	// cephCSI can return error message and volume replication operator can
-	// mark the VolumeReplication status as not resyncing for the volume.
-
-	// If the state is Replaying means the resync is going on.
-	// Once the volume on remote cluster is demoted and resync
-	// is completed the image state will be moved to UNKNOWN.
-	// RBD mirror daemon should be always running on the primary cluster.
-	if !localStatus.Up || (localStatus.State != librbd.MirrorImageStatusStateReplaying &&
-		localStatus.State != librbd.MirrorImageStatusStateUnknown) {
-		return fmt.Errorf(
-			"not resyncing. Local status: daemon up=%t image is in %q state",
-			localStatus.Up, localStatus.State)
+	// we are considering local snapshot timestamp to check if the resync is
+	// started or not, if we dont see local_snapshot_timestamp in the
+	// description of localStatus, we are returning error. if we see the local
+	// snapshot timestamp in the description we return resyncing started.
+	description := localStatus.Description
+	resp, err := getLastSyncInfo(description)
+	if err != nil {
+		return fmt.Errorf("failed to get last sync info: %w", err)
+	}
+	if resp.LastSyncTime == nil {
+		return errors.New("last sync time is nil")
 	}
 
 	return nil

internal/rbd/nodeserver.go

@@ -227,7 +227,7 @@ func (ns *NodeServer) populateRbdVol(
 		}
 	}()
 	// get the image details from the ceph cluster.
-	err = rv.getImageInfo()
+	err = rv.GetImageInfo()
 	if err != nil {
 		log.ErrorLog(ctx, "failed to get image details %s: %v", rv, err)
 

internal/rbd/rbd_journal.go

@@ -166,7 +166,7 @@ func checkSnapCloneExists(
 	}
 	vol.ReservedID = snapUUID
 	// Fetch on-disk image attributes
-	err = vol.getImageInfo()
+	err = vol.GetImageInfo()
 	if err != nil {
 		if errors.Is(err, ErrImageNotFound) {
 			err = parentVol.deleteSnapshot(ctx, rbdSnap)
@@ -292,7 +292,7 @@ func (rv *rbdVolume) Exists(ctx context.Context, parentVol *rbdVolume) (bool, er
 	// save it for size checks before fetching image data
 	requestSize := rv.VolSize //nolint:ifshort // FIXME: rename and split function into helpers
 	// Fetch on-disk image attributes and compare against request
-	err = rv.getImageInfo()
+	err = rv.GetImageInfo()
 	if err != nil {
 		if errors.Is(err, ErrImageNotFound) {
 			// Need to check cloned info here not on createvolume,

internal/rbd/rbd_util.go

@@ -173,7 +173,7 @@ type rbdVolume struct {
 	// Network namespace file path to execute nsenter command
 	NetNamespaceFilePath string
 	// RequestedVolSize has the size of the volume requested by the user and
-	// this value will not be updated when doing getImageInfo() on rbdVolume.
+	// this value will not be updated when doing GetImageInfo() on rbdVolume.
 	RequestedVolSize   int64
 	DisableInUseChecks bool
 	readOnly           bool
@@ -716,7 +716,7 @@ func (ri *rbdImage) getCloneDepth(ctx context.Context) (uint, error) {
 			return depth, err
 		}
 
-		err = vol.getImageInfo()
+		err = vol.GetImageInfo()
 		// FIXME: create and destroy the vol inside the loop.
 		// see https://github.com/ceph/ceph-csi/pull/1838#discussion_r598530807
 		vol.ioctx.Destroy()
@@ -922,13 +922,13 @@ func (ri *rbdImage) checkImageChainHasFeature(ctx context.Context, feature uint6
 			return false, err
 		}
 
-		err = rbdImg.getImageInfo()
+		err = rbdImg.GetImageInfo()
 		// FIXME: create and destroy the vol inside the loop.
 		// see https://github.com/ceph/ceph-csi/pull/1838#discussion_r598530807
 		rbdImg.ioctx.Destroy()
 		rbdImg.ioctx = nil
 		if err != nil {
-			// call to getImageInfo returns the parent name even if the parent
+			// call to GetImageInfo returns the parent name even if the parent
 			// is in the trash, when we try to open the parent image to get its
 			// information it fails because it is already in trash. We should
 			// treat error as nil if the parent is not found.
@@ -1057,7 +1057,7 @@ func updateSnapshotDetails(rbdSnap *rbdSnapshot) error {
 	}
 	defer vol.Destroy()
 
-	err = vol.getImageInfo()
+	err = vol.GetImageInfo()
 	if err != nil {
 		return err
 	}
@@ -1153,7 +1153,7 @@ func generateVolumeFromVolumeID(
 			return rbdVol, err
 		}
 	}
-	err = rbdVol.getImageInfo()
+	err = rbdVol.GetImageInfo()
 
 	return rbdVol, err
 }
@@ -1501,7 +1501,7 @@ func (rv *rbdVolume) cloneRbdImageFromSnapshot(
 	}()
 
 	// get image latest information
-	err = rv.getImageInfo()
+	err = rv.GetImageInfo()
 	if err != nil {
 		return fmt.Errorf("failed to get image info of %s: %w", rv, err)
 	}
@@ -1558,9 +1558,9 @@ func (rv *rbdVolume) setImageOptions(ctx context.Context, options *librbd.ImageO
 	return nil
 }
 
-// getImageInfo queries rbd about the given image and returns its metadata, and returns
+// GetImageInfo queries rbd about the given image and returns its metadata, and returns
 // ErrImageNotFound if provided image is not found.
-func (ri *rbdImage) getImageInfo() error {
+func (ri *rbdImage) GetImageInfo() error {
 	image, err := ri.open()
 	if err != nil {
 		return err
@@ -1607,7 +1607,7 @@ func (ri *rbdImage) getImageInfo() error {
 
 // getParent returns parent image if it exists.
 func (ri *rbdImage) getParent() (*rbdImage, error) {
-	err := ri.getImageInfo()
+	err := ri.GetImageInfo()
 	if err != nil {
 		return nil, err
 	}
@@ -1623,7 +1623,7 @@ func (ri *rbdImage) getParent() (*rbdImage, error) {
 	parentImage.RadosNamespace = ri.RadosNamespace
 	parentImage.RbdImageName = ri.ParentName
 
-	err = parentImage.getImageInfo()
+	err = parentImage.GetImageInfo()
 	if err != nil {
 		return nil, err
 	}