From 12c9a8c2e0b6f7f9c1920410210da8a4d6834e2f Mon Sep 17 00:00:00 2001 From: Rakshith R Date: Fri, 22 Nov 2024 17:15:46 +0530 Subject: [PATCH] rbd: make use of both listSnapshots and listChildren Currently, CephCSI only uses listSnaps to determine number of snapshots on a RBD image and uses snapshot names as child image names to flatten them. But child images may have different name(in case of group snapshot) or they maybe in trash (deleted k8s VolSnapshot with alive restored PVC). The above problems are avoid by making use of both snap and child image lists. Signed-off-by: Rakshith R (cherry picked from commit 09d848e01727343a56e38f64a1daf2d03b234a84) --- internal/rbd/controllerserver.go | 26 +++++++++-- internal/rbd/rbd_util.go | 76 +++++--------------------------- 2 files changed, 34 insertions(+), 68 deletions(-) diff --git a/internal/rbd/controllerserver.go b/internal/rbd/controllerserver.go index 5ccec05f9..7457befc8 100644 --- a/internal/rbd/controllerserver.go +++ b/internal/rbd/controllerserver.go @@ -573,7 +573,7 @@ func (cs *ControllerServer) repairExistingVolume(ctx context.Context, req *csi.C // are more than the `minSnapshotOnImage` Add a task to flatten all the // temporary cloned images. func flattenTemporaryClonedImages(ctx context.Context, rbdVol *rbdVolume, cr *util.Credentials) error { - snaps, err := rbdVol.listSnapshots() + snaps, children, err := rbdVol.listSnapAndChildren() if err != nil { if errors.Is(err, ErrImageNotFound) { return status.Error(codes.InvalidArgument, err.Error()) @@ -589,9 +589,19 @@ func flattenTemporaryClonedImages(ctx context.Context, rbdVol *rbdVolume, cr *ut len(snaps), rbdVol, maxSnapshotsOnImage) + + if len(children) == 0 { + // if none of the child images(are in trash) exist, we can't flatten them. + // return ResourceExhausted error message as we have reached the hard limit. + log.ErrorLog(ctx, "child images of image %q cannot be flatten", rbdVol) + + return status.Errorf(codes.ResourceExhausted, + "rbd image %q has %d snapshots but child images cannot be flattened", + rbdVol, len(snaps)) + } err = flattenClonedRbdImages( ctx, - snaps, + children, rbdVol.Pool, rbdVol.Monitors, rbdVol.RbdImageName, @@ -610,13 +620,21 @@ func flattenTemporaryClonedImages(ctx context.Context, rbdVol *rbdVolume, cr *ut len(snaps), rbdVol, minSnapshotsOnImageToStartFlatten) + if len(children) == 0 { + // if none of the child images(are in trash) exist, we can't flatten them. + // return nil since we have only reach the soft limit. + log.DebugLog(ctx, "child images of image %q cannot be flatten", rbdVol) + + return nil + } // If we start flattening all the snapshots at one shot the volume // creation time will be affected,so we will flatten only the extra // snapshots. - snaps = snaps[minSnapshotsOnImageToStartFlatten-1:] + extraSnapshots := min(len(snaps)-int(minSnapshotsOnImageToStartFlatten), len(children)) + children = children[:extraSnapshots] err = flattenClonedRbdImages( ctx, - snaps, + children, rbdVol.Pool, rbdVol.Monitors, rbdVol.RbdImageName, diff --git a/internal/rbd/rbd_util.go b/internal/rbd/rbd_util.go index 6dc4d139a..d1fc4a328 100644 --- a/internal/rbd/rbd_util.go +++ b/internal/rbd/rbd_util.go @@ -785,13 +785,9 @@ func (ri *rbdImage) getCloneDepth(ctx context.Context) (uint, error) { } } -type trashSnapInfo struct { - origSnapName string -} - func flattenClonedRbdImages( ctx context.Context, - snaps []librbd.SnapInfo, + children []string, pool, monitors, rbdImageName string, cr *util.Credentials, ) error { @@ -807,26 +803,9 @@ func flattenClonedRbdImages( return err } - var origNameList []trashSnapInfo - for _, snapInfo := range snaps { - // check if the snapshot belongs to trash namespace. - isTrash, retErr := rv.isTrashSnap(snapInfo.Id) - if retErr != nil { - return retErr - } - if isTrash { - // get original snap name for the snapshot in trash namespace - origSnapName, retErr := rv.getOrigSnapName(snapInfo.Id) - if retErr != nil { - return retErr - } - origNameList = append(origNameList, trashSnapInfo{origSnapName}) - } - } - - for _, snapName := range origNameList { - rv.RbdImageName = snapName.origSnapName + for _, childName := range children { + rv.RbdImageName = childName err = rv.flattenRbdImage(ctx, true, rbdHardMaxCloneDepth, rbdSoftMaxCloneDepth) if err != nil { log.ErrorLog(ctx, "failed to flatten %s; err %v", rv, err) @@ -2052,57 +2031,26 @@ func (ri *rbdImage) DisableDeepFlatten() error { return image.UpdateFeatures(librbd.FeatureDeepFlatten, false) } -func (ri *rbdImage) listSnapshots() ([]librbd.SnapInfo, error) { +// listSnapAndChildren returns list of names of snapshots and child images. +func (ri *rbdImage) listSnapAndChildren() ([]librbd.SnapInfo, []string, error) { image, err := ri.open() if err != nil { - return nil, err + return nil, nil, err } defer image.Close() - snapInfoList, err := image.GetSnapshotNames() + snaps, err := image.GetSnapshotNames() if err != nil { - return nil, err + return nil, nil, err } - return snapInfoList, nil -} - -// isTrashSnap returns true if the snapshot belongs to trash namespace. -func (ri *rbdImage) isTrashSnap(snapID uint64) (bool, error) { - image, err := ri.open() + // ListChildren() returns pools, images, err. + _, children, err := image.ListChildren() if err != nil { - return false, err - } - defer image.Close() - - // Get namespace type for the snapshot - nsType, err := image.GetSnapNamespaceType(snapID) - if err != nil { - return false, err + return nil, nil, err } - if nsType == librbd.SnapNamespaceTypeTrash { - return true, nil - } - - return false, nil -} - -// getOrigSnapName returns the original snap name for -// the snapshots in Trash Namespace. -func (ri *rbdImage) getOrigSnapName(snapID uint64) (string, error) { - image, err := ri.open() - if err != nil { - return "", err - } - defer image.Close() - - origSnapName, err := image.GetSnapTrashNamespace(snapID) - if err != nil { - return "", err - } - - return origSnapName, nil + return snaps, children, nil } func (ri *rbdImage) isCompatibleEncryption(dst *rbdImage) error {