Merge pull request #7438 from Lyndon-Li/batch-delete-snapshot

Issue 7281: batch delete snapshot
This commit is contained in:
lyndon-li
2024-03-27 13:31:07 +08:00
committed by GitHub
14 changed files with 605 additions and 68 deletions
+66 -13
View File
@@ -68,6 +68,7 @@ type backupDeletionReconciler struct {
newPluginManager func(logrus.FieldLogger) clientmgmt.Manager
backupStoreGetter persistence.ObjectBackupStoreGetter
credentialStore credentials.FileStore
repoEnsurer *repository.Ensurer
}
// NewBackupDeletionReconciler creates a new backup deletion reconciler.
@@ -81,6 +82,7 @@ func NewBackupDeletionReconciler(
newPluginManager func(logrus.FieldLogger) clientmgmt.Manager,
backupStoreGetter persistence.ObjectBackupStoreGetter,
credentialStore credentials.FileStore,
repoEnsurer *repository.Ensurer,
) *backupDeletionReconciler {
return &backupDeletionReconciler{
Client: client,
@@ -93,6 +95,7 @@ func NewBackupDeletionReconciler(
newPluginManager: newPluginManager,
backupStoreGetter: backupStoreGetter,
credentialStore: credentialStore,
repoEnsurer: repoEnsurer,
}
}
@@ -502,20 +505,16 @@ func (r *backupDeletionReconciler) deletePodVolumeSnapshots(ctx context.Context,
return nil
}
snapshots, err := getSnapshotsInBackup(ctx, backup, r.Client)
directSnapshots, err := getSnapshotsInBackup(ctx, backup, r.Client)
if err != nil {
return []error{err}
}
var errs []error
for _, snapshot := range snapshots {
if err := r.repoMgr.Forget(ctx, snapshot); err != nil {
errs = append(errs, err)
}
}
return errs
return batchDeleteSnapshots(ctx, r.repoEnsurer, r.repoMgr, directSnapshots, backup, r.logger)
}
var batchDeleteSnapshotFunc = batchDeleteSnapshots
func (r *backupDeletionReconciler) deleteMovedSnapshots(ctx context.Context, backup *velerov1api.Backup) []error {
if r.repoMgr == nil {
return nil
@@ -532,26 +531,52 @@ func (r *backupDeletionReconciler) deleteMovedSnapshots(ctx context.Context, bac
return []error{errors.Wrapf(err, "failed to retrieve config for snapshot info")}
}
var errs []error
directSnapshots := map[string][]repository.SnapshotIdentifier{}
for i := range list.Items {
cm := list.Items[i]
snapshot := repository.SnapshotIdentifier{}
if cm.Data == nil || len(cm.Data) == 0 {
errs = append(errs, errors.New("no snapshot info in config"))
continue
}
b, err := json.Marshal(cm.Data)
if err != nil {
errs = append(errs, errors.Wrapf(err, "fail to marshal the snapshot info into JSON"))
continue
}
snapshot := repository.SnapshotIdentifier{}
if err := json.Unmarshal(b, &snapshot); err != nil {
errs = append(errs, errors.Wrapf(err, "failed to unmarshal snapshot info"))
continue
}
if err := r.repoMgr.Forget(ctx, snapshot); err != nil {
errs = append(errs, errors.Wrapf(err, "failed to delete snapshot %s, namespace: %s", snapshot.SnapshotID, snapshot.VolumeNamespace))
if snapshot.SnapshotID == "" || snapshot.VolumeNamespace == "" || snapshot.RepositoryType == "" {
errs = append(errs, errors.Errorf("invalid snapshot, ID %s, namespace %s, repository %s", snapshot.SnapshotID, snapshot.VolumeNamespace, snapshot.RepositoryType))
continue
}
r.logger.Infof("Deleted snapshot %s, namespace: %s, repo type: %s", snapshot.SnapshotID, snapshot.VolumeNamespace, snapshot.RepositoryType)
if directSnapshots[snapshot.VolumeNamespace] == nil {
directSnapshots[snapshot.VolumeNamespace] = []repository.SnapshotIdentifier{}
}
directSnapshots[snapshot.VolumeNamespace] = append(directSnapshots[snapshot.VolumeNamespace], snapshot)
r.logger.Infof("Deleting snapshot %s, namespace: %s, repo type: %s", snapshot.SnapshotID, snapshot.VolumeNamespace, snapshot.RepositoryType)
}
for i := range list.Items {
cm := list.Items[i]
if err := r.Client.Delete(ctx, &cm); err != nil {
r.logger.Warnf("Failed to delete snapshot info configmap %s/%s: %v", cm.Namespace, cm.Name, err)
}
}
if len(directSnapshots) > 0 {
deleteErrs := batchDeleteSnapshotFunc(ctx, r.repoEnsurer, r.repoMgr, directSnapshots, backup, r.logger)
errs = append(errs, deleteErrs...)
}
return errs
}
@@ -592,7 +617,7 @@ func (r *backupDeletionReconciler) patchBackup(ctx context.Context, backup *vele
// getSnapshotsInBackup returns a list of all pod volume snapshot ids associated with
// a given Velero backup.
func getSnapshotsInBackup(ctx context.Context, backup *velerov1api.Backup, kbClient client.Client) ([]repository.SnapshotIdentifier, error) {
func getSnapshotsInBackup(ctx context.Context, backup *velerov1api.Backup, kbClient client.Client) (map[string][]repository.SnapshotIdentifier, error) {
podVolumeBackups := &velerov1api.PodVolumeBackupList{}
options := &client.ListOptions{
LabelSelector: labels.Set(map[string]string{
@@ -607,3 +632,31 @@ func getSnapshotsInBackup(ctx context.Context, backup *velerov1api.Backup, kbCli
return podvolume.GetSnapshotIdentifier(podVolumeBackups), nil
}
func batchDeleteSnapshots(ctx context.Context, repoEnsurer *repository.Ensurer, repoMgr repository.Manager,
directSnapshots map[string][]repository.SnapshotIdentifier, backup *velerov1api.Backup, logger logrus.FieldLogger) []error {
var errs []error
for volumeNamespace, snapshots := range directSnapshots {
batchForget := []string{}
for _, snapshot := range snapshots {
batchForget = append(batchForget, snapshot.SnapshotID)
}
// For volumes in one backup, the BSL and repositoryType should always be the same
repoType := snapshots[0].RepositoryType
repo, err := repoEnsurer.EnsureRepo(ctx, backup.Namespace, volumeNamespace, backup.Spec.StorageLocation, repoType)
if err != nil {
errs = append(errs, errors.Wrapf(err, "error to ensure repo %s-%s-%s, skip deleting PVB snapshots %v", backup.Spec.StorageLocation, volumeNamespace, repoType, batchForget))
continue
}
if forgetErrs := repoMgr.BatchForget(ctx, repo, batchForget); len(forgetErrs) > 0 {
errs = append(errs, forgetErrs...)
continue
}
logger.Infof("Batch deleted snapshots %v", batchForget)
}
return errs
}
+195 -29
View File
@@ -18,9 +18,11 @@ package controller
import (
"bytes"
"encoding/json"
"errors"
"fmt"
"io"
"sort"
"reflect"
"time"
"context"
@@ -53,6 +55,7 @@ import (
"github.com/vmware-tanzu/velero/pkg/plugin/velero"
"github.com/vmware-tanzu/velero/pkg/plugin/velero/mocks"
"github.com/vmware-tanzu/velero/pkg/repository"
repomocks "github.com/vmware-tanzu/velero/pkg/repository/mocks"
velerotest "github.com/vmware-tanzu/velero/pkg/test"
)
@@ -94,6 +97,7 @@ func setupBackupDeletionControllerTest(t *testing.T, req *velerov1api.DeleteBack
func(logrus.FieldLogger) clientmgmt.Manager { return pluginManager },
NewFakeSingleObjectBackupStoreGetter(backupStore),
velerotest.NewFakeCredentialsFileStore("", nil),
nil,
),
req: ctrl.Request{NamespacedName: types.NamespacedName{Namespace: req.Namespace, Name: req.Name}},
}
@@ -694,13 +698,13 @@ func TestGetSnapshotsInBackup(t *testing.T) {
tests := []struct {
name string
podVolumeBackups []velerov1api.PodVolumeBackup
expected []repository.SnapshotIdentifier
expected map[string][]repository.SnapshotIdentifier
longBackupNameEnabled bool
}{
{
name: "no pod volume backups",
podVolumeBackups: nil,
expected: nil,
expected: map[string][]repository.SnapshotIdentifier{},
},
{
name: "no pod volume backups with matching label",
@@ -720,7 +724,7 @@ func TestGetSnapshotsInBackup(t *testing.T) {
Status: velerov1api.PodVolumeBackupStatus{SnapshotID: "snap-2"},
},
},
expected: nil,
expected: map[string][]repository.SnapshotIdentifier{},
},
{
name: "some pod volume backups with matching label",
@@ -761,16 +765,18 @@ func TestGetSnapshotsInBackup(t *testing.T) {
Status: velerov1api.PodVolumeBackupStatus{SnapshotID: ""},
},
},
expected: []repository.SnapshotIdentifier{
{
VolumeNamespace: "ns-1",
SnapshotID: "snap-3",
RepositoryType: "restic",
},
{
VolumeNamespace: "ns-1",
SnapshotID: "snap-4",
RepositoryType: "restic",
expected: map[string][]repository.SnapshotIdentifier{
"ns-1": {
{
VolumeNamespace: "ns-1",
SnapshotID: "snap-3",
RepositoryType: "restic",
},
{
VolumeNamespace: "ns-1",
SnapshotID: "snap-4",
RepositoryType: "restic",
},
},
},
},
@@ -814,11 +820,13 @@ func TestGetSnapshotsInBackup(t *testing.T) {
Status: velerov1api.PodVolumeBackupStatus{SnapshotID: ""},
},
},
expected: []repository.SnapshotIdentifier{
{
VolumeNamespace: "ns-1",
SnapshotID: "snap-3",
RepositoryType: "restic",
expected: map[string][]repository.SnapshotIdentifier{
"ns-1": {
{
VolumeNamespace: "ns-1",
SnapshotID: "snap-3",
RepositoryType: "restic",
},
},
},
},
@@ -843,21 +851,179 @@ func TestGetSnapshotsInBackup(t *testing.T) {
res, err := getSnapshotsInBackup(context.TODO(), veleroBackup, clientBuilder.Build())
assert.NoError(t, err)
// sort to ensure good compare of slices
less := func(snapshots []repository.SnapshotIdentifier) func(i, j int) bool {
return func(i, j int) bool {
if snapshots[i].VolumeNamespace == snapshots[j].VolumeNamespace {
return snapshots[i].SnapshotID < snapshots[j].SnapshotID
}
return snapshots[i].VolumeNamespace < snapshots[j].VolumeNamespace
assert.True(t, reflect.DeepEqual(res, test.expected))
})
}
}
func batchDeleteSucceed(ctx context.Context, repoEnsurer *repository.Ensurer, repoMgr repository.Manager, directSnapshots map[string][]repository.SnapshotIdentifier, backup *velerov1api.Backup, logger logrus.FieldLogger) []error {
return nil
}
func batchDeleteFail(ctx context.Context, repoEnsurer *repository.Ensurer, repoMgr repository.Manager, directSnapshots map[string][]repository.SnapshotIdentifier, backup *velerov1api.Backup, logger logrus.FieldLogger) []error {
return []error{
errors.New("fake-delete-1"),
errors.New("fake-delete-2"),
}
}
func generateSnapshotData(snapshot *repository.SnapshotIdentifier) (map[string]string, error) {
if snapshot == nil {
return nil, nil
}
b, err := json.Marshal(snapshot)
if err != nil {
return nil, err
}
data := make(map[string]string)
if err := json.Unmarshal(b, &data); err != nil {
return nil, err
}
return data, nil
}
func TestDeleteMovedSnapshots(t *testing.T) {
tests := []struct {
name string
repoMgr repository.Manager
batchDeleteSucceed bool
backupName string
snapshots []*repository.SnapshotIdentifier
expected []string
}{
{
name: "repoMgr is nil",
},
{
name: "no cm",
repoMgr: repomocks.NewManager(t),
},
{
name: "bad cm info",
repoMgr: repomocks.NewManager(t),
backupName: "backup-01",
snapshots: []*repository.SnapshotIdentifier{nil},
expected: []string{"no snapshot info in config"},
},
{
name: "invalid snapshots",
repoMgr: repomocks.NewManager(t),
backupName: "backup-01",
snapshots: []*repository.SnapshotIdentifier{
{
RepositoryType: "repo-1",
VolumeNamespace: "ns-1",
},
{
SnapshotID: "snapshot-1",
VolumeNamespace: "ns-1",
},
{
SnapshotID: "snapshot-1",
RepositoryType: "repo-1",
},
},
batchDeleteSucceed: true,
expected: []string{
"invalid snapshot, ID , namespace ns-1, repository repo-1",
"invalid snapshot, ID snapshot-1, namespace ns-1, repository ",
"invalid snapshot, ID snapshot-1, namespace , repository repo-1",
},
},
{
name: "batch delete succeed",
repoMgr: repomocks.NewManager(t),
backupName: "backup-01",
snapshots: []*repository.SnapshotIdentifier{
{
SnapshotID: "snapshot-1",
RepositoryType: "repo-1",
VolumeNamespace: "ns-1",
},
},
batchDeleteSucceed: true,
expected: []string{},
},
{
name: "batch delete fail",
repoMgr: repomocks.NewManager(t),
backupName: "backup-01",
snapshots: []*repository.SnapshotIdentifier{
{
RepositoryType: "repo-1",
VolumeNamespace: "ns-1",
},
{
SnapshotID: "snapshot-1",
RepositoryType: "repo-1",
VolumeNamespace: "ns-1",
},
},
expected: []string{"invalid snapshot, ID , namespace ns-1, repository repo-1", "fake-delete-1", "fake-delete-2"},
},
}
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
objs := []runtime.Object{}
for i, snapshot := range test.snapshots {
snapshotData, err := generateSnapshotData(snapshot)
require.NoError(t, err)
cm := corev1api.ConfigMap{
TypeMeta: metav1.TypeMeta{
APIVersion: corev1api.SchemeGroupVersion.String(),
Kind: "ConfigMap",
},
ObjectMeta: metav1.ObjectMeta{
Namespace: "velero",
Name: fmt.Sprintf("du-info-%d", i),
Labels: map[string]string{
velerov1api.BackupNameLabel: test.backupName,
velerov1api.DataUploadSnapshotInfoLabel: "true",
},
},
Data: snapshotData,
}
objs = append(objs, &cm)
}
sort.Slice(test.expected, less(test.expected))
sort.Slice(res, less(res))
veleroBackup := &velerov1api.Backup{}
controller := NewBackupDeletionReconciler(
velerotest.NewLogger(),
velerotest.NewFakeControllerRuntimeClient(t, objs...),
NewBackupTracker(),
test.repoMgr,
metrics.NewServerMetrics(),
nil, // discovery helper
func(logrus.FieldLogger) clientmgmt.Manager { return pluginManager },
NewFakeSingleObjectBackupStoreGetter(backupStore),
velerotest.NewFakeCredentialsFileStore("", nil),
nil,
)
assert.Equal(t, test.expected, res)
veleroBackup.Name = test.backupName
if test.batchDeleteSucceed {
batchDeleteSnapshotFunc = batchDeleteSucceed
} else {
batchDeleteSnapshotFunc = batchDeleteFail
}
errs := controller.deleteMovedSnapshots(context.Background(), veleroBackup)
if test.expected == nil {
assert.Nil(t, errs)
} else {
assert.Equal(t, len(test.expected), len(errs))
for i := range test.expected {
assert.EqualError(t, errs[i], test.expected[i])
}
}
})
}
}
+1 -1
View File
@@ -343,7 +343,7 @@ func (r *DataUploadReconciler) runCancelableDataUpload(ctx context.Context, fsBa
velerov1api.AsyncOperationIDLabel: du.Labels[velerov1api.AsyncOperationIDLabel],
}
if err := fsBackup.StartBackup(path, fmt.Sprintf("%s/%s", du.Spec.SourceNamespace, du.Spec.SourcePVC), "", false, tags, du.Spec.DataMoverConfig); err != nil {
if err := fsBackup.StartBackup(path, datamover.GetRealSource(du.Spec.SourceNamespace, du.Spec.SourcePVC), "", false, tags, du.Spec.DataMoverConfig); err != nil {
return r.errorOut(ctx, du, err, "error starting data path backup", log)
}