backup-operator: Support periodically backup (#1841)

pkg/apis/etcd/v1beta2/backup_types.go

@@ -101,6 +101,9 @@ type BackupPolicy struct {
 	// BackupIntervalInSecond is to specify how often operator take snapshot
 	// 0 is magic number to indicate one-shot backup
 	BackupIntervalInSecond int64 `json:"backupIntervalInSecond,omitempty"`
+	// MaxBackups is to specify how many backups we want to keep
+	// 0 is magic number to indicate un-limited backups
+	MaxBackups int `json:"maxBackups,omitempty"`
 }
 
 // BackupStatus represents the status of the EtcdBackup Custom Resource.

pkg/backup/backup_manager.go

@@ -18,6 +18,7 @@ import (
 	"context"
 	"crypto/tls"
 	"fmt"
+	"sort"
 	"time"
 
 	"github.com/coreos/etcd-operator/pkg/backup/writer"
@@ -77,6 +78,27 @@ func (bm *BackupManager) SaveSnap(ctx context.Context, s3Path string, now time.T
 	return rev, resp.Version, nil
 }
 
+// EnsureMaxbackup to ensure the number of snapshot is under maxcount
+// if the number of snapshot exceeded than maxcount, delete oldest snapshot
+func (bm *BackupManager) EnsureMaxbackup(ctx context.Context, s3Path string, maxCount int) error {
+
+	savedSnapShots, err := bm.bw.List(ctx, s3Path)
+	if err != nil {
+		return fmt.Errorf("failed to get exisiting snapashots: %v", err)
+	}
+	sort.Sort(sort.Reverse(sort.StringSlice(savedSnapShots)))
+	for i, snapshotPath := range savedSnapShots {
+		if i < maxCount {
+			continue
+		}
+		err := bm.bw.Delete(ctx, snapshotPath)
+		if err != nil {
+			return fmt.Errorf("failed to delete snapshot: %v", err)
+		}
+	}
+	return nil
+}
+
 // etcdClientWithMaxRevision gets the etcd endpoint with the maximum kv store revision
 // and returns the etcd client of that member.
 func (bm *BackupManager) etcdClientWithMaxRevision(ctx context.Context) (*clientv3.Client, int64, error) {

pkg/backup/writer/abs_writer.go

@@ -100,3 +100,50 @@ func (absw *absWriter) Write(ctx context.Context, path string, r io.Reader) (int
 
 	return blob.Properties.ContentLength, nil
 }
+
+func (absw *absWriter) List(ctx context.Context, basePath string) ([]string, error) {
+	// TODO: support context.
+	container, _, err := util.ParseBucketAndKey(basePath)
+	if err != nil {
+		return nil, err
+	}
+
+	containerRef := absw.abs.GetContainerReference(container)
+	containerExists, err := containerRef.Exists()
+	if err != nil {
+		return nil, err
+	}
+	if !containerExists {
+		return nil, fmt.Errorf("container %v does not exist", container)
+	}
+
+	blobs, err := containerRef.ListBlobs(
+		storage.ListBlobsParameters{Prefix: basePath})
+	if err != nil {
+		return nil, err
+	}
+	blobKeys := []string{}
+	for _, blob := range blobs.Blobs {
+		blobKeys = append(blobKeys, container+"/"+blob.Name)
+	}
+	return blobKeys, nil
+}
+
+func (absw *absWriter) Delete(ctx context.Context, path string) error {
+	// TODO: support context.
+	container, key, err := util.ParseBucketAndKey(path)
+	if err != nil {
+		return err
+	}
+	containerRef := absw.abs.GetContainerReference(container)
+	containerExists, err := containerRef.Exists()
+	if err != nil {
+		return err
+	}
+	if !containerExists {
+		return fmt.Errorf("container %v does not exist", container)
+	}
+
+	blob := containerRef.GetBlobReference(key)
+	return blob.Delete(&storage.DeleteBlobOptions{})
+}

pkg/backup/writer/gcs_writer.go

@@ -23,6 +23,7 @@ import (
 
 	"cloud.google.com/go/storage"
 	"github.com/sirupsen/logrus"
+	"google.golang.org/api/iterator"
 )
 
 var _ Writer = &gcsWriter{}
@@ -58,3 +59,37 @@ func (gcsw *gcsWriter) Write(ctx context.Context, path string, r io.Reader) (int
 	}
 	return n, err
 }
+
+func (gcsw *gcsWriter) List(ctx context.Context, basePath string) ([]string, error) {
+	bucket, key, err := util.ParseBucketAndKey(basePath)
+	if err != nil {
+		return nil, err
+	}
+	objects := gcsw.gcs.Bucket(bucket).Objects(ctx, &storage.Query{Prefix: key})
+	if objects == nil {
+		return nil, fmt.Errorf("failed to get objects having %s prefix", key)
+	}
+
+	objectKeys := []string{}
+
+	for {
+		objAttrs, err := objects.Next()
+		if err == iterator.Done {
+			break
+		}
+		if err != nil {
+			return nil, err
+		}
+		objectKeys = append(objectKeys, bucket+"/"+objAttrs.Name)
+	}
+	return objectKeys, nil
+}
+
+func (gcsw *gcsWriter) Delete(ctx context.Context, path string) error {
+	bucket, key, err := util.ParseBucketAndKey(path)
+	if err != nil {
+		return err
+	}
+
+	return gcsw.gcs.Bucket(bucket).Object(key).Delete(ctx)
+}

pkg/backup/writer/s3_writer.go

@@ -67,3 +67,39 @@ func (s3w *s3Writer) Write(ctx context.Context, path string, r io.Reader) (int64
 	}
 	return *resp.ContentLength, nil
 }
+
+// List return the file paths which match the given s3 path
+func (s3w *s3Writer) List(ctx context.Context, basePath string) ([]string, error) {
+	bk, key, err := util.ParseBucketAndKey(basePath)
+	if err != nil {
+		return nil, err
+	}
+
+	objects, err := s3w.s3.ListObjectsWithContext(ctx,
+		&s3.ListObjectsInput{
+			Bucket: aws.String(bk),
+			Prefix: aws.String(key),
+		})
+	if err != nil {
+		return nil, err
+	}
+	objectKeys := []string{}
+	for _, object := range objects.Contents {
+		objectKeys = append(objectKeys, bk+"/"+ *object.Key)
+	}
+	return objectKeys, nil
+}
+
+func (s3w *s3Writer) Delete(ctx context.Context, path string) error {
+	bk, key, err := util.ParseBucketAndKey(path)
+	if err != nil {
+		return err
+	}
+
+	_, err = s3w.s3.DeleteObjectWithContext(ctx,
+		&s3.DeleteObjectInput{
+			Bucket: aws.String(bk),
+			Key:    aws.String(key),
+		})
+	return err
+}

pkg/backup/writer/writer.go

@@ -23,4 +23,10 @@ import (
 type Writer interface {
 	// Write writes a backup file to the given path and returns size of written file.
 	Write(ctx context.Context, path string, r io.Reader) (int64, error)
+
+	// List a backup files
+	List(ctx context.Context, basePath string) ([]string, error)
+
+	// Delete a backup file
+	Delete(ctx context.Context, path string) error
 }

pkg/controller/backup-operator/abs_backup.go

@@ -29,7 +29,7 @@ import (
 )
 
 // handleABS saves etcd cluster's backup to specificed ABS path.
-func handleABS(ctx context.Context, kubecli kubernetes.Interface, s *api.ABSBackupSource, endpoints []string, clientTLSSecret, namespace string) (*api.BackupStatus, error) {
+func handleABS(ctx context.Context, kubecli kubernetes.Interface, s *api.ABSBackupSource, endpoints []string, clientTLSSecret, namespace string, maxBackup int) (*api.BackupStatus, error) {
 	// TODO: controls NewClientFromSecret with ctx. This depends on upstream kubernetes to support API calls with ctx.
 	cli, err := absfactory.NewClientFromSecret(kubecli, namespace, s.ABSSecret)
 	if err != nil {
@@ -47,5 +47,11 @@ func handleABS(ctx context.Context, kubecli kubernetes.Interface, s *api.ABSBack
 	if err != nil {
 		return nil, fmt.Errorf("failed to save snapshot (%v)", err)
 	}
+	if maxBackup > 0 {
+		err := bm.EnsureMaxbackup(ctx, s.Path, maxBackup)
+		if err != nil {
+			return nil, fmt.Errorf("succeeded in saving snapshot but failed to delete old snapshot (%v)", err)
+		}
+	}
 	return &api.BackupStatus{EtcdVersion: etcdVersion, EtcdRevision: rev, LastSuccessDate: now}, nil
 }

pkg/controller/backup-operator/gcs_backup.go

@@ -29,7 +29,7 @@ import (
 )
 
 // handleGCS saves etcd cluster's backup to specificed GCS path.
-func handleGCS(ctx context.Context, kubecli kubernetes.Interface, s *api.GCSBackupSource, endpoints []string, clientTLSSecret, namespace string) (*api.BackupStatus, error) {
+func handleGCS(ctx context.Context, kubecli kubernetes.Interface, s *api.GCSBackupSource, endpoints []string, clientTLSSecret, namespace string, maxBackup int) (*api.BackupStatus, error) {
 	// TODO: controls NewClientFromSecret with ctx. This depends on upstream kubernetes to support API calls with ctx.
 	cli, err := gcsfactory.NewClientFromSecret(ctx, kubecli, namespace, s.GCPSecret)
 	if err != nil {
@@ -48,5 +48,11 @@ func handleGCS(ctx context.Context, kubecli kubernetes.Interface, s *api.GCSBack
 	if err != nil {
 		return nil, fmt.Errorf("failed to save snapshot (%v)", err)
 	}
+	if maxBackup > 0 {
+		err := bm.EnsureMaxbackup(ctx, s.Path, maxBackup)
+		if err != nil {
+			return nil, fmt.Errorf("succeeded in saving snapshot but failed to delete old snapshot (%v)", err)
+		}
+	}
 	return &api.BackupStatus{EtcdVersion: etcdVersion, EtcdRevision: rev, LastSuccessDate: now}, nil
 }

pkg/controller/backup-operator/s3_backup.go

@@ -30,7 +30,7 @@ import (
 
 // TODO: replace this with generic backend interface for other options (PV, Azure)
 // handleS3 saves etcd cluster's backup to specificed S3 path.
-func handleS3(ctx context.Context, kubecli kubernetes.Interface, s *api.S3BackupSource, endpoints []string, clientTLSSecret, namespace string) (*api.BackupStatus, error) {
+func handleS3(ctx context.Context, kubecli kubernetes.Interface, s *api.S3BackupSource, endpoints []string, clientTLSSecret, namespace string, maxBackup int) (*api.BackupStatus, error) {
 	// TODO: controls NewClientFromSecret with ctx. This depends on upstream kubernetes to support API calls with ctx.
 	cli, err := s3factory.NewClientFromSecret(kubecli, namespace, s.Endpoint, s.AWSSecret)
 	if err != nil {
@@ -49,5 +49,11 @@ func handleS3(ctx context.Context, kubecli kubernetes.Interface, s *api.S3Backup
 	if err != nil {
 		return nil, fmt.Errorf("failed to save snapshot (%v)", err)
 	}
+	if maxBackup > 0 {
+		err := bm.EnsureMaxbackup(ctx, s.Path, maxBackup)
+		if err != nil {
+			return nil, fmt.Errorf("succeeded in saving snapshot but failed to delete old snapshot (%v)", err)
+		}
+	}
 	return &api.BackupStatus{EtcdVersion: etcdVersion, EtcdRevision: rev, LastSuccessDate: now}, nil
 }

pkg/controller/backup-operator/sync.go

@@ -250,6 +250,10 @@ func (b *Backup) handleBackup(parentContext *context.Context, spec *api.BackupSp
 	if spec.BackupPolicy != nil && spec.BackupPolicy.TimeoutInSecond > 0 {
 		backupTimeout = time.Duration(spec.BackupPolicy.TimeoutInSecond) * time.Second
 	}
+	backupMaxCount := 0
+	if spec.BackupPolicy != nil && spec.BackupPolicy.MaxBackups > 0 {
+		backupMaxCount = spec.BackupPolicy.MaxBackups
+	}
 
 	if parentContext == nil {
 		tmpParent := context.Background()
@@ -259,19 +263,19 @@ func (b *Backup) handleBackup(parentContext *context.Context, spec *api.BackupSp
 	defer cancel()
 	switch spec.StorageType {
 	case api.BackupStorageTypeS3:
-		bs, err := handleS3(ctx, b.kubecli, spec.S3, spec.EtcdEndpoints, spec.ClientTLSSecret, b.namespace)
+		bs, err := handleS3(ctx, b.kubecli, spec.S3, spec.EtcdEndpoints, spec.ClientTLSSecret, b.namespace, backupMaxCount)
 		if err != nil {
 			return nil, err
 		}
 		return bs, nil
 	case api.BackupStorageTypeABS:
-		bs, err := handleABS(ctx, b.kubecli, spec.ABS, spec.EtcdEndpoints, spec.ClientTLSSecret, b.namespace)
+		bs, err := handleABS(ctx, b.kubecli, spec.ABS, spec.EtcdEndpoints, spec.ClientTLSSecret, b.namespace, backupMaxCount)
 		if err != nil {
 			return nil, err
 		}
 		return bs, nil
 	case api.BackupStorageTypeGCS:
-		bs, err := handleGCS(ctx, b.kubecli, spec.GCS, spec.EtcdEndpoints, spec.ClientTLSSecret, b.namespace)
+		bs, err := handleGCS(ctx, b.kubecli, spec.GCS, spec.EtcdEndpoints, spec.ClientTLSSecret, b.namespace, backupMaxCount)
 		if err != nil {
 			return nil, err
 		}
@@ -287,8 +291,13 @@ func validate(spec *api.BackupSpec) error {
 	if len(spec.EtcdEndpoints) == 0 {
 		return errors.New("spec.etcdEndpoints should not be empty")
 	}
-	if spec.BackupPolicy != nil && spec.BackupPolicy.BackupIntervalInSecond < 0 {
-		return errros.New("spec.backupPoloicy.backupIntervalInSecond should not be lower than 0")
+	if spec.BackupPolicy != nil {
+		if spec.BackupPolicy.BackupIntervalInSecond < 0 {
+			return errors.New("spec.backupPoloicy.backupIntervalInSecond should not be lower than 0")
+		}
+		if spec.BackupPolicy.MaxBackups < 0 {
+			return errors.New("spec.backupPolicy.MaxBackups should not be lower than 0")
+		}
 	}
 	return nil
 }