fix: propagate snapshot close error, improve endpoint tracing, rename endpoint to host

Author: cwrau

api/v1alpha1/hostedcontrolplane.go

@@ -52,12 +52,12 @@ func (a *Audit) ModeOrDefault() string {
 	return ptr.Deref(a.Mode, defaultMode)
 }
 
-func (ebs *ETCDBackupSecret) EndpointKeyOrDefault() string {
-	defaultKey := "endpoint"
+func (ebs *ETCDBackupSecret) HostKeyOrDefault() string {
+	defaultKey := "host"
 	if ebs == nil {
 		return defaultKey
 	}
-	return ptr.Deref(ebs.EndpointKey, defaultKey)
+	return ptr.Deref(ebs.HostKey, defaultKey)
 }
 
 func (ebs *ETCDBackupSecret) RegionKeyOrDefault() string {

api/v1alpha1/hostedcontrolplane_types.go

@@ -169,7 +169,7 @@ type ETCDBackupSecret struct {
 	//+kubebuilder:validation:Optional
 	SecretAccessKeyKey *string `json:"secretAccessKeyKey,omitempty"`
 	//+kubebuilder:validation:Optional
-	EndpointKey *string `json:"endpointKey,omitempty"`
+	HostKey *string `json:"hostKey,omitempty"`
 	//+kubebuilder:validation:Optional
 	RegionKey *string `json:"regionKey,omitempty"`
 	// The bucket will be split by `/` and only the first part is the bucket,

api/v1alpha1/zz_generated.deepcopy.go

@@ -34,7 +34,7 @@ var (
 
 type EtcdClient interface {
 	GetStatuses(ctx context.Context) (map[string]*clientv3.StatusResponse, error)
-	CreateSnapshot(ctx context.Context) (*clientv3.SnapshotResponse, func(), error)
+	CreateSnapshot(ctx context.Context) (*clientv3.SnapshotResponse, func() error, error)
 	ListAlarms(ctx context.Context) (*clientv3.AlarmResponse, error)
 	DisarmAlarm(ctx context.Context, alarm *clientv3.AlarmMember) error
 }
@@ -108,19 +108,24 @@ func (e *etcdClient) GetStatuses(ctx context.Context) (map[string]*clientv3.Stat
 	)
 }
 
-func (e *etcdClient) CreateSnapshot(ctx context.Context) (*clientv3.SnapshotResponse, func(), error) {
+func (e *etcdClient) CreateSnapshot(ctx context.Context) (*clientv3.SnapshotResponse, func() error, error) {
 	return tracing.WithSpan3(ctx, tracer, "EtcdClient.CreateSnapshot",
-		func(ctx context.Context, span trace.Span) (_ *clientv3.SnapshotResponse, _ func(), retErr error) {
+		func(ctx context.Context, span trace.Span) (_ *clientv3.SnapshotResponse, _ func() error, retErr error) {
 			endpoint := e.anyEndpoint
 			etcdClient, err := createEtcdClient(e, endpoint)
-			closeFunc := func() { closeEtcdClient(etcdClient, &retErr, endpoint) }
+			closeFunc := func() error {
+				var closeErr error
+				closeEtcdClient(etcdClient, &closeErr, endpoint)
+				return closeErr
+			}
 			if err != nil {
 				return nil, closeFunc, err
 			}
 
 			snapshotResponse, err := callETCDFuncOnAnyMember(
 				ctx,
 				etcdClient,
+				endpoint,
 				clientv3.Client.SnapshotWithVersion,
 			)
 			return snapshotResponse, closeFunc, err
@@ -183,6 +188,7 @@ func (e *etcdClient) ListAlarms(ctx context.Context) (*clientv3.AlarmResponse, e
 			return callETCDFuncOnAnyMember(
 				ctx,
 				etcdClient,
+				endpoint,
 				clientv3.Client.AlarmList,
 			)
 		},
@@ -203,6 +209,7 @@ func (e *etcdClient) DisarmAlarm(ctx context.Context, alarm *clientv3.AlarmMembe
 			_, err = callETCDFuncOnAnyMember(
 				ctx,
 				etcdClient,
+				endpoint,
 				func(client clientv3.Client, ctx context.Context) (*clientv3.AlarmResponse, error) {
 					response, err := client.AlarmDisarm(ctx, alarm)
 					if err != nil {
@@ -313,11 +320,12 @@ func callETCDFuncOnMember[R any](
 func callETCDFuncOnAnyMember[R any](
 	ctx context.Context,
 	etcdClient *clientv3.Client,
+	endpoint string,
 	etcdFunc func(client clientv3.Client, ctx context.Context) (R, error),
 ) (R, error) {
 	return tracing.WithSpan(ctx, tracer, "CallETCDFuncOnAnyMember",
 		func(ctx context.Context, span trace.Span) (R, error) {
-			return callETCDFuncOnMember(ctx, etcdClient, "",
+			return callETCDFuncOnMember(ctx, etcdClient, endpoint,
 				func(client clientv3.Client, ctx context.Context, _ string) (R, error) {
 					return etcdFunc(client, ctx)
 				})

pkg/reconcilers/etcd_cluster/etcd_client/etcd_client.go

@@ -319,7 +319,7 @@ func (er *etcdClusterReconciler) reconcileETCDBackup(
 					"Starting etcd backup",
 				)
 				snapshotResponse, closeClientFunc, err := etcdClient.CreateSnapshot(ctx)
-				defer closeClientFunc()
+				defer func() { err = errors.Join(err, closeClientFunc()) }()
 				if err != nil {
 					return fmt.Errorf("failed to create etcd snapshot: %w", err)
 				}

pkg/reconcilers/etcd_cluster/reconciler.go

@@ -76,7 +76,7 @@ func NewS3Client(
 		secretName := etcdBackupSecretConfig.Name
 		accessKeyIDKey := etcdBackupSecretConfig.AccessKeyIDKeyOrDefault()
 		secretAccessKeyKey := etcdBackupSecretConfig.SecretAccessKeyKeyOrDefault()
-		endpointKey := etcdBackupSecretConfig.EndpointKeyOrDefault()
+		endpointKey := etcdBackupSecretConfig.HostKeyOrDefault()
 		regionKey := etcdBackupSecretConfig.RegionKeyOrDefault()
 		bucketKey := etcdBackupSecretConfig.BucketKeyOrDefault()
 		spanAttributes := []attribute.KeyValue{
@@ -119,6 +119,9 @@ func NewS3Client(
 
 		parts := strings.SplitN(bucketString, "/", 2)
 		bucket := parts[0]
+		if bucket == "" {
+			return nil, fmt.Errorf("bucket secret key %q resolved to an empty bucket name", bucketKey)
+		}
 		prefix := ""
 		if len(parts) == 2 {
 			prefix = parts[1]

pkg/reconcilers/etcd_cluster/s3_client/s3_client.go

@@ -38,14 +38,14 @@ func (s *EtcdClientStub) GetStatuses(_ context.Context) (map[string]*clientv3.St
 	return s.StatusResponses, nil
 }
 
-func (s *EtcdClientStub) CreateSnapshot(_ context.Context) (*clientv3.SnapshotResponse, func(), error) {
+func (s *EtcdClientStub) CreateSnapshot(_ context.Context) (*clientv3.SnapshotResponse, func() error, error) {
 	if s.SnapshotError != nil {
-		return nil, func() {}, s.SnapshotError
+		return nil, func() error { return nil }, s.SnapshotError
 	}
 	return &clientv3.SnapshotResponse{
 		Header:   &etcdserverpb.ResponseHeader{ClusterId: 1},
 		Snapshot: io.NopCloser(bytes.NewReader(EtcdSnapshotData)),
-	}, func() {}, nil
+	}, func() error { return nil }, nil
 }
 
 func (s *EtcdClientStub) ListAlarms(_ context.Context) (*clientv3.AlarmResponse, error) {