add restore doc and fix restore bug

Signed-off-by: Xieql <xieqianglong@huawei.com>

docs/content/en/docs/fleet-manager/backup/restore.md

@@ -0,0 +1,278 @@
+---
+title: "Unified Restore"
+linkTitle: "Unified Restore"
+weight: 30
+description: >
+ A comprehensive guide on Kurator's Unified Restore solution, detailing the methodology and steps for data recovery in a distributed cloud-native infrastructure.
+---
+
+Unified Restore offers a streamlined method to restore application and cluster resources across all clusters in Fleet, based on backups produced using the Unified Backup approach. 
+This feature aids in recovery, aiming for reduced downtime and operational effectiveness.
+
+## Introduction
+
+## Restore Overview
+
+Kurator's Unified Restore feature is designed around the Unified Backup object. 
+Based on the type of backup chosen, either Immediate or Scheduled, the restoration method is determined accordingly.
+
+### Restore from an **Immediate Backup**
+
+- **Use Case**: Responding to sudden data losses or application issues.
+- **Referred Backup**: The specific **Immediate Backup** designated by the user.
+- **Restore Result**: Restoration from the selected backup into designated clusters.
+
+### Restore from a **Scheduled Backup**
+
+- **Use Case**: Restoring to a recent state in scenarios such as post-accidental data modifications, compliance verifications, or disaster recovery after unforeseen system failures.
+- **Referred Backup**: When a **Scheduled Backup** is selected, Kurator will automatically target the latest successful backup within that series.
+- **Restore Result**: Restoration using the latest successful backup to maintain data relevance and integrity.
+
+### Advanced Backup Options
+
+#### Specific Cluster Restore within Fleet:
+
+Users can specify clusters as the restore destination.
+However, these selected clusters must be a subset of those included in the backup.
+This is because the restore process relies on the data from the backup.
+
+**Note**: To restore resources from one cluster to another not in the original backup, utilize the Unified Migration feature, detailed in a later section.
+
+#### Resource Filtering:
+Users can apply a secondary filter to the data from the backup, enabling selective restoration.
+Define the scope using attributes like backup name, namespace, or label to ensure only desired data is restored. For details, refer to the [Fleet API](https://kurator.dev/docs/references/fleet-api/#fleet)
+
+## How to Perform a Unified Restore
+
+### Pre-requisites
+
+Before diving into the restore steps, ensure that:
+
+- You have successfully installed the backup plugin as described in the [backup plugin installation guide](/docs/fleet-manager/backup/backup-plugin).
+- You have correctly configured `fleet` and `attachedcluster` based on the instructions from the previous guide.
+
+**Note:** The examples provided in this section correspond directly with those outlined in the [unified backup](/docs/fleet-manager/backup/restore) documentation.
+
+### Steps to Follow
+
+1. **Backup Creation**: This step involves setting up a backup using the existed config. More details can be found in [unified backup](/docs/fleet-manager/backup/restore)
+
+2. **Disaster Simulation**: This involves activities that represent scenarios of data losses or application disruptions.
+
+3. **Restore Execution**: Based on the previously created backup, execute restore config.
+
+4. **Restore Object Review**: This involves examining the details about restore object.
+
+Throughout these operations, users can check the pod's status using the commands below. 
+This will help in verifying the initial state of the pod, confirming if the pod was lost due to the simulated disaster, and ascertaining if the pod was restored through unified recovery:
+
+```console
+kubectl get po -n kurator-backup --kubeconfig=/root/.kube/kurator-member1.config
+kubectl get po -n kurator-backup --kubeconfig=/root/.kube/kurator-member2.config
+```
+
+### 1. Restore from an Immediate Backup
+
+#### Backup Immediate Creation
+
+Deploy a test application using the following command:
+
+```console
+kubectl apply -f examples/backup/app-backup-demo.yaml 
+```
+
+Trigger an immediate backup using the command:
+
+```console
+kubectl apply -f examples/backup/backup-select-labels.yaml
+```
+
+#### Disaster Simulation
+
+Simulate a disaster event that deletes all previously installed resources using the command:
+
+```console
+kubectl delete applications.apps.kurator.dev unified-backup-demo 
+```
+
+#### Restore Execution
+
+Apply a restore action that refers to the backup created earlier:
+
+```console
+k apply -f examples/backup/restore-minimal.yaml
+```
+
+After executing, you'll observe that the `busybox` pod is restored in two clusters.
+
+
+#### Examine the Restore Object
+
+Use the following command:
+
+```console
+kubectl get restores.backup.kurator.dev minimal -o yaml
+```
+
+You can expect the output to resemble the provided structure. 
+The status section displays the processing status for the two clusters within the fleet.
+
+```console
+apiVersion: backup.kurator.dev/v1alpha1
+kind: Restore
+metadata:
+  annotations:
+    kubectl.kubernetes.io/last-applied-configuration: |
+      {"apiVersion":"backup.kurator.dev/v1alpha1","kind":"Restore","metadata":{"annotations":{},"name":"minimal","namespace":"default"},"spec":{"backupName":"select-labels"}}
+  creationTimestamp: "2023-10-28T09:24:05Z"
+  finalizers:
+  - restore.kurator.dev
+  generation: 1
+  name: minimal
+  namespace: default
+  resourceVersion: "9441827"
+  uid: 6cf6154b-4a68-4431-8366-354bd3cb6250
+spec:
+  backupName: select-labels
+status:
+  restoreDetails:
+  - clusterKind: AttachedCluster
+    clusterName: kurator-member1
+    restoreNameInCluster: kurator-member1-restore-default-minimal
+    restoreStatusInCluster:
+      completionTimestamp: "2023-10-28T09:24:07Z"
+      phase: Completed
+      progress:
+        itemsRestored: 2
+        totalItems: 2
+      startTimestamp: "2023-10-28T09:24:05Z"
+  - clusterKind: AttachedCluster
+    clusterName: kurator-member2
+    restoreNameInCluster: kurator-member2-restore-default-minimal
+    restoreStatusInCluster:
+      completionTimestamp: "2023-10-28T09:24:07Z"
+      phase: Completed
+      progress:
+        itemsRestored: 2
+        totalItems: 2
+      startTimestamp: "2023-10-28T09:24:05Z"
+```
+
+**Interpreting the Output:**
+
+- **The spec `destination` field:** The absence of a `destination` field indicates that the restore process will occur on all clusters where the backup was executed. Consequently, the `busybox` pod is successfully restored on both clusters.
+
+- **The spec `policy` field:** The absence of a `policy` field means that the restore strategy is entirely based on the initial backup without any secondary filtering.
+
+- **The `status` section:** This provides insights into the processing status of the two clusters.
+
+### 2. Restore from a Scheduled Backup
+
+Deploy a test application using the following command:
+
+```console
+kubectl apply -f examples/backup/app-backup-demo.yaml 
+```
+
+Trigger a schedule backup using the command:
+
+```console
+kubectl apply -f examples/backup/backup-schedule.yaml
+```
+
+> Please note: since scheduled backups aren't executed immediately, you'll need to wait at least 5 minutes for a backup to complete before proceeding with the subsequent steps.
+
+```console
+kubectl get backups.backup.kurator.dev schedule -o yaml
+```
+
+#### Disaster Simulation
+
+Simulate a disaster event that deletes all previously installed resources using the command:
+
+```console
+kubectl delete applications.apps.kurator.dev unified-backup-demo 
+```
+
+#### Restore Execution
+
+Apply a restore action that refers to the backup created earlier:
+
+```console
+kubectl apply -f examples/backup/restore-schedule.yaml
+```
+
+After executing, you'll observe that the `busybox` pod is only restored in the second cluster.
+
+#### Examine the Restore Object
+
+Use the following command:
+
+```console
+kubectl get restores.backup.kurator.dev schedule -o yaml
+```
+
+You can expect the output to resemble the provided structure.
+The status section displays the processing status for the two clusters within the fleet.
+
+```console
+apiVersion: backup.kurator.dev/v1alpha1
+kind: Restore
+metadata:
+  annotations:
+    kubectl.kubernetes.io/last-applied-configuration: |
+      {"apiVersion":"backup.kurator.dev/v1alpha1","kind":"Restore","metadata":{"annotations":{},"name":"schedule","namespace":"default"},"spec":{"backupName":"schedule","destination":{"clusters":[{"kind":"AttachedCluster","name":"kurator-member2"}],"fleet":"quickstart"},"policy":{"resourceFilter":{"labelSelector":{"matchLabels":{"app":"busybox"}}}}}}
+  creationTimestamp: "2023-10-28T13:27:03Z"
+  finalizers:
+  - restore.kurator.dev
+  generation: 1
+  name: schedule
+  namespace: default
+  resourceVersion: "9485496"
+  uid: 1eab7122-a6bc-4b64-8196-b512078abfa0
+spec:
+  backupName: schedule
+  destination:
+    clusters:
+    - kind: AttachedCluster
+      name: kurator-member2
+    fleet: quickstart
+  policy:
+    resourceFilter:
+      labelSelector:
+        matchLabels:
+          app: busybox
+status:
+  restoreDetails:
+  - clusterKind: AttachedCluster
+    clusterName: kurator-member2
+    restoreNameInCluster: kurator-member2-restore-default-schedule
+    restoreStatusInCluster:
+      completionTimestamp: "2023-10-28T13:27:10Z"
+      phase: Completed
+      progress: {}
+      startTimestamp: "2023-10-28T13:27:03Z"
+```
+
+**Interpreting the Output:**
+
+Notice that the backup referenced here is "schedule", which backed up all cluster resources. However, the restore here did not use the "minimal" method to execute the default full restore strategy. 
+Instead, a second filter was applied during the restore phase.
+
+- **The spec `destination` field:** Here, only the second cluster is specified, so the resources won't be restored in all clusters, but only in the second one.
+
+- **The spec `policy` field:** It specifies `app: busybox`, so only resources with this label will be restored.
+
+- **The `status` section:** As it's a scheduled backup, the status here shows that the current restore is targeting the most recent successful backup. This provides insights into the processing status of restore.
+
+### Cleanup
+
+To remove the backup examples used for testing, execute:
+
+```console
+kubectl delete backups.backup.kurator.dev specific-ns schedule-matchlabels
+```
+
+> Please note: This command only deletes the current object in the k8s API.
+For data security, deletion of the data in object storage should be performed using the tools provided by the object storage solution you adopted.
+Kurator currently does not offer capabilities to delete data inside the object storage.

examples/backup/app-backup-demo.yaml

@@ -16,6 +16,6 @@ spec:
         fleet: quickstart
       kustomization:
         interval: 5m0s
-        path: ./examples/fleet/backup/busybox-demo
+        path: ./examples/backup/busybox-demo
         prune: true
         timeout: 2m0s

examples/backup/restore-minimal.yaml

@@ -0,0 +1,7 @@
+apiVersion: backup.kurator.dev/v1alpha1
+kind: Restore
+metadata:
+  name: minimal
+  namespace: default
+spec:
+  backupName: select-labels

examples/backup/restore-schedule.yaml

@@ -0,0 +1,17 @@
+apiVersion: backup.kurator.dev/v1alpha1
+kind: Restore
+metadata:
+  name: schedule
+  namespace: default
+spec:
+  backupName: schedule
+  destination:
+    fleet: quickstart
+    clusters:
+      - kind: AttachedCluster
+        name: kurator-member2
+  policy:
+    resourceFilter:
+      labelSelector:
+        matchLabels:
+          app: busybox

pkg/fleet-manager/restore_controller.go

@@ -92,8 +92,23 @@ func (r *RestoreManager) reconcileRestore(ctx context.Context, restore *backupap
 		return ctrl.Result{}, err
 	}
 
+	// Fetch the backup object referred to by the restore. if not found, return directly.
+	key := client.ObjectKey{
+		Name:      restore.Spec.BackupName,
+		Namespace: restore.Namespace,
+	}
+	referredBackup := &backupapi.Backup{}
+	if err := r.Client.Get(ctx, key, referredBackup); err != nil {
+		log.Error(err, "Failed to get backup object", "backupName", restore.Spec.BackupName)
+		return ctrl.Result{}, nil
+	}
+
 	// Apply restore resource in target clusters
-	result, err := r.reconcileRestoreResources(ctx, restore, destinationClusters, fleetName)
+	result, err := r.reconcileRestoreResources(ctx, restore, referredBackup, destinationClusters, fleetName)
+	if err == ErrNoCompletedBackups {
+		log.Error(err, "no completed Velero backups available for restore. stop reconcile", "referred backup", restore.Spec.BackupName)
+		return result, nil
+	}
 	if err != nil {
 		return result, err
 	}
@@ -112,15 +127,22 @@ func (r *RestoreManager) reconcileRestore(ctx context.Context, restore *backupap
 	}
 }
 
+var ErrNoCompletedBackups = errors.New("No completed Velero backups available for restore.")
+
 // reconcileRestoreResources converts the restore resources into velero restore resources on the target clusters, and applies those velero restore resources.
-func (r *RestoreManager) reconcileRestoreResources(ctx context.Context, restore *backupapi.Restore, destinationClusters map[ClusterKey]*fleetCluster, fleetName string) (ctrl.Result, error) {
+func (r *RestoreManager) reconcileRestoreResources(ctx context.Context, restore *backupapi.Restore, referredBackup *backupapi.Backup, destinationClusters map[ClusterKey]*fleetCluster, fleetName string) (ctrl.Result, error) {
 	log := ctrl.LoggerFrom(ctx)
 
 	restoreLabels := generateVeleroInstanceLabel(RestoreNameLabel, restore.Name, fleetName)
 
 	var tasks []func() error
 	for clusterKey, clusterAccess := range destinationClusters {
-		veleroBackupName := generateVeleroResourceName(clusterKey.Name, BackupKind, restore.Namespace, restore.Spec.BackupName)
+		// veleroBackupName is depends on the referred backup type, immediate or schedule.
+		veleroBackupName, err := r.fetchRestoreVeleroBackupName(ctx, restore, referredBackup, clusterAccess, clusterKey.Name, BackupKind, restore.Namespace, restore.Spec.BackupName)
+		if err != nil {
+			return ctrl.Result{}, ErrNoCompletedBackups
+		}
+
 		veleroRestoreName := generateVeleroResourceName(clusterKey.Name, RestoreKind, restore.Namespace, restore.Name)
 		veleroRestore := buildVeleroRestoreInstance(&restore.Spec, restoreLabels, veleroBackupName, veleroRestoreName)
 
@@ -171,3 +193,34 @@ func (r *RestoreManager) reconcileDeleteRestore(ctx context.Context, restore *ba
 
 	return ctrl.Result{}, nil
 }
+
+// fetchRestoreVeleroBackupName retrieves the name of the Velero backup associated with the provided restore.
+// If the referred backup is an immediate backup, it returns the generated Velero backup name.
+// If the referred backup is a scheduled backup, it fetches the name of the most recent completed backup.
+func (r *RestoreManager) fetchRestoreVeleroBackupName(ctx context.Context, restore *backupapi.Restore, referredBackup *backupapi.Backup, clusterAccess *fleetCluster, clusterName, creatorKind, creatorNamespace, creatorName string) (string, error) {
+	log := ctrl.LoggerFrom(ctx)
+
+	// Return the generated name directly if it's an immediate backup.
+	if !isScheduleBackup(referredBackup) {
+		return generateVeleroResourceName(clusterName, BackupKind, referredBackup.Namespace, referredBackup.Name), nil
+	}
+
+	// Fetch the most recent completed backup name if it's a scheduled backup.
+	veleroScheduleName := generateVeleroResourceName(clusterName, BackupKind, referredBackup.Namespace, referredBackup.Name)
+	backupList := &velerov1.BackupList{}
+	err := listResourcesFromClusterClient(ctx, VeleroNamespace, velerov1.ScheduleNameLabel, veleroScheduleName, *clusterAccess, backupList)
+	if err != nil {
+		log.Error(err, "Unable to list velero backups for schedule", "scheduleName", veleroScheduleName)
+		return "", err
+	}
+
+	// Return an error if no completed backups are found for the referred schedule.
+	veleroBackup := MostRecentCompletedBackup(backupList.Items)
+	if veleroBackup.Name == "" {
+		errMsg := fmt.Sprintf("No completed backups found for referred schedule backup: %s", restore.Spec.BackupName)
+		log.Error(errors.New(errMsg), "")
+		return "", errors.New(errMsg)
+	}
+
+	return veleroBackup.Name, nil
+}