implement volume snapshot backups and restore

charts/tembo-operator/Chart.yaml

@@ -3,10 +3,10 @@ name: tembo-operator
 description: 'Helm chart to deploy the tembo-operator'
 type: application
 icon: https://cloud.tembo.io/images/TemboElephant.png
-version: 0.3.0
+version: 0.3.1
 home: https://tembo.io
 sources:
-  - https://github.com/tembo-io/tembo-stacks
+  - https://github.com/tembo-io/tembo
   - https://github.com/cloudnative-pg/cloudnative-pg
 keywords:
   - postgresql

charts/tembo-operator/templates/crd.yaml

@@ -1344,6 +1344,8 @@ spec:
                   endpointURL: null
                   s3Credentials:
                     inheritFromIAMRole: true
+                  volumeSnapshot:
+                    enabled: false
                 description: |-
                   The backup configuration for the CoreDB instance to facilitate database backups and WAL archive uploads to an S3 compatible object store.
 
@@ -1432,6 +1434,22 @@ spec:
                     description: The backup schedule set with cron syntax
                     nullable: true
                     type: string
+                  volumeSnapshot:
+                    default:
+                      enabled: false
+                    description: Enable using Volume Snapshots for backups instead of Object Storage
+                    nullable: true
+                    properties:
+                      enabled:
+                        description: Enable the volume snapshots for backups
+                        type: boolean
+                      snapshotClass:
+                        description: The reference to the snapshot class
+                        nullable: true
+                        type: string
+                    required:
+                    - enabled
+                    type: object
                 type: object
               connectionPooler:
                 default:

tembo-operator/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "controller"
 description = "Tembo Operator for Postgres"
-version = "0.35.4"
+version = "0.36.0"
 edition = "2021"
 default-run = "controller"
 license = "Apache-2.0"

tembo-operator/src/apis/coredb_types.rs

@@ -157,11 +157,29 @@ pub struct S3CredentialsSessionToken {
     pub name: String,
 }
 
+/// VolumeSnapshots is the type for the configuration of the volume snapshots
+/// to be used for backups instead of object storage
+#[derive(Serialize, Deserialize, Clone, Debug, Default, JsonSchema)]
+pub struct VolumeSnapshot {
+    /// Enable the volume snapshots for backups
+    pub enabled: bool,
+
+    /// The reference to the snapshot class
+    #[serde(
+        default,
+        skip_serializing_if = "Option::is_none",
+        rename = "snapshotClass"
+    )]
+    pub snapshot_class: Option<String>,
+}
+
 /// CoreDB Backup configuration
 /// The backup configuration for the CoreDB instance to facilitate database
-/// backups and WAL archive uploads to an S3 compatible object store.
+/// backups uploads to an S3 compatible object store or using Volume Snapshots
+/// For WAL archive uploads utilite an S3 compatible object store.
 ///
 /// **Example**: A typical S3 backup configuration using IAM Role for authentication
+/// with Volume Snapshots enabled
 ///
 /// See `ServiceAccountTemplate` for to map the IAM role ARN to a Kubernetes service account.
 ///
@@ -178,6 +196,9 @@ pub struct S3CredentialsSessionToken {
 ///     s3Credentials:
 ///       inheritFromIAMRole: true
 ///     schedule: "0 0 * * *" #every day at midnight
+///     volumeSnapshots:
+///       enabled: true
+///       snapshotClass: my-snapshot-class-name
 /// ```
 #[derive(Deserialize, Serialize, Clone, Debug, Default, JsonSchema)]
 #[allow(non_snake_case)]
@@ -205,6 +226,13 @@ pub struct Backup {
     /// The S3 credentials to use for backups (if not using IAM Role)
     #[serde(default = "defaults::default_s3_credentials", rename = "s3Credentials")]
     pub s3_credentials: Option<S3Credentials>,
+
+    /// Enable using Volume Snapshots for backups instead of Object Storage
+    #[serde(
+        default = "defaults::default_volume_snapshot",
+        rename = "volumeSnapshot"
+    )]
+    pub volume_snapshot: Option<VolumeSnapshot>,
 }
 
 /// Restore configuration provides a way to restore a database from a backup

tembo-operator/src/cloudnativepg/clusters.rs

@@ -1157,7 +1157,7 @@ pub enum ClusterBackupTarget {
 }
 
 /// VolumeSnapshot provides the configuration for the execution of volume snapshot backups.
-#[derive(Serialize, Deserialize, Clone, Debug, Default, JsonSchema)]
+#[derive(Serialize, Deserialize, Clone, Debug, Default, JsonSchema, PartialEq)]
 pub struct ClusterBackupVolumeSnapshot {
     /// Annotations key-value pairs that will be added to .metadata.annotations snapshot resources.
     #[serde(default, skip_serializing_if = "Option::is_none")]
@@ -1202,7 +1202,7 @@ pub struct ClusterBackupVolumeSnapshot {
 }
 
 /// Configuration parameters to control the online/hot backup with volume snapshots
-#[derive(Serialize, Deserialize, Clone, Debug, Default, JsonSchema)]
+#[derive(Serialize, Deserialize, Clone, Debug, Default, JsonSchema, PartialEq)]
 pub struct ClusterBackupVolumeSnapshotOnlineConfiguration {
     /// Control whether the I/O workload for the backup initial checkpoint will be limited, according to the `checkpoint_completion_target` setting on the PostgreSQL server. If set to true, an immediate checkpoint will be used, meaning PostgreSQL will complete the checkpoint as soon as possible. `false` by default.
     #[serde(
@@ -1221,7 +1221,7 @@ pub struct ClusterBackupVolumeSnapshotOnlineConfiguration {
 }
 
 /// VolumeSnapshot provides the configuration for the execution of volume snapshot backups.
-#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema)]
+#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema, PartialEq)]
 pub enum ClusterBackupVolumeSnapshotSnapshotOwnerReference {
     #[serde(rename = "none")]
     None,

tembo-operator/src/cloudnativepg/cnpg.rs

@@ -15,9 +15,12 @@ use crate::{
             ClusterBackupBarmanObjectStoreS3CredentialsSecretAccessKey,
             ClusterBackupBarmanObjectStoreS3CredentialsSessionToken,
             ClusterBackupBarmanObjectStoreWal, ClusterBackupBarmanObjectStoreWalCompression,
-            ClusterBackupBarmanObjectStoreWalEncryption, ClusterBootstrap, ClusterBootstrapInitdb,
-            ClusterBootstrapRecovery, ClusterBootstrapRecoveryRecoveryTarget, ClusterCertificates,
-            ClusterExternalClusters, ClusterExternalClustersBarmanObjectStore,
+            ClusterBackupBarmanObjectStoreWalEncryption, ClusterBackupVolumeSnapshot,
+            ClusterBackupVolumeSnapshotOnlineConfiguration,
+            ClusterBackupVolumeSnapshotSnapshotOwnerReference, ClusterBootstrap,
+            ClusterBootstrapInitdb, ClusterBootstrapRecovery,
+            ClusterBootstrapRecoveryRecoveryTarget, ClusterCertificates, ClusterExternalClusters,
+            ClusterExternalClustersBarmanObjectStore,
             ClusterExternalClustersBarmanObjectStoreS3Credentials,
             ClusterExternalClustersBarmanObjectStoreS3CredentialsAccessKeyId,
             ClusterExternalClustersBarmanObjectStoreS3CredentialsRegion,
@@ -41,7 +44,7 @@ use crate::{
         },
         scheduledbackups::{
             ScheduledBackup, ScheduledBackupBackupOwnerReference, ScheduledBackupCluster,
-            ScheduledBackupSpec,
+            ScheduledBackupMethod, ScheduledBackupSpec,
         },
     },
     config::Config,
@@ -66,6 +69,8 @@ use std::{collections::BTreeMap, sync::Arc};
 use tokio::time::Duration;
 use tracing::{debug, error, info, instrument, warn};
 
+const VOLUME_SNAPSHOT_CLASS_NAME: &str = "cnpg-snapshot-class";
+
 pub struct PostgresConfig {
     pub postgres_parameters: Option<BTreeMap<String, String>>,
     pub shared_preload_libraries: Option<Vec<String>>,
@@ -103,7 +108,7 @@ fn create_cluster_backup_barman_wal(cdb: &CoreDB) -> Option<ClusterBackupBarmanO
         Some(ClusterBackupBarmanObjectStoreWal {
             compression: Some(ClusterBackupBarmanObjectStoreWalCompression::Snappy),
             encryption,
-            max_parallel: Some(5),
+            max_parallel: Some(8),
         })
     } else {
         None
@@ -152,6 +157,28 @@ fn create_cluster_certificates(cdb: &CoreDB) -> Option<ClusterCertificates> {
     }
 }
 
+fn create_cluster_backup_volume_snapshot(cdb: &CoreDB) -> ClusterBackupVolumeSnapshot {
+    let class_name = cdb
+        .spec
+        .backup
+        .volume_snapshot
+        .as_ref()
+        .and_then(|vs| vs.snapshot_class.as_ref())
+        .cloned()
+        .unwrap_or_else(|| VOLUME_SNAPSHOT_CLASS_NAME.to_string());
+
+    ClusterBackupVolumeSnapshot {
+        class_name: Some(class_name),
+        online: Some(true),
+        online_configuration: Some(ClusterBackupVolumeSnapshotOnlineConfiguration {
+            wait_for_archive: Some(true),
+            immediate_checkpoint: Some(true),
+        }),
+        snapshot_owner_reference: Some(ClusterBackupVolumeSnapshotSnapshotOwnerReference::Cluster),
+        ..ClusterBackupVolumeSnapshot::default()
+    }
+}
+
 fn create_cluster_backup(
     cdb: &CoreDB,
     endpoint_url: &str,
@@ -171,14 +198,23 @@ fn create_cluster_backup(
         },
     };
 
+    let volume_snapshot = cdb.spec.backup.volume_snapshot.as_ref().and_then(|vs| {
+        if vs.enabled {
+            Some(create_cluster_backup_volume_snapshot(cdb))
+        } else {
+            None
+        }
+    });
+
     Some(ClusterBackup {
         barman_object_store: Some(create_cluster_backup_barman_object_store(
             cdb,
             endpoint_url,
             backup_path,
             s3_credentials,
         )),
-        retention_policy: Some(retention_days), // Adjust as needed
+        retention_policy: Some(retention_days),
+        volume_snapshot,
         ..ClusterBackup::default()
     })
 }
@@ -1331,6 +1367,13 @@ fn schedule_expression_from_cdb(cdb: &CoreDB) -> String {
 fn cnpg_scheduled_backup(cdb: &CoreDB) -> ScheduledBackup {
     let name = cdb.name_any();
     let namespace = cdb.namespace().unwrap();
+    let method = cdb.spec.backup.volume_snapshot.as_ref().map(|vs| {
+        if vs.enabled {
+            ScheduledBackupMethod::VolumeSnapshot
+        } else {
+            ScheduledBackupMethod::BarmanObjectStore
+        }
+    });
 
     ScheduledBackup {
         metadata: ObjectMeta {
@@ -1344,6 +1387,7 @@ fn cnpg_scheduled_backup(cdb: &CoreDB) -> ScheduledBackup {
             immediate: Some(true),
             schedule: schedule_expression_from_cdb(cdb),
             suspend: Some(false),
+            method,
             ..ScheduledBackupSpec::default()
         },
         status: None,
@@ -2292,6 +2336,8 @@ mod tests {
             encryption: AES256
             retentionPolicy: "45"
             schedule: 55 7 * * *
+            volumeSnapshot:
+              enabled: false
           image: quay.io/tembo/tembo-pg-cnpg:15.3.0-5-48d489e 
           port: 5432
           postgresExporterEnabled: true
@@ -2323,6 +2369,11 @@ mod tests {
             "45d".to_string()
         );
 
+        assert_eq!(
+            scheduled_backup.spec.method,
+            Some(ScheduledBackupMethod::BarmanObjectStore)
+        );
+
         // Assert to make sure that backup destination path is set
         assert_eq!(
             backup
@@ -2634,4 +2685,68 @@ mod tests {
         let cdb_no_storage_class: CoreDB = from_str(cdb_no_storage_class_yaml).unwrap();
         assert_eq!(cnpg_cluster_storage_class(&cdb_no_storage_class), None);
     }
+
+    #[test]
+    fn test_cnpg_cluster_volume_snapshot() {
+        let cdb_yaml = r#"
+        apiVersion: coredb.io/v1alpha1
+        kind: CoreDB
+        metadata:
+          name: test
+          namespace: default
+        spec:
+          backup:
+            destinationPath: s3://tembo-backup/sample-standard-backup
+            encryption: ""
+            retentionPolicy: "30"
+            schedule: 17 9 * * *
+            endpointURL: http://minio:9000
+            volumeSnapshot:
+              enabled: true
+              snapshotClass: "csi-vsc"
+          image: quay.io/tembo/tembo-pg-cnpg:15.3.0-5-48d489e 
+          port: 5432
+          replicas: 1
+          resources:
+            limits:
+              cpu: "1"
+              memory: 0.5Gi
+          serviceAccountTemplate:
+            metadata:
+              annotations:
+                eks.amazonaws.com/role-arn: arn:aws:iam::012345678901:role/aws-iam-role-iam
+          sharedirStorage: 1Gi
+          stop: false
+          storage: 1Gi
+          storageClass: "gp3-enc"
+          uid: 999
+        "#;
+
+        let cdb: CoreDB = serde_yaml::from_str(cdb_yaml).expect("Failed to parse YAML");
+        let snapshot = create_cluster_backup_volume_snapshot(&cdb);
+        let scheduled_backup = cnpg_scheduled_backup(&cdb);
+
+        // Set an expected ClusterBackupVolumeSnapshot object
+        let expected_snapshot = ClusterBackupVolumeSnapshot {
+            class_name: Some("csi-vsc".to_string()), // Expected to match the YAML input
+            online: Some(true),
+            online_configuration: Some(ClusterBackupVolumeSnapshotOnlineConfiguration {
+                wait_for_archive: Some(true),
+                immediate_checkpoint: Some(true),
+            }),
+            snapshot_owner_reference: Some(
+                ClusterBackupVolumeSnapshotSnapshotOwnerReference::Cluster,
+            ),
+            ..ClusterBackupVolumeSnapshot::default()
+        };
+
+        // Assert to make sure that the snapshot.snapshot_class and expected_snapshot.snapshot_class are the same
+        assert_eq!(snapshot, expected_snapshot);
+
+        // Assert to make sure that the ScheduledBackup method is set to VolumeSnapshot
+        assert_eq!(
+            scheduled_backup.spec.method,
+            Some(ScheduledBackupMethod::VolumeSnapshot)
+        );
+    }
 }

tembo-operator/src/cloudnativepg/scheduledbackups.rs

@@ -71,7 +71,7 @@ pub struct ScheduledBackupCluster {
 }
 
 /// Specification of the desired behavior of the ScheduledBackup. More info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#spec-and-status
-#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema)]
+#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema, PartialEq)]
 pub enum ScheduledBackupMethod {
     #[serde(rename = "barmanObjectStore")]
     BarmanObjectStore,

tembo-operator/src/defaults.rs

@@ -6,7 +6,7 @@ use std::collections::BTreeMap;
 
 use crate::{
     apis::coredb_types::{
-        Backup, ConnectionPooler, PgBouncer, S3Credentials, ServiceAccountTemplate,
+        Backup, ConnectionPooler, PgBouncer, S3Credentials, ServiceAccountTemplate, VolumeSnapshot,
     },
     cloudnativepg::poolers::{PoolerPgbouncerPoolMode, PoolerTemplateSpecContainersResources},
     extensions::types::{Extension, TrunkInstall},
@@ -139,6 +139,7 @@ pub fn default_backup() -> Backup {
         retentionPolicy: default_retention_policy(),
         schedule: default_backup_schedule(),
         s3_credentials: default_s3_credentials(),
+        volume_snapshot: default_volume_snapshot(),
         ..Default::default()
     }
 }
@@ -210,3 +211,10 @@ pub fn default_s3_credentials() -> Option<S3Credentials> {
         ..Default::default()
     })
 }
+
+pub fn default_volume_snapshot() -> Option<VolumeSnapshot> {
+    Some(VolumeSnapshot {
+        enabled: false,
+        snapshot_class: None,
+    })
+}