Merge 2a12a52211 into e30154b938

Upgrade barman-cloud library to v0.5.1 which includes lz4 support for
base backups. Update CRD schema and manifest to expose lz4 as a valid
compression option, and clarify docs that lz4 is no longer WAL-only.

Closes #867

Signed-off-by: Marco Nenciarini <marco.nenciarini@enterprisedb.com>
Signed-off-by: Leonardo Cecchi <leonardo.cecchi@gmail.com>
Co-authored-by: Leonardo Cecchi <leonardo.cecchi@gmail.com>

.wordlist.txt

@@ -15,6 +15,7 @@ DigitalOcean
 Docusaurus
 EDB
 EKS
+EnterpriseDB
 Enum
 EnvVar
 GCP
@@ -76,6 +77,7 @@ backends
 barmanObjectName
 barmanObjectStore
 barmancloud
+benchmarked
 boto
 bzip
 cd

config/crd/bases/barmancloud.cnpg.io_objectstores.yaml

@@ -144,10 +144,11 @@ spec:
                         description: |-
                           Compress a backup file (a tar file per tablespace) while streaming it
                           to the object store. Available options are empty string (no
-                          compression, default), `gzip`, `bzip2`, and `snappy`.
+                          compression, default), `gzip`, `bzip2`, `lz4`, and `snappy`.
                         enum:
                         - bzip2
                         - gzip
+                        - lz4
                         - snappy
                         type: string
                       encryption:

go.mod

@@ -7,7 +7,7 @@ toolchain go1.26.2
 require (
 	github.com/cert-manager/cert-manager v1.20.2
 	github.com/cloudnative-pg/api v1.29.0
-	github.com/cloudnative-pg/barman-cloud v0.5.0
+	github.com/cloudnative-pg/barman-cloud v0.5.1
 	github.com/cloudnative-pg/cloudnative-pg v1.29.0
 	github.com/cloudnative-pg/cnpg-i v0.5.0
 	github.com/cloudnative-pg/cnpg-i-machinery v0.4.2

go.sum

@@ -20,8 +20,8 @@ github.com/cespare/xxhash/v2 v2.3.0 h1:UL815xU9SqsFlibzuggzjXhog7bL6oX9BbNZnL2UF
 github.com/cespare/xxhash/v2 v2.3.0/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XLFGgcrjCOs=
 github.com/cloudnative-pg/api v1.29.0 h1:mNx6yJ5qi+Xrjs0NYrUy6V4MlXBkVJxGKwvTJZIuTX4=
 github.com/cloudnative-pg/api v1.29.0/go.mod h1:bF3HI8UVVcllZ7M8CfBufnb+Us8FyQArhD+4qtX0qhM=
-github.com/cloudnative-pg/barman-cloud v0.5.0 h1:DykSaX4o7ee2vyu5FQoG1RJsntHd+EIttIKZbJPlB1Q=
+github.com/cloudnative-pg/barman-cloud v0.5.1 h1:vjkXrrxo2DQXHT9u9usqhtaHiPZ/lTfDVs/pIWYTepQ=
-github.com/cloudnative-pg/barman-cloud v0.5.0/go.mod h1:SO2HzLa+GWlSIpGyxnISoJAFPIcaa/qDa33Bb3jefac=
+github.com/cloudnative-pg/barman-cloud v0.5.1/go.mod h1:XPc5IUFP1y4cZX1sg+Pd8j9V4tmUEVnv3BGCpfQOOg8=
 github.com/cloudnative-pg/cloudnative-pg v1.29.0 h1:49Dm8+y4va7RODspJjeaK8uMWP3OGAD0gMsxhjm16Mo=
 github.com/cloudnative-pg/cloudnative-pg v1.29.0/go.mod h1:0Sgb/50VyaCnQm3IbWqgnhQG8Kb6mgqo8Jo1J+KtkSI=
 github.com/cloudnative-pg/cnpg-i v0.5.0 h1:/TOzpNT6cwNgrpftTtrnLKdoHgMwd+88vZgXjlVgXeE=

internal/cnpgi/operator/lifecycle.go

@@ -145,10 +145,16 @@ func (impl LifecycleImplementation) reconcileJob(
 		return nil, err
 	}
 
+	useAzureWorkloadIdentity, err := impl.collectAzureWorkloadIdentityUsage(ctx, pluginConfiguration)
+	if err != nil {
+		return nil, err
+	}
+
 	return reconcileJob(ctx, cluster, request, sidecarConfiguration{
 		env:                      env,
 		certificates:             certificates,
 		resources:                resources,
+		useAzureWorkloadIdentity: useAzureWorkloadIdentity,
 	})
 }
 
@@ -157,6 +163,7 @@ type sidecarConfiguration struct {
 	certificates             []corev1.VolumeProjection
 	resources                corev1.ResourceRequirements
 	additionalArgs           []string
+	useAzureWorkloadIdentity bool
 }
 
 func reconcileJob(
@@ -243,11 +250,17 @@ func (impl LifecycleImplementation) reconcilePod(
 		return nil, err
 	}
 
+	useAzureWorkloadIdentity, err := impl.collectAzureWorkloadIdentityUsage(ctx, pluginConfiguration)
+	if err != nil {
+		return nil, err
+	}
+
 	return reconcileInstancePod(ctx, cluster, request, pluginConfiguration, sidecarConfiguration{
 		env:                      env,
 		certificates:             certificates,
 		resources:                resources,
 		additionalArgs:           additionalArgs,
+		useAzureWorkloadIdentity: useAzureWorkloadIdentity,
 	})
 }
 
@@ -301,6 +314,25 @@ func (impl LifecycleImplementation) collectAdditionalInstanceArgs(
 	return nil, nil
 }
 
+func (impl LifecycleImplementation) collectAzureWorkloadIdentityUsage(
+	ctx context.Context,
+	pluginConfiguration *config.PluginConfiguration,
+) (bool, error) {
+	for _, objectKey := range pluginConfiguration.GetReferredBarmanObjectsKey() {
+		var objectStore barmancloudv1.ObjectStore
+		if err := impl.Client.Get(ctx, objectKey, &objectStore); err != nil {
+			return false, fmt.Errorf("while getting object store %s: %w", objectKey.String(), err)
+		}
+
+		if objectStore.Spec.Configuration.Azure != nil &&
+			objectStore.Spec.Configuration.Azure.UseDefaultAzureCredentials {
+			return true, nil
+		}
+	}
+
+	return false, nil
+}
+
 func reconcileInstancePod(
 	ctx context.Context,
 	cluster *cnpgv1.Cluster,
@@ -379,6 +411,13 @@ func reconcilePodSpec(
 		},
 	)
 
+	if config.useAzureWorkloadIdentity {
+		envs = append(envs, corev1.EnvVar{
+			Name:  "AZURE_FEDERATED_TOKEN_FILE",
+			Value: azureFederatedTokenFilePath,
+		})
+	}
+
 	envs = append(envs, config.env...)
 
 	baseProbe := &corev1.Probe{
@@ -466,6 +505,34 @@ func reconcilePodSpec(
 		spec.Volumes = removeVolume(spec.Volumes, barmanCertificatesVolumeName)
 	}
 
+	if config.useAzureWorkloadIdentity {
+		sidecarTemplate.VolumeMounts = ensureVolumeMount(
+			sidecarTemplate.VolumeMounts,
+			corev1.VolumeMount{
+				Name:      azureFederatedTokenVolumeName,
+				MountPath: azureFederatedTokenMountPath,
+				ReadOnly:  true,
+			},
+		)
+
+		spec.Volumes = ensureVolume(spec.Volumes, corev1.Volume{
+			Name: azureFederatedTokenVolumeName,
+			VolumeSource: corev1.VolumeSource{
+				Projected: &corev1.ProjectedVolumeSource{
+					Sources: []corev1.VolumeProjection{
+						{
+							ServiceAccountToken: &corev1.ServiceAccountTokenProjection{
+								Path:              azureFederatedTokenFileName,
+								Audience:          azureFederatedTokenAudience,
+								ExpirationSeconds: ptr.To[int64](azureFederatedTokenExpirationSeconds),
+							},
+						},
+					},
+				},
+			},
+		})
+	}
+
 	if err := injectPluginSidecarPodSpec(spec, &sidecarTemplate, mainContainerName); err != nil {
 		return err
 	}
@@ -473,6 +540,15 @@ func reconcilePodSpec(
 	return nil
 }
 
+const (
+	azureFederatedTokenVolumeName        = "azure-identity-token"
+	azureFederatedTokenMountPath         = "/var/run/secrets/azure/tokens"
+	azureFederatedTokenFileName          = "azure-identity-token"
+	azureFederatedTokenFilePath          = azureFederatedTokenMountPath + "/" + azureFederatedTokenFileName
+	azureFederatedTokenAudience          = "api://AzureADTokenExchange"
+	azureFederatedTokenExpirationSeconds = 3600
+)
+
 // TODO: move to machinery once the logic is finalized
 
 // InjectPluginVolumePodSpec injects the plugin volume into a CNPG Pod spec.

internal/cnpgi/operator/lifecycle_test.go

@@ -22,6 +22,7 @@ package operator
 import (
 	"encoding/json"
 
+	barmanapi "github.com/cloudnative-pg/barman-cloud/pkg/api"
 	cnpgv1 "github.com/cloudnative-pg/cloudnative-pg/api/v1"
 	"github.com/cloudnative-pg/cloudnative-pg/pkg/utils"
 	"github.com/cloudnative-pg/cnpg-i/pkg/lifecycle"
@@ -387,6 +388,134 @@ var _ = Describe("LifecycleImplementation", func() {
 			Expect(args).To(Equal([]string{"--log-level=info"}))
 		})
 	})
+
+	Describe("collectAzureWorkloadIdentityUsage", func() {
+		It("returns true when any referred object store uses default Azure credentials", func(ctx SpecContext) {
+			ns := "test-ns"
+			cluster := &cnpgv1.Cluster{ObjectMeta: metav1.ObjectMeta{Name: "c", Namespace: ns}}
+			pc := &config.PluginConfiguration{
+				Cluster:                  cluster,
+				BarmanObjectName:         "primary-store",
+				RecoveryBarmanObjectName: "recovery-store",
+			}
+			primaryStore := &barmancloudv1.ObjectStore{
+				ObjectMeta: metav1.ObjectMeta{Name: pc.BarmanObjectName, Namespace: ns},
+				Spec: barmancloudv1.ObjectStoreSpec{
+					Configuration: barmanapi.BarmanObjectStoreConfiguration{
+						BarmanCredentials: barmanapi.BarmanCredentials{
+							Azure: &barmanapi.AzureCredentials{UseDefaultAzureCredentials: true},
+						},
+					},
+				},
+			}
+			recoveryStore := &barmancloudv1.ObjectStore{
+				ObjectMeta: metav1.ObjectMeta{Name: pc.RecoveryBarmanObjectName, Namespace: ns},
+			}
+			cli := buildClientFunc(primaryStore, recoveryStore).Build()
+
+			impl := LifecycleImplementation{Client: cli}
+			useWorkloadIdentity, err := impl.collectAzureWorkloadIdentityUsage(ctx, pc)
+			Expect(err).NotTo(HaveOccurred())
+			Expect(useWorkloadIdentity).To(BeTrue())
+		})
+
+		It("returns false when none of the referred object stores use default Azure credentials", func(ctx SpecContext) {
+			ns := "test-ns"
+			cluster := &cnpgv1.Cluster{ObjectMeta: metav1.ObjectMeta{Name: "c", Namespace: ns}}
+			pc := &config.PluginConfiguration{
+				Cluster:          cluster,
+				BarmanObjectName: "primary-store",
+			}
+			primaryStore := &barmancloudv1.ObjectStore{
+				ObjectMeta: metav1.ObjectMeta{Name: pc.BarmanObjectName, Namespace: ns},
+				Spec: barmancloudv1.ObjectStoreSpec{
+					Configuration: barmanapi.BarmanObjectStoreConfiguration{
+						BarmanCredentials: barmanapi.BarmanCredentials{
+							Azure: &barmanapi.AzureCredentials{InheritFromAzureAD: true},
+						},
+					},
+				},
+			}
+			cli := buildClientFunc(primaryStore).Build()
+
+			impl := LifecycleImplementation{Client: cli}
+			useWorkloadIdentity, err := impl.collectAzureWorkloadIdentityUsage(ctx, pc)
+			Expect(err).NotTo(HaveOccurred())
+			Expect(useWorkloadIdentity).To(BeFalse())
+		})
+	})
+})
+
+var _ = Describe("reconcilePodSpec", func() {
+	It("injects the Azure federated token volume and env when workload identity is enabled", func() {
+		cluster := &cnpgv1.Cluster{ObjectMeta: metav1.ObjectMeta{Name: "cluster-1", Namespace: "ns-1"}}
+		spec := &corev1.PodSpec{
+			Containers: []corev1.Container{
+				{
+					Name: "postgres",
+					Env: []corev1.EnvVar{
+						{Name: "AZURE_CLIENT_ID", Value: "client-id"},
+					},
+				},
+			},
+		}
+
+		err := reconcilePodSpec(
+			cluster,
+			spec,
+			"postgres",
+			corev1.Container{Args: []string{"instance"}},
+			sidecarConfiguration{useAzureWorkloadIdentity: true},
+		)
+		Expect(err).NotTo(HaveOccurred())
+		Expect(spec.Volumes).To(ContainElement(HaveField("Name", azureFederatedTokenVolumeName)))
+		Expect(spec.InitContainers).To(HaveLen(1))
+		Expect(spec.InitContainers[0].Env).To(ContainElement(corev1.EnvVar{
+			Name:  "AZURE_FEDERATED_TOKEN_FILE",
+			Value: azureFederatedTokenFilePath,
+		}))
+		Expect(spec.InitContainers[0].Env).To(ContainElement(corev1.EnvVar{
+			Name:  "AZURE_CLIENT_ID",
+			Value: "client-id",
+		}))
+		Expect(spec.InitContainers[0].VolumeMounts).To(ContainElement(corev1.VolumeMount{
+			Name:      azureFederatedTokenVolumeName,
+			MountPath: azureFederatedTokenMountPath,
+			ReadOnly:  true,
+		}))
+	})
+
+	It("does not override an existing federated token file env", func() {
+		cluster := &cnpgv1.Cluster{ObjectMeta: metav1.ObjectMeta{Name: "cluster-1", Namespace: "ns-1"}}
+		spec := &corev1.PodSpec{
+			Containers: []corev1.Container{
+				{
+					Name: "postgres",
+					Env: []corev1.EnvVar{
+						{Name: "AZURE_FEDERATED_TOKEN_FILE", Value: "/custom/token"},
+					},
+				},
+			},
+		}
+
+		err := reconcilePodSpec(
+			cluster,
+			spec,
+			"postgres",
+			corev1.Container{Args: []string{"instance"}},
+			sidecarConfiguration{useAzureWorkloadIdentity: true},
+		)
+		Expect(err).NotTo(HaveOccurred())
+		Expect(spec.InitContainers).To(HaveLen(1))
+		Expect(spec.InitContainers[0].Env).To(ContainElement(corev1.EnvVar{
+			Name:  "AZURE_FEDERATED_TOKEN_FILE",
+			Value: "/custom/token",
+		}))
+		Expect(spec.InitContainers[0].Env).NotTo(ContainElement(corev1.EnvVar{
+			Name:  "AZURE_FEDERATED_TOKEN_FILE",
+			Value: azureFederatedTokenFilePath,
+		}))
+	})
 })
 
 var _ = Describe("Volume utilities", func() {

manifest.yaml

@@ -143,10 +143,11 @@ spec:
                         description: |-
                           Compress a backup file (a tar file per tablespace) while streaming it
                           to the object store. Available options are empty string (no
-                          compression, default), `gzip`, `bzip2`, and `snappy`.
+                          compression, default), `gzip`, `bzip2`, `lz4`, and `snappy`.
                         enum:
                         - bzip2
                         - gzip
+                        - lz4
                         - snappy
                         type: string
                       encryption:

web/docs/compression.md

@@ -15,7 +15,7 @@ for space, speed, or a balance of both.
 
 - `bzip2`
 - `gzip`
-- `lz4` (WAL only)
+- `lz4`
 - `snappy`
 - `xz` (WAL only)
 - `zstd` (WAL only)
@@ -41,3 +41,5 @@ network throughput.
 | bzip2       | 25,404           | 13,886            | 395                    | 67                   | 5.9:1 |
 | gzip        | 116,281          | 3,077             | 395                    | 91                   | 4.3:1 |
 | snappy      | 8,134            | 8,341             | 395                    | 166                  | 2.4:1 |
+
+Numbers come from a 2021 Barman proof of concept ([EnterpriseDB/barman#344](https://github.com/EnterpriseDB/barman/issues/344#issuecomment-992547396)), which predates `lz4` support for base backups. `lz4` is not yet benchmarked.

web/docs/object_stores.md

@@ -272,9 +272,10 @@ flow, which uses [`DefaultAzureCredential`](https://learn.microsoft.com/en-us/py
 to automatically discover and use available credentials in the following order:
 
 1. **Environment Variables** — `AZURE_CLIENT_ID`, `AZURE_CLIENT_SECRET`, and `AZURE_TENANT_ID` for Service Principal authentication
-2. **Managed Identity** — Uses the managed identity assigned to the pod
+2. **Workload Identity** — Uses `AZURE_CLIENT_ID`, `AZURE_TENANT_ID`, and a federated service account token
-3. **Azure CLI** — Uses credentials from the Azure CLI if available
+3. **Managed Identity** — Uses the managed identity assigned to the pod
-4. **Azure PowerShell** — Uses credentials from Azure PowerShell if available
+4. **Azure CLI** — Uses credentials from the Azure CLI if available
+5. **Azure PowerShell** — Uses credentials from Azure PowerShell if available
 
 This approach is particularly useful for getting started with development and testing; it allows
 the SDK to attempt multiple authentication mechanisms seamlessly across different environments.
@@ -295,6 +296,30 @@ spec:
   [...]
 ```
 
+When `useDefaultAzureCredentials: true` is set, the plugin sidecar projects a
+service account token with the Azure workload identity audience and exposes it
+as `AZURE_FEDERATED_TOKEN_FILE`. If your platform does not already inject
+`AZURE_CLIENT_ID` and `AZURE_TENANT_ID`, you can provide them through
+`.spec.instanceSidecarConfiguration.env`:
+
+```yaml
+apiVersion: barmancloud.cnpg.io/v1
+kind: ObjectStore
+metadata:
+  name: azure-store
+spec:
+  configuration:
+    destinationPath: "<destination path here>"
+    azureCredentials:
+      useDefaultAzureCredentials: true
+  instanceSidecarConfiguration:
+    env:
+    - name: AZURE_CLIENT_ID
+      value: "<managed-identity-client-id>"
+    - name: AZURE_TENANT_ID
+      value: "<tenant-id>"
+```
+
 ### Access Key, SAS Token, or Connection String
 
 Store credentials in a Kubernetes secret: