Storage Backends¶

OpenDepot supports four storage backends. Each is configured via the storageConfig field on Depot.spec.global.storageConfig, ModuleConfig.storageConfig, or directly on a Module.spec.moduleConfig.storageConfig.

Amazon S3¶

Recommended for production. Stores module archives in S3 buckets with SHA256 checksum validation.

CRD Fields:

Authentication: Uses the AWS SDK v2 default credentials chain. In Kubernetes, this typically means:

• EKS with IRSA (recommended): Annotate the Version controller's ServiceAccount with an IAM role ARN
• Environment variables: Set AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, and optionally AWS_SESSION_TOKEN
• EC2 instance profile: Automatically used when running on EC2/EKS nodes

Required IAM Permissions:

{
  "Effect": "Allow",
  "Action": [
    "s3:GetObject",
    "s3:PutObject",
    "s3:DeleteObject",
    "s3:GetObjectAttributes"
  ],
  "Resource": "arn:aws:s3:::your-bucket-name/*"
}

Example Configuration:

storageConfig:
  s3:
    bucket: opendepot-modules
    region: us-west-2

Azure Blob Storage¶

Recommended for production. Stores module archives in Azure Blob Storage containers with checksum metadata.

CRD Fields:

Authentication: Uses Azure DefaultAzureCredential. In Kubernetes, this typically means:

• AKS with Workload Identity (recommended): Configure federated identity credentials on the Version controller's ServiceAccount
• Managed Identity: Assign a managed identity to the AKS node pool or pod
• Environment variables: Set AZURE_CLIENT_ID, AZURE_TENANT_ID, and AZURE_CLIENT_SECRET

Required Azure RBAC Roles:

• Storage Blob Data Contributor on the Storage Account (for read, write, and delete)
• Reader on the Storage Account resource (for container metadata operations)
• Storage Blob Delegator on the Storage Account (required when pre-signed URLs are enabled — used to obtain a User Delegation Key for SAS generation)

Example Configuration:

storageConfig:
  azureStorage:
    accountName: opendepotmodules
    accountUrl: https://opendepotmodules.blob.core.windows.net
    subscriptionID: 00000000-0000-0000-0000-000000000000
    resourceGroup: opendepot-rg

Google Cloud Storage¶

CRD Fields:

Authentication: Uses Application Default Credentials (ADC). In Kubernetes, this typically means:

• GKE with Workload Identity (recommended): Bind a Google service account to the Version controller's Kubernetes ServiceAccount
• Service account key: Mount a JSON key file and set GOOGLE_APPLICATION_CREDENTIALS

Required GCS Permissions:

• storage.objects.create
• storage.objects.get
• storage.objects.delete
• storage.objects.getMetadata (or the Storage Object Admin role)

Example Configuration:

storageConfig:
  gcs:
    bucket: opendepot-modules

Local Filesystem¶

Stores module archives on a shared volume mounted to both the Version controller and the Server pods. Suitable for development, testing, and air-gapped environments when paired with a PersistentVolumeClaim.

CRD Fields:

How it works: The Helm chart creates a shared volume (either a PersistentVolumeClaim or a hostPath) and mounts it to both the Version controller and the Server. The Version controller writes module archives to the volume, and the Server reads and serves them.

Helm Storage Configuration:

Local Development with kind (hostPath):

helm install opendepot opendepot/opendepot \
  -n opendepot-system \
  --create-namespace \
  --set storage.filesystem.enabled=true \
  --set storage.filesystem.hostPath=/tmp/opendepot-modules

When using hostPath, the chart adds an initContainer that runs as root to set ownership of the volume to uid 65532 (the non-root user the containers run as).

Production with PVC (ReadWriteMany):

helm install opendepot opendepot/opendepot \
  -n opendepot-system \
  --create-namespace \
  --set storage.filesystem.enabled=true \
  --set storage.filesystem.storageClassName=efs-sc \
  --set storage.filesystem.size=50Gi

The PVC requires a StorageClass that supports ReadWriteMany (e.g., AWS EFS, Azure Files, NFS).

Example CRD Configuration:

storageConfig:
  fileSystem:
    directoryPath: /data/modules

Pre-signed URL Redirects¶

When enabled, the server issues an HTTP 307 Temporary Redirect to a time-limited URL signed by the storage backend instead of proxying the binary itself. The OpenTofu client fetches the artifact directly from the storage backend, reducing network egress through the server.

Pre-signed URLs are supported on Amazon S3, Google Cloud Storage, and Azure Blob Storage. The filesystem backend does not support them.

Configure pre-signing via the presign field on any StorageConfig:

storageConfig:
  s3:
    bucket: opendepot-providers
    region: us-east-1
  presign:
    enabled: true
    ttl: "15m"
    fallbackToProxy: true

PresignConfig fields:

Additional IAM permissions required for S3 pre-signing:

No additional IAM permissions are needed. s3:GetObject (already required for the proxy path) is sufficient to generate pre-signed GET URLs.

Additional permissions required for Azure pre-signing:

Azure pre-signed URLs use user delegation SAS tokens. Assign the Storage Blob Delegator role to the controller's identity in addition to the existing Storage Blob Data Contributor role.

GCS signed URLs:

GCS pre-signed URLs use the HMAC/V4 signing method. The workload identity must have the iam.serviceAccounts.signBlob permission (included in the Service Account Token Creator role) in addition to the existing storage permissions.

Storage Backend Comparison¶