Skip to content

Configuring Your Artifact Repository

To run Argo workflows that use artifacts, you must configure and use an artifact repository. Argo supports any S3 compatible artifact repository such as AWS, GCS and MinIO. This section shows how to configure the artifact repository. Subsequent sections will show how to use it.

Name Inputs Outputs Garbage Collection Usage (Feb 2020)
Artifactory Yes Yes No 11%
Azure Blob Yes Yes Yes -
GCS Yes Yes Yes -
Git Yes No No -
HDFS Yes Yes No 3%
HTTP Yes Yes No 2%
OSS Yes Yes No -
Raw Yes No No 5%
S3 Yes Yes Yes 86%

The actual repository used by a workflow is chosen by the following rules:

  1. Anything explicitly configured using Artifact Repository Ref. This is the most flexible, safe, and secure option.
  2. From a config map named artifact-repositories if it has the workflows.argoproj.io/default-artifact-repository annotation in the workflow's namespace.
  3. From a workflow controller config-map.

Configuring MinIO

You can install MinIO into your cluster via Helm.

First, install helm. Then, install MinIO with the below commands:

helm repo add minio https://charts.min.io/ # official minio Helm charts
helm repo update
helm install argo-artifacts minio/minio --set service.type=LoadBalancer --set fullnameOverride=argo-artifacts

Login to the MinIO UI using a web browser (port 9000) after obtaining the external IP using kubectl.

kubectl get service argo-artifacts

On Minikube:

minikube service --url argo-artifacts

NOTE: When MinIO is installed via Helm, it generates credentials, which you will use to login to the UI: Use the commands shown below to see the credentials

  • AccessKey: kubectl get secret argo-artifacts -o jsonpath='{.data.accesskey}' | base64 --decode
  • SecretKey: kubectl get secret argo-artifacts -o jsonpath='{.data.secretkey}' | base64 --decode

Create a bucket named my-bucket from the MinIO UI.

If MinIO is configured to use TLS you need to set the parameter insecure to false. Additionally, if MinIO is protected by certificates generated by a custom CA, you first need to save the CA certificate in a Kubernetes secret, then set the caSecret parameter accordingly. This will allow Argo to correctly verify the server certificate presented by MinIO. For example:

kubectl create secret generic my-root-ca --from-file=my-ca.pem
artifacts:
  - s3:
      insecure: false
      caSecret:
        name: my-root-ca
        key: my-ca.pem
      ...

Configuring AWS S3

First, create a bucket:

mybucket=my-bucket-name
aws s3 mb s3://mybucket [--region xxx]

AWS S3 IAM Access

Next, create a policy file. You will attach this in one of the sections below based on your chosen authentication method.

cat > policy.json <<EOF
{
   "Version":"2012-10-17",
   "Statement":[
      {
         "Effect":"Allow",
         "Action":[
            "s3:PutObject",
            "s3:GetObject",
            "s3:DeleteObject"
         ],
         "Resource":"arn:aws:s3:::$mybucket/*"
      },
      {
         "Effect":"Allow",
         "Action":[
            "s3:ListBucket"
         ],
         "Resource":"arn:aws:s3:::$mybucket"
      }
   ]
}
EOF

If you do not have Artifact Garbage Collection configured, you should remove s3:DeleteObject from the list of Actions above.

Region discovery

Argo can discover the region of your buckets with the additional policy below. Without this, you must specify the region in your artifact configuration.

      {
        "Effect":"Allow",
        "Action":[
            "s3:GetBucketLocation"
        ],
        "Resource":"arn:aws:s3:::*"
      }
    ...

AWS S3 IRSA

IAM Roles for Service Accounts (IRSA) is the recommended Kubernetes native mechanism to authenticate to S3. If you are using EKS, follow the IRSA setup guide. If not, follow the Pod Identity Webhook self-hosted setup guide.

With the bucket and policy as described above, create an IAM role and add the policy:

aws iam create-role --role-name $mybucket-role
aws iam put-role-policy --role-name $mybucket-user --policy-name $mybucket-policy --policy-document file://policy.json

Attach this IAM role to a service account with an annotation:

apiVersion: v1
kind: ServiceAccount
metadata:
  annotations:
    eks.amazonaws.com/role-arn: arn:aws:iam::012345678901:role/mybucket-role
  name: myserviceaccount
  namespace: mynamespace

Use the service account in a workflow:

apiVersion: argoproj.io/v1alpha1
kind: Workflow
spec:
  serviceAccountName: myserviceaccount

AWS S3 with IAM Access Keys

Least privilege user

To reduce the privileges of an access key, create a user with only the necessary permissions and no more.

With the bucket and policy described above, create an IAM user and add the policy:

aws iam create-user --user-name $mybucket-user
aws iam put-user-policy --user-name $mybucket-user --policy-name $mybucket-policy --policy-document file://policy.json
aws iam create-access-key --user-name $mybucket-user > access-key.json

Configure an artifact with the access keys:

artifacts:
  - name: my-output-artifact
    path: /my-output-artifact
    s3:
      endpoint: s3.amazonaws.com
      bucket: my-s3-bucket
      key: path/in/bucket/my-output-artifact.tgz
      # The following fields are secret selectors.
      # They reference the k8s secret named 'my-s3-credentials'.
      # This secret is expected to have the keys 'accessKey' and 'secretKey',
      # containing the base64 encoded credentials to the bucket.
      accessKeySecret:
        name: my-s3-credentials
        key: accessKey
      secretKeySecret:
        name: my-s3-credentials
        key: secretKey

AWS S3 with IAM Assume Role

v3.6 and after

You can use an IAM role for temporary access.

With the bucket and policy described above, create an IAM role and add the policy:

aws iam create-role --role-name $mybucket-role
aws iam put-role-policy --role-name $mybucket-user --policy-name $mybucket-policy --policy-document file://policy.json

Retrieve the role credentials:

aws sts assume-role --role-arn arn:aws:iam::012345678901:role/$mybucket-role

Configure an artifact with the credentials:

artifacts:
  - name: my-output-artifact
    path: /my-output-artifact
    s3:
      endpoint: s3.amazonaws.com
      bucket: my-s3-bucket
      key: path/in/bucket/my-output-artifact.tgz
      # The following fields are secret selectors.
      # They reference the k8s secret named 'my-s3-credentials'.
      # This secret is expected to have the keys 'accessKey', 'secretKey', and 'sessionToken',
      # containing the base64 encoded credentials to the bucket.
      accessKeySecret:
        name: my-s3-credentials
        key: accessKey
      secretKeySecret:
        name: my-s3-credentials
        key: secretKey
      sessionTokenSecret:
        name: my-s3-credentials
        key: sessionToken

Temporary

IAM role credentials are temporary, so you must refresh them periodically via an external mechanism.

AWS S3 with S3 Access Grants

v3.6 and after

You can use S3 Access Grants for temporary, reduced scope access. Follow the AWS guide to set this up in your AWS account and retrieve access grant credentials.

Configure an artifact with the access grant credentials:

artifacts:
  - name: my-output-artifact
    path: /my-output-artifact
    s3:
      endpoint: s3.amazonaws.com
      bucket: my-s3-bucket
      key: path/in/bucket/my-output-artifact.tgz
      # The following fields are secret selectors.
      # They reference the k8s secret named 'my-s3-credentials'.
      # This secret is expected to have the keys 'accessKey', 'secretKey', and 'sessionToken',
      # containing the base64 encoded credentials to the bucket.
      accessKeySecret:
        name: my-s3-credentials
        key: accessKey
      secretKeySecret:
        name: my-s3-credentials
        key: secretKey
      sessionTokenSecret:
        name: my-s3-credentials
        key: sessionToken

Temporary

S3 Access Grants are temporary, so you must refresh them periodically via an external mechanism.

Configuring GCS (Google Cloud Storage)

Create a bucket from the GCP Console (https://console.cloud.google.com/storage/browser).

There are 2 ways to configure a Google Cloud Storage.

Through Native GCS APIs

  • Create and download a Google Cloud service account key.
  • Create a kubernetes secret to store the key.
  • Configure gcs artifact as following in the yaml.
artifacts:
  - name: message
    path: /tmp/message
    gcs:
      bucket: my-bucket-name
      key: path/in/bucket
      # serviceAccountKeySecret is a secret selector.
      # It references the k8s secret named 'my-gcs-credentials'.
      # This secret is expected to have the key 'serviceAccountKey',
      # containing the base64 encoded credentials
      # to the bucket.
      #
      # If it's running on GKE and Workload Identity is used,
      # serviceAccountKeySecret is not needed.
      serviceAccountKeySecret:
        name: my-gcs-credentials
        key: serviceAccountKey

If it's a GKE cluster, and Workload Identity is configured, there's no need to create the service account key and store it as a Kubernetes secret, serviceAccountKeySecret is also not needed in this case. Please follow the link to configure Workload Identity (https://cloud.google.com/kubernetes-engine/docs/how-to/workload-identity).

Use S3 APIs

Enable S3 compatible access and create an access key. Note that S3 compatible access is on a per project rather than per bucket basis.

artifacts:
  - name: my-output-artifact
    path: /my-output-artifact
    s3:
      endpoint: storage.googleapis.com
      bucket: my-gcs-bucket-name
      # NOTE that, by default, all output artifacts are automatically tarred and
      # gzipped before saving. So as a best practice, .tgz or .tar.gz
      # should be incorporated into the key name so the resulting file
      # has an accurate file extension.
      key: path/in/bucket/my-output-artifact.tgz
      accessKeySecret:
        name: my-gcs-s3-credentials
        key: accessKey
      secretKeySecret:
        name: my-gcs-s3-credentials
        key: secretKey

Configuring Alibaba Cloud OSS (Object Storage Service)

Create your bucket and access key for the bucket. Suggest to limit the permission for the access key, you will need to create a user with just the permissions you want to associate with the access key. Otherwise, you can just create an access key using your existing user account.

Setup Alibaba Cloud CLI and follow the steps to configure the artifact storage for your workflow:

$ export mybucket=bucket-workflow-artifect
$ export myregion=cn-zhangjiakou
$ # limit permission to read/write the bucket.
$ cat > policy.json <<EOF
{
    "Version": "1",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
              "oss:PutObject",
              "oss:GetObject"
            ],
            "Resource": "acs:oss:*:*:$mybucket/*"
        }
    ]
}
EOF
$ # create bucket.
$ aliyun oss mb oss://$mybucket --region $myregion
$ # show endpoint of bucket.
$ aliyun oss stat oss://$mybucket
$ #create a ram user to access bucket.
$ aliyun ram CreateUser --UserName $mybucket-user
$ # create ram policy with the limit permission.
$ aliyun ram CreatePolicy --PolicyName $mybucket-policy --PolicyDocument "$(cat policy.json)"
$ # attch ram policy to the ram user.
$ aliyun ram AttachPolicyToUser --UserName $mybucket-user --PolicyName $mybucket-policy --PolicyType Custom
$ # create access key and secret key for the ram user.
$ aliyun ram CreateAccessKey --UserName $mybucket-user > access-key.json
$ # create secret in demo namespace, replace demo with your namespace.
$ kubectl create secret generic $mybucket-credentials -n demo\
  --from-literal "accessKey=$(cat access-key.json | jq -r .AccessKey.AccessKeyId)" \
  --from-literal "secretKey=$(cat access-key.json | jq -r .AccessKey.AccessKeySecret)"
$ # create configmap to config default artifact for a namespace.
$ cat > default-artifact-repository.yaml << EOF
apiVersion: v1
kind: ConfigMap
metadata:
  # If you want to use this config map by default, name it "artifact-repositories". Otherwise, you can provide a reference to a
  # different config map in `artifactRepositoryRef.configMap`.
  name: artifact-repositories
  annotations:
    # v3.0 and after - if you want to use a specific key, put that key into this annotation.
    workflows.argoproj.io/default-artifact-repository: default-oss-artifact-repository
data:
  default-oss-artifact-repository: |
    oss:
      endpoint: http://oss-cn-zhangjiakou-internal.aliyuncs.com
      bucket: $mybucket
      # accessKeySecret and secretKeySecret are secret selectors.
      # It references the k8s secret named 'bucket-workflow-artifect-credentials'.
      # This secret is expected to have the keys 'accessKey'
      # and 'secretKey', containing the base64 encoded credentials
      # to the bucket.
      accessKeySecret:
        name: $mybucket-credentials
        key: accessKey
      secretKeySecret:
        name: $mybucket-credentials
        key: secretKey
EOF
# create cm in demo namespace, replace demo with your namespace.
$ k apply -f default-artifact-repository.yaml -n demo

You can also set createBucketIfNotPresent to true to tell the artifact driver to automatically create the OSS bucket if it doesn't exist yet when saving artifacts. Note that you'll need to set additional permission for your OSS account to create new buckets.

Alibaba Cloud OSS RRSA

If you wish to use OSS RRSA instead of passing in an accessKey and secretKey, you need to perform the following actions:

  • Install pod-identity-webhook in your cluster to automatically inject the OIDC tokens and environment variables.
  • Add the label pod-identity.alibabacloud.com/injection: 'on' to the target workflow namespace.
  • Add the annotation pod-identity.alibabacloud.com/role-name: $your_ram_role_name to the service account of running workflow.
  • Set useSDKCreds: true in your target artifact repository cm and remove the secret references to AK/SK.
apiVersion: v1
kind: Namespace
metadata:
  name: my-ns
  labels:
    pod-identity.alibabacloud.com/injection: 'on'

---
apiVersion: v1
kind: ServiceAccount
metadata:
  name: my-sa
  namespace: rrsa-demo
  annotations:
    pod-identity.alibabacloud.com/role-name: $your_ram_role_name

---
apiVersion: v1
kind: ConfigMap
metadata:
  # If you want to use this config map by default, name it "artifact-repositories". Otherwise, you can provide a reference to a
  # different config map in `artifactRepositoryRef.configMap`.
  name: artifact-repositories
  annotations:
    # v3.0 and after - if you want to use a specific key, put that key into this annotation.
    workflows.argoproj.io/default-artifact-repository: default-oss-artifact-repository
data:
  default-oss-artifact-repository: |
    oss:
      endpoint: http://oss-cn-zhangjiakou-internal.aliyuncs.com
      bucket: $mybucket
      useSDKCreds: true

Configuring Azure Blob Storage

Create an Azure Storage account and a container within your account. You can use the Azure Portal, the CLI, or other tools.

You can authenticate Argo to your Azure storage account in multiple ways:

Using Azure Managed Identities

Azure Managed Identities is the preferred method for managing access to Azure resources securely. You can set useSDKCreds: true if a Managed Identity is assigned. In this case, the accountKeySecret is not used and authentication uses DefaultAzureCredential.

artifacts:
  - name: message
    path: /tmp/message
    azure:
      endpoint: https://mystorageaccountname.blob.core.windows.net
      container: my-container-name
      blob: path/in/container
      # If a managed identity has been assigned to the machines running the
      # workflow (for example, https://docs.microsoft.com/en-us/azure/aks/use-managed-identity)
      # then useSDKCreds should be set to true. The accountKeySecret is not required
      # and will not be used in this case.
      useSDKCreds: true  

Using Azure Access Keys

You can also use an Access Key.

  1. Retrieve the blob service endpoint for the storage account:

    az storage account show -n mystorageaccountname --query 'primaryEndpoints.blob' -otsv
    # https://mystorageaccountname.blob.core.windows.net
    
  2. Retrieve the Access Key for the storage account:

    ACCESS_KEY="$(az storage account keys list -n mystorageaccountname --query '[0].value' -otsv)"
    
  3. Create a Kubernetes Secret to hold the storage account key:

    kubectl create secret generic my-azure-storage-credentials \
      --from-literal "account-access-key=$ACCESS_KEY"
    
  4. Configure an azure artifact:

    artifacts:
      - name: message
        path: /tmp/message
        azure:
          endpoint: https://mystorageaccountname.blob.core.windows.net
          container: my-container-name
          blob: path/in/container
          # accountKeySecret is a secret selector.
          # It references the Kubernetes Secret named 'my-azure-storage-credentials'.
          # This secret is expected to have the key 'account-access-key',
          # containing the base64 encoded credentials to the storage account.
          accountKeySecret:
            name: my-azure-storage-credentials
            key: account-access-key     
    

Using Azure Shared Access Signatures (SAS)

v3.6 and after

You can also use a Shared Access Signature (SAS).

  1. Retrieve the blob service endpoint for the storage account:

    az storage account show -n mystorageaccountname --query 'primaryEndpoints.blob' -otsv
    # https://mystorageaccountname.blob.core.windows.net
    
  2. Generate a Shared Access Signature for the storage account:

    SAS_TOKEN="$(az storage container generate-sas --account-name <storage-account> --name <container> --permissions acdlrw --expiry <date-time> --auth-mode key)"
    
  3. Create a Kubernetes Secret to hold the storage account key:

    kubectl create secret generic my-azure-storage-credentials \
      --from-literal "shared-access-key=$SAS_TOKEN"
    
  4. Configure an azure artifact:

    artifacts:
      - name: message
        path: /tmp/message
        azure:
          endpoint: https://mystorageaccountname.blob.core.windows.net
          container: my-container-name
          blob: path/in/container
          # accountKeySecret is a secret selector.
          # It references the Kubernetes Secret named 'my-azure-storage-credentials'.
          # This secret is expected to have the key 'shared-access-key',
          # containing the base64 encoded shared access signature to the storage account.
          accountKeySecret:
            name: my-azure-storage-credentials
            key: shared-access-key
    

Configure the Default Artifact Repository

In order for Argo to use your artifact repository, you can configure it as the default repository. Edit the workflow-controller config map with the correct endpoint and access/secret keys for your repository.

S3 compatible artifact repository bucket (such as AWS, GCS, MinIO, and Alibaba Cloud OSS)

Use the endpoint corresponding to your provider:

  • AWS: s3.amazonaws.com
  • GCS: storage.googleapis.com
  • MinIO: my-minio-endpoint.default:9000
  • Alibaba Cloud OSS: oss-cn-hangzhou-zmf.aliyuncs.com

The key is name of the object in the bucket The accessKeySecret and secretKeySecret are secret selectors that reference the specified kubernetes secret. The secret is expected to have the keys accessKey and secretKey, containing the base64 encoded credentials to the bucket.

For AWS, the accessKeySecret and secretKeySecret correspond to AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY respectively.

EC2 provides a meta-data API via which applications using the AWS SDK may assume IAM roles associated with the instance. If you are running argo on EC2 and the instance role allows access to your S3 bucket, you can configure the workflow step pods to assume the role. To do so, simply omit the accessKeySecret and secretKeySecret fields.

For GCS, the accessKeySecret and secretKeySecret for S3 compatible access can be obtained from the GCP Console. Note that S3 compatible access is on a per project rather than per bucket basis.

For MinIO, the accessKeySecret and secretKeySecret naturally correspond the AccessKey and SecretKey.

For Alibaba Cloud OSS, the accessKeySecret and secretKeySecret corresponds to accessKeyID and accessKeySecret respectively.

Example:

$ kubectl edit configmap workflow-controller-configmap -n argo  # assumes argo was installed in the argo namespace
...
data:
  artifactRepository: |
    s3:
      bucket: my-bucket
      keyFormat: prefix/in/bucket     #optional
      endpoint: my-minio-endpoint.default:9000        #AWS => s3.amazonaws.com; GCS => storage.googleapis.com
      insecure: true                  #omit for S3/GCS. Needed when minio runs without TLS
      accessKeySecret:                #omit if accessing via AWS IAM
        name: my-minio-cred
        key: accessKey
      secretKeySecret:                #omit if accessing via AWS IAM
        name: my-minio-cred
        key: secretKey
      useSDKCreds: true               #tells argo to use AWS SDK's default provider chain, enable for things like IRSA support

The secrets are retrieved from the namespace you use to run your workflows. Note that you can specify a keyFormat.

Google Cloud Storage (GCS)

Argo also can use native GCS APIs to access a Google Cloud Storage bucket.

serviceAccountKeySecret references to a Kubernetes secret which stores a Google Cloud service account key to access the bucket.

Example:

$ kubectl edit configmap workflow-controller-configmap -n argo  # assumes argo was installed in the argo namespace
...
data:
  artifactRepository: |
    gcs:
      bucket: my-bucket
      keyFormat: prefix/in/bucket/{{workflow.name}}/{{pod.name}}     #it should reference workflow variables, such as "{{workflow.name}}/{{pod.name}}"
      serviceAccountKeySecret:
        name: my-gcs-credentials
        key: serviceAccountKey

Azure Blob Storage

Argo can use native Azure APIs to access a Azure Blob Storage container.

accountKeySecret references to a Kubernetes secret which stores an Azure Blob Storage account shared key to access the container.

Example:

$ kubectl edit configmap workflow-controller-configmap -n argo  # assumes argo was installed in the argo namespace
...
data:
  artifactRepository: |
    azure:
      container: my-container
      blobNameFormat: prefix/in/container     #optional, it could reference workflow variables, such as "{{workflow.name}}/{{pod.name}}"
      accountKeySecret:
        name: my-azure-storage-credentials
        key: account-access-key

Accessing Non-Default Artifact Repositories

This section shows how to access artifacts from non-default artifact repositories.

The endpoint, accessKeySecret and secretKeySecret are the same as for configuring the default artifact repository described previously.

  templates:
  - name: artifact-example
    inputs:
      artifacts:
      - name: my-input-artifact
        path: /my-input-artifact
        s3:
          endpoint: s3.amazonaws.com
          bucket: my-aws-bucket-name
          key: path/in/bucket/my-input-artifact.tgz
          accessKeySecret:
            name: my-aws-s3-credentials
            key: accessKey
          secretKeySecret:
            name: my-aws-s3-credentials
            key: secretKey
    outputs:
      artifacts:
      - name: my-output-artifact
        path: /my-output-artifact
        s3:
          endpoint: storage.googleapis.com
          bucket: my-gcs-bucket-name
          # NOTE that, by default, all output artifacts are automatically tarred and
          # gzipped before saving. So as a best practice, .tgz or .tar.gz
          # should be incorporated into the key name so the resulting file
          # has an accurate file extension.
          key: path/in/bucket/my-output-artifact.tgz
          accessKeySecret:
            name: my-gcs-s3-credentials
            key: accessKey
          secretKeySecret:
            name: my-gcs-s3-credentials
            key: secretKey
          region: my-GCS-storage-bucket-region
    container:
      image: debian:latest
      command: [sh, -c]
      args: ["cp -r /my-input-artifact /my-output-artifact"]

Artifact Streaming

With artifact streaming, artifacts don’t need to be saved to disk first. Artifact streaming is only supported in the following artifact drivers: S3 (v3.4+), Azure Blob (v3.4+), HTTP (v3.5+), Artifactory (v3.5+), and OSS (v3.6+).

Previously, when a user would click the button to download an artifact in the UI, the artifact would need to be written to the Argo Server’s disk first before downloading. If many users tried to download simultaneously, they would take up disk space and fail the download.


Have a question?

Search on GitHub Discussions and Slack.