Configuring Object Storage

A variety of Deis Workflow components rely on an object storage system to do their work including storing application slugs, Docker images and database logs.

Deis Workflow ships with Minio by default, which provides in-cluster, ephemeral object storage. This means that if the Minio server crashes, all data will be lost. Therefore, Minio should be used for development or testing only.

Configuring off-cluster Object Storage

Every component that relies on object storage uses two inputs for configuration:

  1. Component-specific environment variables (e.g. BUILDER_STORAGE and REGISTRY_STORAGE)
  2. Access credentials stored as a Kubernetes secret named objectstorage-keyfile

The helm chart for Deis Workflow can be easily configured to connect Workflow components to off-cluster object storage. Deis Workflow currently supports Google Compute Storage, Amazon S3, Azure Blob Storage and OpenStack Swift Storage.

Step 1: Create storage buckets

Create storage buckets for each of the Workflow subsystems: builder, registry, and database.

Depending on your chosen object storage you may need to provide globally unique bucket names. If you are using S3, use hyphens instead of periods in the bucket names. Using periods in the bucket name will cause an ssl certificate validation issue with S3.

If you provide credentials with sufficient access to the underlying storage, Workflow components will create the buckets if they do not exist.

Step 2: Generate storage credentials

If applicable, generate credentials that have create and write access to the storage buckets created in Step 1.

If you are using AWS S3 and your Kubernetes nodes are configured with appropriate IAM API keys via InstanceRoles, you do not need to create API credentials. Do, however, validate that the InstanceRole has appropriate permissions to the configured buckets!

Step 3: Add Deis Repo

Helm changed its UX adding support for OCI, so helm repo add is not a thing anymore.

The Team Hephy Chart Repository contains everything needed to install Hephy Workflow onto a Kubernetes cluster, but Helm's legacy chart repository data structures are chronically unscalable.

As a result, there's no need to ever run this command again:

$ helm repo add hephy https://charts.teamhephy.com/

(Note: at the time of this writing, the chartmuseum host is currently down for the count.)

Step 4: Configure Workflow Chart

Operators should configure object storage by editing the Helm values file before running helm install. To do so:

  • Fetch the Helm values by running helm inspect values deis/workflow > values.yaml
  • Update the global/storage parameter to reference the platform you are using, e.g. s3, azure, gcs, or swift
  • Find the corresponding section for your storage type and provide appropriate values including region, bucket names, and access credentials.
  • Save your changes.

Note

All values will be automatically (base64) encoded except the key_json values under gcs/gcr. These must be base64-encoded. This is to support cleanly passing said encoded text via helm --set cli functionality rather than attempting to pass the raw JSON data. For example:

$ helm install workflow --namespace deis \
    --set global.storage=gcs,gcs.key_json="$(cat /path/to/gcs_creds.json | base64 -w 0)"

You are now ready to run helm install deis/workflow --namespace deis -f values.yaml using your desired object storage.