#
S3 Object Storage
Phoeniqs Cloud provides S3-compatible object storage for buckets used by your applications, backups, media, and data lakes. Object storage is backed by Ceph RGW (Red Hat OpenShift Data Foundation / NooBaa) and is served from Phoeniqs' Swiss sovereign infrastructure.
S3 object storage is not available on every cluster
S3 object storage is provided by the RGW provider, part of the OpenShift Data Foundation (ODF) storage platform. This provider is only deployed on Phoeniqs' standard Kvant clusters, such as the Basel cluster used throughout this guide.
The AI-2 and AI-3 clusters do not run the RGW / ODF storage provider, so S3 object storage (buckets and ObjectBucketClaims) is not natively available on them. If you are purchasing S3, provision your buckets on a cluster that includes the RGW provider (Basel).
Existing customers already running workloads on AI-2 or AI-3 can still consume S3 by provisioning the bucket on Basel and using it cross-cluster over the public RGW endpoint. See
Object storage is different from the block storage in your Capacity Pool. Block storage is purchased and consumed through a Capacity Pool and PersistentVolumeClaims inside a Namespace. S3 object storage is allocated in the portal, but provisioned on OpenShift through an ObjectBucketClaim. Using it therefore takes two steps:
- In the portal, add the S3 storage subscription to the resource group you want it billed to.
- On OpenShift, create an
ObjectBucketClaimYAML in your project to provision the bucket and obtain access credentials.
Setting up S3 is not complete by adding it to a resource group alone — the bucket is only created on OpenShift.
#
Step 1: Purchase & allocate in the portal
- Sign in to the Phoeniqs Portal.
- Go to Products → Infrastructure Services.
- Click Add on the S3 Object Storage tile.
- At checkout, select the Resource Group you want the storage allocated and billed to, agree to the terms, and click Buy subscription.
This mirrors the checkout flow for the other IaaS products. See the IaaS Quickstart Guide for the general purchase flow and Resource Pricing for rates.
#
Step 2: Provision the bucket on OpenShift
The bucket must be provisioned on OpenShift with an ObjectBucketClaim. Do not use the OpenShift console's GUI wizard to create the bucket — it does not expose the required maxSize field, so you must apply the manifest directly.
Log in to the OpenShift console. See Access Your OpenShift for the login and project navigation steps.
Open the project (namespace) you want to hold the bucket.
Create a persistent claim using the manifest below, adjusting
metadata.nameandmetadata.namespaceto match your project:apiVersion: objectbucket.io/v1alpha1 kind: ObjectBucketClaim metadata: name: obc-smoke-test namespace: d01000-e26-4229-test-15657 spec: generateBucketName: obc-smoke-test storageClassName: ocs-storagecluster-ceph-rgw additionalConfig: maxSize: 10Gi
#
Field notes
Apply the manifest with oc apply -f <file>.yaml or paste it into the OpenShift console's + Add → Import YAML.
#
Step 3: Connect & use
Once the ObjectBucketClaim is provisioned, OpenShift creates an S3 access Secret in the same namespace containing your access key and secret key. These credentials are required to use the bucket.
Retrieve the Secret in your project (OpenShift console Workloads → Secrets, or
oc get secret).Point any S3-compatible client (AWS CLI,
mc, or an application library) at the public Ceph RGW S3 endpoint and authenticate with those keys:https://ocs-storagecluster-cephobjectstore-secure-openshift-storage.apps.basel.kvant.cloud/Create and use your buckets through the client.
note Access requires valid S3 credentials
The endpoint is publicly reachable, but it does not grant access on its own. You must present the access key and secret from the Secret provisioned by your ObjectBucketClaim.
#
Existing customers on AI-2 / AI-3
The AI-2 and AI-3 clusters do not run the RGW provider, so you cannot provision an ObjectBucketClaim directly on them. If your workloads already run on AI-2 or AI-3 and you need S3, you can still use object storage by provisioning the bucket on Basel and consuming it cross-cluster over the public RGW route. This pattern is verified to work from AI-3 against Basel.
The single most important point: the bucket's credentials and settings are created on Basel, but you must point your S3 client at the public RGW endpoint — not at the internal BUCKET_HOST that OpenShift provisions alongside the bucket.
#
Step 1: Provision the bucket on Basel
Create the ObjectBucketClaim on the Basel cluster, exactly as described in spec.storageClassName set to ocs-storagecluster-ceph-rgw and a maxSize.
Once the claim becomes Bound, OpenShift creates a Secret and a ConfigMap named after the claim in the same namespace on Basel:
# On Basel
oc -n <basel-namespace> get secret <obc-name> -o jsonpath='{.data}' | jq 'keys'
# AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY
oc -n <basel-namespace> get cm <obc-name> -o jsonpath='{.data}'
# BUCKET_NAME, BUCKET_HOST, BUCKET_PORT, BUCKET_REGION, BUCKET_SUBREGION
#
Step 2: Copy the credentials to your AI-2 / AI-3 namespace
Copy the access key and secret from Basel into the namespace on your consumer cluster (AI-2 or AI-3). Pipe the values directly between the two clusters with oc, so the credentials are never printed or pasted. You only need the two keys, the bucket name, and the public RGW endpoint:
# From the Basel cluster to the AI-2 / AI-3 consumer cluster
oc --context <basel-ctx> -n <basel-namespace> get secret <obc-name> -o json \
| jq --arg ns <consumer-namespace> '{apiVersion: "v1", kind: "Secret", type: "Opaque",
metadata: {name: "s3-creds", namespace: $ns},
data: {AWS_ACCESS_KEY_ID: .data.AWS_ACCESS_KEY_ID,
AWS_SECRET_ACCESS_KEY: .data.AWS_SECRET_ACCESS_KEY}}' \
| oc --context <consumer-ctx> -n <consumer-namespace> apply -f -
oc --context <consumer-ctx> -n <consumer-namespace> create configmap s3-bucket \
--from-literal=BUCKET_NAME="$(oc --context <basel-ctx> -n <basel-namespace> \
get cm <obc-name> -o jsonpath='{.data.BUCKET_NAME}')" \
--from-literal=AWS_ENDPOINT_URL="https://ocs-storagecluster-cephobjectstore-secure-openshift-storage.apps.basel.kvant.cloud"
Use the public endpoint, not the internal BUCKET_HOST
The ConfigMap created alongside the bucket on Basel contains a BUCKET_HOST that is the cluster-internal service name — it is valid only on Basel. A consumer on AI-2 / AI-3 must override the endpoint with the public RGW route shown above.
#
Step 3: Connect & use from your consumer cluster
Point any S3-compatible client at the public endpoint (AWS_ENDPOINT_URL above) using the copied credentials. The verifiable pattern uses an aws-cli container, for example:
apiVersion: v1
kind: Pod
metadata:
name: s3-consumer
annotations:
# Required on our clusters; without it the pod lands on the RHOAI image-volume SCC
openshift.io/required-scc: restricted-v2
spec:
restartPolicy: Never
securityContext:
runAsNonRoot: true
seccompProfile:
type: RuntimeDefault
containers:
- name: aws
# Pinned below 2.23: newer CLIs send default CRC checksums that Ceph RGW may reject
image: docker.io/amazon/aws-cli:2.17.0
command: ["/bin/bash", "-c"]
args:
- |
aws s3 ls s3://$BUCKET_NAME
aws s3 cp /tmp/object.bin s3://$BUCKET_NAME/object.bin
env:
- name: HOME
value: /tmp
- name: AWS_DEFAULT_REGION
value: us-east-1
envFrom:
- secretRef:
name: s3-creds
- configMapRef:
name: s3-bucket
#
Caveats for cross-cluster consumption
- SCC annotation — add
openshift.io/required-scc: restricted-v2to the pod or workload. Without it, pods are admitted under the RHOAI image-volume SCC on our clusters. - aws-cli version — pin to
2.17.0. Newer CLI versions send default CRC checksums that Ceph RGW may reject. - Presigned URLs and addressing style — presigned GET URLs may fail with a DNS or certificate error if the client resolves them as
bucket.host...(virtual-host addressing), because the wildcard route/certificate does not cover that style. If so, switch to path-style addressing (aws configure set default.s3.addressing_style path, or in boto3Config(s3={'addressing_style': 'path'})). - Egress — if the consumer namespace has restrictive NetworkPolicies, allow egress TCP 443 to the Basel ingress routers so the client can reach the public endpoint.
- Throughput — traffic crosses the Basel ingress routers, so upload/download throughput reflects this path rather than raw RGW speed.
#
Capacity Pool S3 allowance & deprovisioning
Each Capacity Pool has an S3 object-storage allowance. The total maxSize of the ObjectBucketClaims tied to a pool counts against that allowance, so the pool only reserves the capacity your buckets actually need.
Do not deprovision a Capacity Pool that still has ObjectBucketClaims tied to it
If any ObjectBucketClaim is still tied to the Capacity Pool, deprovisioning the pool is blocked by the platform and will fail. Remove the affected ObjectBucketClaims first (or reduce their maxSize) so the remaining total stays within the pool's allowance, then deprovision.