Storage for the FLAME Hub

The Hub contains several stateful services. Storage is therefore an infrastructure decision, not just a chart setting. The chart uses the cluster's default StorageClass unless an explicit class is configured for a component.

Recommended model

Use a CSI-backed volume service that provides durable block storage and supports detaching a volume from a failed VM and reattaching it to the replacement node.

For FLAME clusters running as virtual machines on OpenStack, the recommended choice is the OpenStack Cinder CSI driver. When the Cinder volume type is backed by Ceph, replication already happens below Kubernetes. Do not add a second Mayastor replication layer on top: it will slow everything down.

Verify the cluster storage

List available classes and identify the default:

kubectl get storageclass

Configure the chart

If the correct class is the cluster default, no storage override is necessary. Otherwise set the same durable class on each persistent component in your override file. The exact keys can change, so use the current values.yaml as the reference.

The first-party PostgreSQL and SeaweedFS settings look like this:

postgresql:
  persistence:
    storageClass: <cinder-storage-class>
    size: 16Gi

seaweedfs:
  allInOne:
    data:
      storageClass: <cinder-storage-class>
      size: 80Gi

Also review persistence for Harbor registry/jobservice/Trivy, RabbitMQ, Grafana, Prometheus, VictoriaLogs, and serverCoreWorker. Size these claims for the expected analysis and retention load.

Storage classes are effectively immutable for existing claims

Changing a value does not migrate existing data.