Every setting under spec.rdb.blockStorage is fixed once your RegattaCluster resource is created, because Kubernetes does not allow changing a StatefulSet’s storage templates. Choose these values carefully up front.
accessModes is a list mainly so you can request ReadWriteOncePod instead of the default ReadWriteOnce. ReadWriteOncePod restricts the volume to exactly one Pod cluster-wide, stricter than ReadWriteOnce (which technically still permits multiple Pods on the same Kubernetes node); some customers prefer it for raw block storage. ReadOnlyMany and ReadWriteMany are part of the underlying Kubernetes API but are not meaningful choices here - RDB’s raw block storage is always single-writer, one device per Pod.- Every RDB Pod can be given more than one raw block device: set
spec.rdb.devices to the total device count per Pod. All devices on an RDB Pod share the same spec.rdb.blockStorage settings; there is no per-device size or StorageClass. Device indexes start at 0: the first device is /dev/rdb0, the second is /dev/rdb1, and so on. A RegattaCluster with spec.rdb.replicas: 3 and spec.rdb.devices: 2 creates 6 raw block PVCs in total (2 per RDB Pod):
Choosing dynamic or static RDB storage
In short: use Dynamic if your cluster has a StorageClass that provisions raw block volumes on demand, and use Static if your raw block storage is provisioned outside Kubernetes (for example pre-attached local disks) or your environment does not support dynamic block provisioning.
spec.rdb.blockStorage.provisioning selects how Kubernetes obtains the raw block PersistentVolumes for RDB:
Dynamic (the default): a StorageClass, named in spec.rdb.blockStorage.storageClassName, provisions a new PersistentVolume for every raw block PVC on demand. Use this when your Kubernetes cluster has a StorageClass that supports volumeMode: Block. Do not set spec.rdb.blockStorage.selector with Dynamic provisioning: Kubernetes never dynamically provisions a PersistentVolumeClaim that has a selector, so the PVC would stay Pending indefinitely; the operator rejects this combination up front instead.Static: you pre-create the PersistentVolumes yourself, and spec.rdb.blockStorage.selector (matchLabels or matchExpressions) tells the operator which ones to bind. Use this when your raw block storage is provisioned outside Kubernetes, for example pre-attached local disks, or your environment does not allow dynamic provisioning for raw block storage.
With Static provisioning, you need at least spec.rdb.replicas * spec.rdb.devices PersistentVolumes that match the selector, StorageClass, requested access modes, volumeMode: Block, and requested capacity, and that are available or already bound to your claims.
The cluster-scoped read-only permission you installed in Installation lets the operator check these conditions before applying resources, so problems surface early. If that permission is ever missing or denied, the operator skips this check and relies on normal PVC binding as the readiness signal instead.
See Example: Two Local NVMe Devices, One RDB Replica for a full walkthrough of static provisioning.
spec.repoStorage and spec.logs are Pod-level settings, applying equally to the SM Pod and every RDB Pod. spec.repoStorage backs RegattaDB’s mutable runtime repository paths. spec.logs backs RegattaDB’s own log output at /var/log/regatta. See Prepare For Deployment for the underlying storage-sizing guidance this maps to. Neither field has a default size - you must size both for your workload; storageClassName falls back to your cluster’s default StorageClass if omitted.
Storage size, StorageClass, and access modes for spec.repoStorage and spec.logs cannot change once your RegattaCluster resource is created, for the same reason as spec.rdb.blockStorage. Choose these values carefully up front.
Multiple RDB devices: naming and readiness
Every raw block device on an RDB Pod is backed by its own PersistentVolumeClaim, generated from the claim-template name, the RDB StatefulSet name, and the Pod ordinal. For a RegattaCluster named my-regattadb, device 0 of RDB Pod 0 produces the PVC rdb-block-0-my-regattadb-rdb-0; device 1 of the same Pod produces rdb-block-1-my-regattadb-rdb-0, and so on.
Storage readiness requires every one of these PVCs, for every RDB Pod, to be bound before RegattaDB is configured. An RDB Pod with spec.rdb.devices: 2 stays pending until both of its raw block PVCs are bound, not just one.
Non-root access to raw block devices
RDB Pods run as the non-root user and group configured in spec.security (UID 1000/GID 1001 by default) and additionally receive Linux supplemental group 6, the conventional disk group on most Linux distributions, so that user can open the raw block devices without running as root. The SM Pod does not receive this supplemental group, because it has no raw block devices.
Whether group 6 actually grants access depends on which Linux group your storage and container runtime assign to the device node inside the Pod. For deterministic, non-root device ownership, configure your container runtime to derive device ownership from the Pod’s security context: set device_ownership_from_security_context = true under containerd, or the equivalent option under [crio.runtime] for CRI-O. Validate this on one RDB Pod, for example by checking device ownership inside the Pod, before relying on it for production. 
