Describe the issue:

Installing ODF with a rack based failure domain using VMware vSphere CSI driver backed dynamic storage for ODF deployment. The installation process added a `portable: true` parameter to the `storageclassdeviceset` and we couldn't find Red Hat documentation which explains this parameter. The eventual outcome was that we needed `portable: false` to align the with the intended design. However without docs to cover the usage, this was something which required a support case to ascertain.

Describe the task you were trying to accomplish:

Install ODF, with a rack based failure domain, using VMware vSphere CSI driver backed dynamic storage. 

Suggestions for improvement:

The Rook documentation [1] gives a basic explanation of what it controls. It does not however provide guidance regarding the benefits/use cases in which a customer should have this enabled/disabled.

The impact of the parameter results in a distinct difference in the CRUSH map since enabling the parameter means that the OSD can change hosts and consequently the OSD is decoupled from the host in the CRUSH map when enabled.

 

Clarification should be provided regarding:

Support for both scenarios (true/false)
Implications of each scenario (performance, resilience, troubleshooting etc.)

Document URL:

No Red Hat doc to cover this was found.

[1] https://rook.io/docs/rook/v1.12/CRDs/Cluster/ceph-cluster-crd/#storage-class-device-sets

Chapter/Section Number and Title:

Product Version:

Environment Details:

OCP 4.14
VMware vSphere

Any other versions of this document that also needs this update:

Would imagine that this will impact all OCP/ODF versions on VMware using dynamic provisioning after CSI migration so potentially 4.13 and newer.

Additional information: