CSI Capabilities

Feature Overview

The CSI API still has some key capabilities that have not reached maturity. This may be due to the API itself or the Kubernetes API objects that facilitate this. Some of these features may be picked up automatically as part of effort upstream, however some capabilities will need to be push to resolution by the team.

In addition we need to be able to migrate storage from using intree drivers to the equivalent CSI drivers with minimal impact to end users and administrators.

Goals

• Bring the following CSI features to GA in OpenShift
  • Resize
  • Clone
  • Snapshot/Restore
• Operational improvements such as allowing PVs to be detached when a node is shutdown ungracefully, better capacity tracking, etc.
• Ephemeral & Generic Ephemeral volumes are tracked by a dedicated feature OCPPLAN-9193 

Requirements

Out of Scope

This work will only cover the API and surrounding Kubernetes API objects, it will not include work on individual drivers

Application consistent snapshots is not included in this initial work

Background, and strategic fit
In a future Kubernetes release (currently 1.21) intree cloud provider drivers will be deprecated and replaced with CSI equivalents, if you do not have completed migration capability you cannot remove the intree drivers.

Snapshot/Clone in this case will be for Crash Consistent snapshots which is pretty standard as far as the industry is concerned, there is the need for significant enhancement/change to the Kubernetes API and the applications themselves to enable full Application Consistent snapshots.

Assumptions

• Storage SIG won't move out the changeover to a later Kubernetes release

Customer Considerations
Customers will need to be able to use the storage they want.

Documentation Considerations

• Target audience: cluster admins
• Updated content: update storage docs to show how to use these capabilities (specific focus on migration maybe but as this will be a diminishing function maybe not in the docs for a long time)