Epic Goal

• Allow using pre-created VM instances for peer-pods

Why is this important?

• Booting of confidential GPU instances (and large memory instances) takes very long in the cloud. Upwards of 10 min. This results in timeouts in different parts of the K8s stack. While some timeouts are configurable some are not. Especially crio doesn't allow to configure container start timeout. Hence we need a mechanism to use pre-created instances for pods to speed up the container startup time.
• Using pre-created instances also enables few additional capabilities - it becomes easier to use instances from different providers, instances with specific network and storage configurations etc.

Scenarios

1. Create a pod using confidential GPUs in public cloud
2. ...
3. ...

Acceptance Criteria 

(The Epic is complete when...)

1. It's possible to create a pod on pre-created instance
2. ..
3. ..

Additional context: