Current Behavior
Today, when NVCF needs to terminate a number of instances, it follows these steps:
- Fetch all instances related to the current deployment from ICMS.
- Select the
instance_ids that should be terminated.
- Send a termination request for each individual
instance_id.
The same flow applies to autoscaler requests and InstanceManagementTask.
Proposed Change
We want to simplify this process by creating a new ICMS delete REST endpoint. The endpoint will accept:
deployment_idgpu_spec_id- the number of instances to terminate
It should use exactly the same algorithm that NVCF uses today: fetch all instances, select the eligible ones, and terminate the requested number.
Current Behavior
Today, when NVCF needs to terminate a number of instances, it follows these steps:
instance_ids that should be terminated.instance_id.The same flow applies to autoscaler requests and
InstanceManagementTask.Proposed Change
We want to simplify this process by creating a new ICMS delete REST endpoint. The endpoint will accept:
deployment_idgpu_spec_idIt should use exactly the same algorithm that NVCF uses today: fetch all instances, select the eligible ones, and terminate the requested number.