What features would you like to see with MCAD Dashboard V2

[mamy-CS]

• Please include below the features you would like to see with version 2 of the MCAD dashboard:
  [Include as many details as possible]
• You can add your feature requests as comments below.
• Follow the template as a guideline.
• Focus points:
  • Admin User Vs Normal User
  • Cluster Metrics in relation to appwrappers
  • Time series appwrapper data
  • .....

Feature Request

Description

[Provide a clear and concise description of the feature you would like to suggest.]

Use Case

[Explain the specific use case or problem that this feature would address.]

Proposed Solution

[Describe your proposed solution or implementation for the new feature.]

Benefits

[Explain the benefits or advantages of implementing this feature from an end-user point of view.]

Additional Context

[Provide any additional information or context that might be helpful for understanding the feature request.]

[asm582]

To begin with I think we should have below functionalities.
There are two personas

• Cluster user who is submitting AW to a namespace:
  • Status of AW
  • Priority number of AW
  • Priority class of AW
  • Quota of AW
• User who is admin:
  • AWs running across namespaces
  • AWs preempted across namespaces
  • AWs pending across namespaces
  • AWs priority across namespaces
  • AWs exceeding quotas

[tardieu]

In addition to showing information about past or present AppWrappers the dashboard should also give easy-to-read views of the resource available on the cluster (and later quota availability).

Resource availability should reflect priority levels, i.e., more resources are available at a high level due to the possibility of preempting lower-priority AppWrappers. Ideally GPU availability should reflect fragmentation.

In the end, the dashboard should make it easy to answer questions such as: Could an AppWrapper with 16 pods with 8 gpus each be successfully dispatched now with e.g., high priority. If not, why and would 8 pods be ok?

[asm582]

The dashboard should be a read-only source for whatever information MCAD exposes. I think if not yet done, we need these issues to be created on the MCAD side. Also, dashboard information will be point-in-time with the best effort to show available resources in the cluster.

[tardieu]

I don't know that MCAD should provide global availability info as opposed to specific reasons for dispatching/not dispatching an AppWrapper. Related MCAD issue: project-codeflare/multi-cluster-app-dispatcher#421

[mamy-CS]

So should we wait on this to be available within MCAD, and dashboard can pull reason/ summary?

[asm582]

We should start implementing with current information available in status section