totalResource calculation error in proportion plugin

[bysph]

Issue Description:

I found proportion plugin has some issues:

1. It does not consider node label isolation scenarios: such as impala nodes being unavailable for use by the spark queue, but are still factored into the calculation of available resources for the flink queue. (I think the node-group plugin may also fails to consider this)
2. it does not filter out cordoned nodes
3. it does not consider about nodeSelectors: some nodes does not managed by volcano will also be included in the calculation.
4. It does not consider multi-scheduler scenarios: for example, in the case of the "deserved" parameter calculated through the cluster's overall allocatable resources, in a multi-scheduler setting, Volcano's available resource amount is actually less, resulting in each queue's "deserved" being larger than actual, rendering the constraints ineffective. (This scenario might be fine, as handling multiple schedulers can indeed be challenging)

These issues may result in an inflated deserved value for all queues, potentially rendering the guarantee ineffective.

Description of case 1 (node label isolation scenarios)：

Expected Behavior:

The total resource of the queue needs to be recalculated, rather than being calculated uniformly based on the total resources of the entire cluster.

Steps to Reproduce:

for example of case 2:

1. Set a nodeSelector for nodes in the cluster.
2. Use the proportion plugin to manage resources.
3. Enter vscode debug mode to see the value of ssn.TotalResource or the list of ssn.Nodes

Additional Information:

Volcano Version: v1.8.2
Kubernetes Version: v1.19.3

Proposed Solution:

The total resources of the queue need to be recalculated as follows:

1. Sum up the allocated resources of all available nodes (nodeLabels and nodegroup).
2. Deduct resources from cordoned nodes.

And the "deserved" value also needs to be recalculated:

1. Accumulate weights for the same nodegroup, then calculate proportionally.
   .....

[bysph]

I found proportion plugin has some issues:

1. It does not consider multi-scheduler scenarios: for example, in the case of the "deserved" parameter calculated through the cluster's overall allocatable resources, in a multi-scheduler setting, Volcano's available resource amount is actually less, resulting in each queue's "deserved" being larger than actual, rendering the constraints ineffective.
2. It also fails to consider node label isolation scenarios: such as impala nodes being unavailable for use by the flink queue, but are still factored into the calculation of available resources for the flink queue.
3. Additionally:it does not filter out cordoned nodes, which should be excluded from the calculation of queue available resources even when set to disallow scheduling.

@Monokaix @lowang-bh Hi, Could you help confirm this issue? thx~

[Monokaix]

3. it does not consider about nodeSelectors: some nodes does not managed by volcano will also be included in the calculation.

You mean what node-selector specified？

[bysph]

3. it does not consider about nodeSelectors: some nodes does not managed by volcano will also be included in the calculation.

You mean what node-selector specified？

Yes，I found that AddPod will add all nodes to cache, althrough AddNode filter them out.

[bysph]

3. it does not consider about nodeSelectors: some nodes does not managed by volcano will also be included in the calculation.

You mean what node-selector specified？

Yes，I found that AddPod will add all nodes to cache, althrough AddNode filter them out.

Sorry, the response here is incorrect; the nodeselector issue may be caused by the csinode informer.