Is the Score plugin ClusterAffinity necessary?

[zhzhuang-zju]

Please provide an in-depth description of the question you have:
During the scheduling process, the clusters filtered by the filter plugin will be scored by the score plugin, currently there are two score plugins

• ClusterLocality

• ClusterAffinity
  Score plugin ClusterAffinity will give each member cluster a score of zero

ClusterAffinity also functions as a filter plugin, removing member clusters that do not meet its specified criteria. It is already effective during the filtering stage of the clustering process. However, is it necessary for ClusterAffinity to also serve as a scoring plugin to evaluate the clusters, given that it does not impact the results of the scoring phase?

What do you think about this question?:
I did a local test by propagating two workloads to the member clusters, and using the metrics collected during the scheduling process to find out how long each step took.
If keep the Score plugin ClusterAffinity

# in container karmada-scheduler
$  curl 127.0.0.1:8080/metrics | grep Score
karmada_scheduler_scheduling_algorithm_duration_seconds_sum{schedule_step="Score"} 0.00019877799999999999
karmada_scheduler_framework_extension_point_duration_seconds_sum{extension_point="Score",result="Success"} 5.5774e-05

without Score plugin ClusterAffinity

# in container karmada-scheduler
$  curl 127.0.0.1:8080/metrics | grep Score
karmada_scheduler_scheduling_algorithm_duration_seconds_sum{schedule_step="Score"} 0.00010575999999999999
karmada_scheduler_framework_extension_point_duration_seconds_sum{extension_point="Score",result="Success"} 1.3979999999999998e-05

While the time consumption impact of a single instance is minor, it can also save some time when there are a larger number of resources.

Environment:

• Karmada version:
• Kubernetes version:
• Others: