Yasong Li

@first-sight12 @luoMonkeyKing We will make a constraint that crd resources need to be installed on the karmada control-plane. Please refer to #2361

@XiShanYongYe-Chang @RainbowMango The e2e test doesn't seem to be very stable, i need to check it. The failed action run: https://github.com/karmada-io/karmada/actions/runs/2704453205 ``` 2022-07-20T12:09:24.0123665Z stderr F I0720 12:09:24.012150 1 options.go:73] karmada-search...

@RainbowMango @XiShanYongYe-Chang Sorry, I will update today

@RainbowMango @XiShanYongYe-Chang Hi I'm testing the "search api" and found that aa doesn't request each pod evenly, it always forwards requests to the same pod over a period of time

@XiShanYongYe-Chang I don't know what its forwarding strategy is, and the timing of switching backends is uncertain. It can only forward to one backend at a time.  

What is its load balancing strategy? Does it not use `service`?

I know this, it should use rr as load balancing strategy, but it doesn't

Some other test cases fail occasionally. The time required for different e2e tests may vary by 10 minutes

@XiShanYongYe-Chang I found that querying the data via the search api automatically starts the informer steps: 1. `kubectl get --raw /apis/search.karmada.io/v1alpha1/search/cache/apis/apps/v1/deployments` (A new informer will be created) https://github.com/karmada-io/karmada/blob/66af6ef5023be78cb81d2260f86a6e5b7ab9fc3a/pkg/registry/search/storage/cache.go#L101 https://github.com/karmada-io/karmada/blob/66af6ef5023be78cb81d2260f86a6e5b7ab9fc3a/pkg/util/fedinformer/genericmanager/single-cluster-manager.go#L125 2....

/cc @XiShanYongYe-Chang