Does Linkerd2 not support disaster recovery drills?

[Maple521]

Hi，everyone
I have 4 apps deployed in K8s with 8 replicas of each app in the ZoneA and ZoneB availability zones and they are all brokered by Linkerd2.
I am doing a pressure test now and the TPS is 4000TPS. When I want to shut down all ECS in ZoneB, I expect the TPS to be reduced to 2000 and the system will still work normally. But the reality is that the whole system is down and no requests can come in, But when I didn't use Linkerd2, I just used K8s, and the remaining 4 Pods still worked. TPS dropped to 2000, so as a ServiceMesh product, this shouldn't happen. Am I using it the wrong way?
I just use it in
annotations: linkerd.io/inject: enabled
It was injected into Linkerd2

Below are screenshots of my tests, both with and without Linkerd2
Use Linkerd2：

No Use Linkerd2：