Downscale old replica one by one.

[rudzm]

Hi, I have a specific case with the Argo rollout. I am using traffic management and canary deployment. I want to apply this to deployment, which uses WebSockets, which means connections are established once, and the whole communication is done inside a tunnel. So with the help of rollouts, I want to deploy a new version. To do that, I set up three steps.

- setWeight: 2
- analysis: 
       templates: 
           - templateName: my-template
- setWeight: 100

So in the first step, I create only one pod with a new revision and redirect 2% of traffic.
In the second step, I check if traffic redirected to the new instance doesn't return 500.
If the second steps go well, I redirect the whole new traffic to the new revision, which scales up to the number of replicas set up in the deployment.
After that old replica is terminated, all pods are terminated at once. And I have a problem with this part. Because, like I already mentioned, this is WS tunnels, so if I terminate them all at once, I will flood my service with reconnect attempts, which can cause my own DDoS. I am looking for a method to scale down old revision one by one in some intervals. Is this even possible? Or maybe this can be a new feature for the Argo rollout.

Enabling DynamicStableScale and adding more steps with more Weight is not a case because tunnels from old replicas can connect again to the old revision, and one client can reconnect that many times, how many steps setup.